Michelson-Morley, Occam and Fisher:

The Radical Implications of Stable Inflation at

Near-Zero Interest Rates

John H. Cochrane

December 7, 2016

Abstract

The long period of quiet inflation at near-zero interest rates, with large quan-
titative easing, suggests that core monetary doctrines are wrong. It suggests that
inflation can be stable and determinate under a nominal interest rate peg, and that
arbitrary amounts of interest-paying reserves are not inflationary. Of the known
alternatives, only the new-Keynesian model merged with the fiscal theory of the
price level is consistent with this simple interpretation of the facts.
I explore two implications of this conclusion. First, what happens if central
banks raise interest rates? Inflation stability suggests that higher nominal interest
rates will result in higher long-run inflation. But can higher interest rates temporar-
ily reduce inflation? Yes, but only by a novel mechanism that depends crucially on
fiscal policy. Second, what are the implications for the stance of monetary policy
and the urgency to normalize? Inflation stability implies that low-interest rate
monetary policy is, perhaps unintentionally, benign, producing a stable Friedman-
optimal quantity of money, that a large interest-paying balance sheet can be main-
tained indefinitely, and that it might not be wise for central bankers to exploit a
temporary negative inflation effect.
The fiscal anchoring required by this interpretation of the data responds to dis-
count rates, however, and may not be as strong as it appears.

Hoover Institution, Stanford University and NBER. Also SIEPR, Stanford GSB, and Cato Institute.
http://faculty.chicagobooth.edu/john.cochrane. This paper supersedes a working paper titled Do higher
interest rates raise or lower inflation? I thank Edward Nelson and Jordi Gal for helpful comments.
2 COCHRANE

Contents

1 Summary and overview 1

2 Nothing happened 5

3 Theories 9
3.1 A very simple model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.1.1 Old-Keynesian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.1.2 New Keynesian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.1.3 Fiscal theory of the price level . . . . . . . . . . . . . . . . . . . . . 16
3.2 Language . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

4 Michelson-Morley and Occam 19

4.1 Offsetting instabilities? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.2 Really slow unstable dynamics? . . . . . . . . . . . . . . . . . . . . . . . . 20
4.3 Not really a peg? Selection from future actions . . . . . . . . . . . . . . 21
4.4 Sunspot volatility? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.5 Occam summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

5 Quantitative easing and monetarism 29

6 Do higher interest rates raise or lower inflation? 35

6.1 Pure Fisher . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

7 Sticky prices 38
7.1 Basic impulse-response function . . . . . . . . . . . . . . . . . . . . . . . 39
7.2 Mean-reverting and stairstep rates . . . . . . . . . . . . . . . . . . . . . . 42
7.3 Taylor rules disclaimer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

8 A fiscal-theoretic model with long term debt 45

8.1 Long term debt and sticky prices . . . . . . . . . . . . . . . . . . . . . . . 49
8.2 Critical assumptions and intuition . . . . . . . . . . . . . . . . . . . . . . 55

9 Money 59
9.1 Impulse-response functions . . . . . . . . . . . . . . . . . . . . . . . . . . 61
 INFLATION AND INTEREST RATES 3

9.2 Money and transitory rate shocks . . . . . . . . . . . . . . . . . . . . . . . 65

9.3 Interest spread policy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
9.4 Money summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

10 The Phillips curve 70

11 Other equilibria 74
11.1 Multiple equilibria in the impulse-response function . . . . . . . . . . . . 74
11.2 Choosing equilibria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
11.3 Anticipated movements and backwards stability . . . . . . . . . . . . . . 76
11.4 Fiscal index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
11.5 Using fiscal policy to choose equilibria . . . . . . . . . . . . . . . . . . . . 81

12 Taylor rules 84
12.1 Constructing Taylor rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
12.2 Open-mouth policy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
12.3 Reasonable disturbances . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
12.4 The standard three-equation model . . . . . . . . . . . . . . . . . . . . . 91

13 Other modeling directions; necessary vs. sufficient assumptions 93

14 VARs 96

15 Fiscal policy anchoring and warning 99

15.1 Exogenous surpluses? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
15.2 On and off equilibria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
15.3 What about Japan? And Europe? And the US? . . . . . . . . . . . . . . . . 102
15.4 What about 2008? And cyclical inflation? . . . . . . . . . . . . . . . . . . . 103
15.5 What about 1951? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
15.6 The uncomfortable future . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

16 Literature 107

17 Concluding comments 112

18 Appendix. A Phillips curve with lags 119

4 COCHRANE

18.1 Roots, moving average, and response . . . . . . . . . . . . . . . . . . . . . 120

18.2 Backward-looking Phillips algebra . . . . . . . . . . . . . . . . . . . . . . 127

19 Algebra Appendix 130

19.1 Variance of inflation in forward vs. backward looking models . . . . . . . 130
19.2 Fiscal theory formulas for delayed and temporary rate rises . . . . . . . . 132
19.3 Sticky-price model solution . . . . . . . . . . . . . . . . . . . . . . . . . . 134
19.4 Model and solutions in continuous time . . . . . . . . . . . . . . . . . . . 137
19.5 Impulse response function explicit solution . . . . . . . . . . . . . . . . 140
19.6 Impulse-response with long-term debt and price stickiness . . . . . . . . 142
19.7 Linearized valuation equation . . . . . . . . . . . . . . . . . . . . . . . . . 144
19.8 The Model with Money . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
19.8.1 CES functional form . . . . . . . . . . . . . . . . . . . . . . . . . . 146
19.8.2 Money demand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
19.8.3 Intertemporal Substitution . . . . . . . . . . . . . . . . . . . . . . 148
19.9 Three-equation model solution . . . . . . . . . . . . . . . . . . . . . . . . 150
 INFLATION AND INTEREST RATES 1

1 Summary and overview

For nearly a decade in US, UK, and Europe, and three decades in Japan, short-term in-
terest rates have been near zero, and thus stopped moving more than one-for-one with
inflation as prescribed by the Taylor principle. In the last decade, central banks also
embarked on immense open market operations. US quantitative easing (QE) raised
bank reserves from $50 billion to $3,000 billion, a factor of 60.

The response to this important experiment in monetary policy has been silence.
Inflation is stable, and if anything less volatile than before. There is no visually apparent
difference in macroeconomic dynamics in the near-zero-rate and large-reserves state
than before.

Existing theories of inflation make sharp predictions about this circumstance: Old-
Keynesian models, characterized by adaptive expectations, and in use throughout the
policy world, predict that inflation is unstable under passive (it = t + vti ; < 1)
monetary policy, and therefore predict a deflation spiral at the zero bound. It did not
happen. Monetarist thought, M V = P Y with V stable in the long run, predicts that a
massive increase in reserves must lead to galloping inflation. It did not happen.

New-Keynesian models, featuring rational expectations and interest rate targets,

predict that inflation is stable under passive policy. Unless one adds frictions, those
models also predict that quantitative easing operations are irrelevant. The observed
inflation stability is thus a big feather in the new-Keynesian cap. But standard new-
Keynesian models predict that inflation becomes indeterminate when interest rates
cannot or do not move, and so suffers from additional sunspot volatility. These mod-
els also predict a menagerie of policy paradoxes: Productivity improvements are bad,
promises further in the future have larger effects today, and reducing price stickiness
makes matters worse, without limit.

This is a Michelson-Morley1 moment for monetary policy. We observe a decisive

experiment, in which previously hard-to-distinguish theories clearly predict large out-

1
In 1887, Albert A. Michelson and Edward W. Morley set out to measure the speed of Earth through the
ether, the substance thought in its day to carry light waves, by measuring the difference between the speed
of light in various directions. They found nothing: the speed of light is the same in all directions, and the
Earth appears to be still. Special relativity follows pretty much from this observation alone.
2 COCHRANE

comes. That experiment yields a null result, which cleanly invalidates those theories.

Now, any theory, especially in economics, invites rescue by epicycles. Perhaps infla-
tion really is unstable, but artful quantitative easing just offset the deflation vortex, Or
perhaps wages are much stickier than we thought, or money is taking a long time to
leak from reserves to broader aggregates, so we just need to wait a bit more for unstable
inflation to show itself. Perhaps a peg really does lead to indeterminacy and sunspots,
but expectations about active ( > 1) monetary policy in the far future takes the place
of current Taylor rule responses to select equilibria. Perhaps the Earth drags the ether
along with it.

Occam responds: Perhaps. Or, perhaps one should take seriously the simplest an-
swer: Perhaps inflation can be stable and determinate under passive monetary policy,
including an interest rate peg, and with arbitrarily large interest-bearing reserves. Clas-
sic contrary doctrines were simply wrong.

We are not left, as Michelson and Morley were, with a puzzle a set of facts that
existing theories cannot account for. Adding the fiscal theory of the price level to to the
standard rational-expectations framework, including new-Keynesian price stickiness,
we obtain a simple economic model in which inflation can be stable and determinate
under passive policy, zero bound, or even a peg, and despite arbitrary quantitative eas-
ing. The model also has a smooth frictionless limit, and resolves new-Keynesian policy
paradoxes.

What does this experience, and theoretical interpretation, imply about monetary
policy going forward?

First, if inflation is stable under an interest rate peg, then it would seem to follow
that were the central bank to raise interest rates and leave them there, then inflation
must eventually rise. This reversal of the usual sign of monetary policy has become
known as the neo-Fisherian hypothesis.

However, higher interest rates might still temporarily lower inflation before eventu-
ally raising it. I investigate what minimal set of ingredients it takes to produce a nega-
tive short-run impact of interest rates on inflation.

This quest has a larger goal. We do not have a simple economic baseline model
 INFLATION AND INTEREST RATES 3

that produces a negative response of inflation to a rise in interest rates, in our world
of interest rate targets and abundant excess reserves. If there is a short-run negative
relationship, what is its basic economic nature?

The natural starting place in this quest is the simple frictionless Fisherian model,
it = r + Et t+1 . A rise in interest rates i produces an immediate and permanent rise in
expected inflation. In the search for a temporary negative sign I add to this basic fric-
tionless model 1) new-Keynesian pricing frictions 2) backwards-looking Phillips curves
3) monetary frictions. These ingredients robustly fail to produce the short-run nega-
tive sign. You cannot truthfully explain, say, to an undergraduate or policy maker that
higher interest rates produce lower inflation because prices are sticky, or because lower
money supply drives up rates and down prices, and our fancy models build on this
basic intuition.

One ingredient can robustly and simply produce the desired temporary negative
sign. If we add long-term debt, and if we assume that fiscal policy does not respond to
variation in the real cost of debt, then a rise in interest rates can produce a temporary
decline in inflation. Higher nominal rates lower the nominal present value of long-term
debt; absent any change in expected surpluses, the price level must fall to restore the
real present value of the debt. That works, but it is a rather dramatically novel mecha-
nism relative to all standard economic stories and policy discussion.

We are left with a logical conundrum: Either 1) The world really is Fisherian, higher
interest rates raise inflation in both short and long run; 2) more complex ingredients,
including frictions or irrationalities, are necessary as well as sufficient to deliver the
negative sign, so this hallowed belief relies on those complex ingredients; 3) the nega-
tive sign ultimately relies on the fiscal theory story involving long-term debt and has
nothing to do with any of the mechanisms commonly alluded for it.

The first view is not as crazy as it seems. The VAR evidence for the traditional sign,
reviewed below, is weak. Perhaps the persistent price puzzle was trying to tell us
something for all these decades. Raw correlations are of little use, as interest rates and
inflation move closely together under either theoretical view, at least away from the
zero bound.

The second set of policy issues: Is it important for central banks to raise expected in-
4 COCHRANE

flation, raise nominal rates, reduce the size of their balance sheets, ration non-interest-
bearing reserves, and return to active Taylor rules? (The Fed refers to this package,
roughly what it was doing 1982-2007, as normalizing policy, though its not clear that
the package deserves the normative flavor of that blessing, or that alternatives deserve
the implied abnormal status.)

The experience of stable inflation at near-zero interest rates suggests that we can
instead live the Friedman optimal quantity of money forever at least a large balance
sheet of interest-bearing reserves financed by short-term government debt, potentially
low or zero rates with corresponding low inflation or even slight deflation, and con-
sequently permanent abandonment of the Taylor principle. That conclusion requires
verification in theory at least the existence of a theory that argues current experience
can continue, and examination of its auxiliary assumptions, which I provide.

Whether we should do so requires listing (here) and quantifying (eventually) trade-

offs the ability of a stochastic peg to accommodate adverse natural rate shocks
without violating the lower bound, the advantages of the Friedman optimal quantity
of money, the practicality of exploiting a poorly understood temporary negative sign.

Finally, stable but temporary negative reaction is a quite different beast than unsta-
ble. Even if we find a model that is stable and determinate, but produces a negative
reaction of inflation to expected interest rate rises, that remains quite different from
restoring the classic view of instability and a permanent negative effect. In particular,
the former model does not imply Taylor-rule style stabilizations. The Taylor rule is wise
for an unstable model (old-Keynesian) or an indeterminate model (new-Keynesian).

I address a few common objections. How can the fiscal theory be consistent with
low inflation, given huge debts and ongoing deficits? Fortunately, the fiscal theory does
not predict a tight linkage between current debts, deficits and inflation. Discount rates
matter as well, and discount rates for government debt are very low. What about other
pegs, which did fall apart? Answer: fiscal policy fell apart.

That last observation leads to a final warning. My careful hedging, that an interest
rate peg can be stable, refers to the necessary fiscal foundations. If fiscal foundations
evaporate, that theory warns, and harsh experience reminds us, so can our benign mo-
ment of subdued and quiet inflation. I sketch the mechanism.
 INFLATION AND INTEREST RATES 5

2 Nothing happened

Figure 1 presents the last 20 years of interest rates, inflation and reserves in the U.S. The

4
10Y Govt

3
Core CPI

1 Fed Funds
Reserves
0
1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016

Figure 1: Recent US experience.

federal funds rate follows its familiar cyclical pattern, until it hits essentially zero in 2008
and stays there. In 2008-2009, the severity of the recession and low inflation required
sharply negative interest rates, in most observers eyes and in most specifications of
a Taylor rule. The zero bound was binding. If you see data only up to the bottom
of inflation in late 2010, and if you view inflation as unstable under passive monetary
policy, fear of a deflation spiral is natural and justified.

But it never happened. Despite interest rates stuck at zero, inflation rebounded with
about the same pattern as it did following the previous two much milder recessions.

Interest rates have remained near zero ever since. The interpretation of this period
is less obvious. To many, we still are in a period in which the natural rate is low, the
zero bound binds, and consequently a deflation spiral still looms. To others, the Taylor
rule recommended interest rates to rise by about 2012, so we should regard near-zero
6 COCHRANE

rates as a policy choice, not a bound one that threatens inflation, not deflation. But
interest rates are stuck either way. (Since inflation is low, the difference in views comes
down to ones views about the desirable response of inflation to output gaps, y in it =
t + y yt + vti , and views on how big the output gap is.)

Still, nothing happened. Inflation gently declined. Overall, inflation is dominated

by a slow 20 year downward trend. On top of that trend there is the usual business cycle
pattern, that inflation declines in a recession and then rises again when the recession
is over.

Moreover, to the unaided eye, inflation dynamics are unaffected by the long period
of immobile interest rates. In fact, after 2012, when the financial crisis and deep reces-
sion receded, inflation volatility is lower than it was before 2009, when interest rates
could actively stabilize inflation. A theory in which > 1 vs = 0 is an important
state variable for stability, determinacy, or other economic dynamics is challenged by
this period.

The Fed also increased bank reserves, from about $50 billion to nearly $3,000 bil-
lion, in three quantitative easing (QE) operations, as shown in Figure 1. Once again,
nothing visible happened. QE2 is associated with a rise in inflation, but QE1 and QE3
are associated with a decline. And the rise in inflation coincident with QE2 mirrors
the QE-free rise coming out of the much milder 2001 recessions. QE2 and QE3 were
supposed to lower long-term interest rates. To the eye, the 20 year downward trend in
long term rates is essentially unaffected by QE. If anything, long-term interest rates rose
coincident with QE operations.

Figure 2 plots the unemployment rate and GDP growth rate. Together with Figure 1,
these figures also show no visible difference in macroeconomic dynamics in and out of
the zero rate / QE state. Yes, there was a bigger shock in 2008. But the unemployment
recovery looks if anything a bit faster than previous recessions. Output growth, though
too low in most opinions, is if anything less volatile than before.

Figure 3 tells a similar but longer story for Japan. Japanese interest rates declined
swiftly in the early 1990s, and essentially hit zero in 1995. Again, armed with the tradi-
tional theory that inflation is unstable unless interest rates can move more than one-
for-one in response, and seeing data up to the bottom of inflation in 2001, or again
 INFLATION AND INTEREST RATES 7

10

8 Unemployment

Fed Funds
2

GDP Growth
-2
1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016

Figure 2: US federal funds rate, unemployment rate, and real GDP growth rate.

in late 2010, predicting a deflation spiral is natural and justified. But again, it never
happened. Despite large fiscal stimulus and quantitative easing operations, Japanese
interest rates stuck at zero with slight deflation for nearly two decades. The 10 year
government bond rate never budged from its steady downward trend.

The bottom panel of Figure 3 repeats the story for Europe. Here the spread of low
rates and slight deflation is even stronger than in the US.

Both Japan and Europe diverge from the U.S. in the last few years, with less inflation
and lower interest rates. But are Japanese and European inflation lower despite their
lower or even negative interest rates, or because of them?
8 COCHRANE

3
Percent

2
10Y Govt

Int rate
0

-1

Core CPI
-2
1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016
5

4 1 mo Euro Libor

3
Percent

1 Euro Core CPI

-1
2000 2002 2004 2006 2008 2010 2012 2014 2016

Figure 3: Japan and Europe. Top: Discount rate, Call rate, Core CPI, and 10 year Gov-
ernment Bond Yield in Japan. The thin presents the raw CPI data. Thick line adjusts
the CPI for the consumption tax by forcing the April 2014 CPI rise to equal the rise in
March 2014. Bottom: Europe
 INFLATION AND INTEREST RATES 9

3 Theories

Old-Keynesian models predict that passive policy including an interest rate peg is un-
stable, and the Taylor rule stabilizes an otherwise unstable economy. In that model, the
central bank must first lower interest rates to get inflation going, then quickly catch up
to keep inflation from spiraling out of control. New-Keynesian models predict that pas-
sive policy including an interest rate peg is stable, but leave indeterminacies. The Tay-
lor rule destabilizes an otherwise stable economy to remove local indeterminacies. In
response to a permanent interest rate increase, this model predicts inflation will even-
tually rise, but the sign of the immediate response can go either way.

The key distinction between the models is rational vs. adaptive expectations. Since
this is review of well-known results, my contribution here is a model that maximizes
simplicity rather than realism or generality.

A visual metaphor may help to map out the purpose of the equations. Imagine a
Fed chair balancing a walking stick upside down. Metaphorically, the end of the stick
she holds is the interest rate, and the other end is inflation. The stick wants to fall
over inflation is unstable. But the Fed chair can keep inflation steady if she quickly
moves the bottom of the stick more than one for one with little movements in the top,
balancing it as a seal balances a ball on its nose. The Taylor principle stabilizes the
dynamic system. If she wants more inflation, she must first move the bottom stick
slightly the wrong way, get the top going, and then swiftly catch up.

Suppose instead our Fed chair holds her walking stick right side up. Now the stick is
stable. Even just holding the top, the bottom will swing back in to place. To get the stick
going, she need only move the top in the direction she wants to go. Still, however, the
bottom of the stick may go the opposite way temporarily if she moves the top quickly.
(I cant do determinacy in this metaphor alas.)

How could we tell from data on the position of the top and bottom of the stick which
way it is being held? In normal times it would be very hard to do. The top and bottom
of the stick would move together at low frequencies, and sometimes moves in opposite
directions at high frequencies. But if someone stopped the Fed chairs hand, we would
quickly sort it out: If the stick were being held from the bottom, unstable, it would
10 COCHRANE

topple over. If it were held from the top, stable, the inflation end would settle down to
follow the fed Chairs hand.

3.1 A very simple model

Consider a Fisher equation, a Phillips curve, a static IS curve, and a Taylor rule for mon-
etary policy:

it = rt + te Fisher (1)

t = te + xt Phillips (2)

xt = (rt vtr ) IS (3)

it = t + vti Taylor (4)

where i is the nominal interest rate, r is the real interest rate, is inflation, e is expected
inflation, x is the output gap, v i is a monetary policy disturbance, and we can label v r
as a natural rate disturbance, the latter two potentially serially correlated.

By specifying a static IS equation, without the usual term Et xt+1 on the right hand
side, we can solve the model trivially without matrices or eigenvalues. The same points
hold in more general and realistic models. Eliminating x and r, equating the resulting
two expressions for the nominal interest rate i, we reduce the model to the solution of
a single equation in :
 
1 1
t + vti = t + 1 + te + vtr (= it ) (5)

The right-hand term of equation (5) is a Fisher equation. i = = e is a steady state.

But the dynamics are smeared out in time due to price stickiness. Those dynamics are
the key.
 INFLATION AND INTEREST RATES 11

3.1.1 Old-Keynesian

Old-Keynesian models specify adaptive expectations, te = t1 . Substituting that spec-

ification in (5) we obtain

1 +
t = t1 (v i vtr ) (6)
1 + 1 + t

For < 1, and at a peg = 0 in particular, the dynamics of this system are unstable
and determinate. The coefficient on lagged inflation is above one. There is only one
solution.

The coefficient on the disturbance term is negative. Therefore, higher interest rates
or a lower natural rate send inflation spiraling down, as Figure 4 illustrates.
2

Interest i
0

Inflation
-2

, less sticky
percent

-4

-6

-8

-10
-3 -2 -1 0 1 2 3 4 5
time

Figure 4: Simulation of a permanent interest rate rise or natural rate fall in the simple
old-Keynesian model, with passive = 0 policy. The baseline uses = 1/2, = 1. The
less sticky case uses = 1.

In this model the Taylor rule stabilizes an otherwise unstable system. If > 1, the
coefficient on lagged inflation is less than one. A monetary contraction now has a uni-
12 COCHRANE

formly negative, temporary (if v i is temporary), but bounded effect on inflation.

But when < 1, as it must be at the zero bound, then the model reverts to unstable
dynamics. This model makes a clear prediction: At the zero bound, deflation must
spiral.

Figure 4 includes the response for less price stickiness, = 1 instead of = 1/2.
Sensibly, less sticky prices speed up dynamics. But that just makes the explosion hap-
pen faster, without limit. The adaptive expectations model does not approach the fric-
tionless rational expectations limit in which interest rates rise, and expected inflation
rises contemporaneously.

In the old-Keynesian view, we do not see exploding inflation, the opposite of Figure
4, because central banks arent dumb enough to keep interest rates constant in the face
of inflation.

To illustrate, Figure 5 plots the response of this very simple old-Keynesian model to
a permanent monetary policy shock vti .

Here, the tightening sends inflation uniformly down. Interest rates rise at first, to
get disinflation going, but then must quickly follow inflation in order to stop it from
going too far. This graph embodies exactly the sequence of events Friedman (1968) (p.
6) described of an interest rate change.

The distinction between monetary policy disturbance vti and the path of observed
interest rates it is important. A positive policy shock leads quickly to lower interest
rates.

In the long run, interest rates embody the Fisher relationship, and interest rates
move one for one with inflation. The model is not Fisherian, however, as it is unstable
and interest rates must initially go the opposite way.

Figure 5 helps to illustrate why it is hard to tell an unstable model, whose central
bank is following active policies and not letting instabilities erupt, from a stable model.
Equilibrium interest rates and inflation will track up and down together in both cases.
Until the interest rate cannot move. Thats why the recent episode in which interest
rates cannot move downward and did not move upward is so important.
 INFLATION AND INTEREST RATES 13

1 vi

0.5

i
0
Percent

-0.5

-1

-1.5

-2
-2 0 2 4 6 8 10 12 14
Time

Figure 5: Response to a step-function monetary policy disturbance in the simple old-

Keynesian model. The model is xt = (it t1 ); t = t1 + x t; it = t + vti . The
figure plots the response to a permanent 1% increase in vti . = 1.5, = 0.2, = 1.

3.1.2 New Keynesian

The new Keynesian tradition instead uses rational expectations: te = Et t+1 . Substi-
tuting this specification into (5), we obtain

1 +
Et t+1 = t + (v i vtr ). (7)
1 + 1 + t

For < 1, the coefficient on t is less than one, so this model is stable all on its own,
even under an interest rate peg = 0. Adaptive, backward-looking expectations make
price dynamics unstable, like driving a car by looking in the rear-view mirror. Ratio-
nal, forward-looking expectations make price dynamics stable, as when drivers look
forward and veer back on the road without outside help.

The last term in (7) is positive, so in this model, a monetary policy tightening v i or
a natural rate fall v r raise inflation. Figure 6 illustrates the response of inflation to an
14 COCHRANE

interest rate rise in this case.

Already, the new-Keynesian model reverses the hallowed doctrine that interest rate
pegs are unstable, and the widespread presumption that higher interest rates lower in-
flation.

A temporary fall in the natural rate v r under an interest rate peg likewise raises in-
flation, and then inflation gradually declines as the natural rate reverts to normal. In-
flation accommodates needed changes in the natural rate, albeit slowly, all by itself
without the need for active Fed action or announcements. One might read the history
of slowly decreasing inflation during recovery at the zero bound as an instance of this
mechanism.

1 Interest i

0.8
, less sticky

0.6
Inflation
percent

0.4

0.2 Multiple equilibria

-2 -1 0 1 2 3 4 5
time

Figure 6: Simulation of an interest rate rise or natural rate fall in the simple new-
Keynesian model with passive = 0 monetary policy. The baseline uses = 1/2, = 1,
the less sticky case uses = 1. Dashed lines indicate potential multiple equilibria.

This model with < 1 is, however indeterminate. It only ties down expected infla-
tion Et t+1 , where the old-Keynesian model ties down actual inflation. To the solutions
of this model we can add any expectational error, t+1 , such that Et t+1 = 0, and then
 INFLATION AND INTEREST RATES 15

write the models solutions as

1 +
t+1 = t + (v i vtr ) + t+1 . (8)
1 + 1 + t

The shocks that index multiple equilibria are sunspots. In the usual causal interpre-
tation of the equations, small changes in expectations about the future Et t+j , perhaps
coordinated by economically unimportant events such as sunspots or Federal Re-
serve officials speeches induce jumps between equilibria t . I indicate such multiple
equilibria by thin dashed lines in Figure 6. The model can jump between any of these,
and then the dashed line indicates expected inflation after such a jump. Sunspots can
induce additional inflation volatility.

One might restore the belief that higher interest rates at least temporarily lower in-
flation by engineering a negative sunspot shock coincident with the interest rate rise. I
explore these issues below.

In this model, a Taylor rule induces instability into an otherwise stable model in
order to try to render it determinate to select a particular choice of {t+1 }. For > 1,
expected inflation blows up for all values of inflation t other than


1 + j+1
X 
i r
t = Et (vt+j vt+j ). (9)
1 + 1 +
j=0

Equivalently, taking Et Et1 ,


1 + j+1
X 
i r
t = (Et Et1 ) (vt+j vt+j ). (10)
1 + 1 +
j=0

The economy jumps by an expectational error t just enough so that expected inflation
does not explode.

This method of inducing determinacy is not entirely uncontroversial (Cochrane

(2011)). That controversy is one motivation to look for another theory. But that con-
troversy is not central here. If, because of the zero bound or other reasons, the interest
rate follows a peg or passive policy < 1, we are back to the conclusion of (8): stability
plus indeterminacy.
16 COCHRANE

3.1.3 Fiscal theory of the price level

To show how the fiscal theory of the price level enters this kind of model in the simplest
way, I specify one-period or floating-rate debt. Then the equation stating that the real
value of nominal debt equals the present value of primary (net of interest) surpluses
reads
Bt1 X X 1 X
= Et Mt,t+j st+j = Et st+j = Et j st+j . (11)
Pt Rt,t+j
j=0 j=0 j=0

Here, Bt1 is the face value of one-period debt, issued at t 1 and coming due at t, Pt
is the price level and st is the real primary surplus. In the first equality, we discount
the future with a general stochastic discount factor M. In the second equality, we dis-
count the future with the ex-post real rate of return on government debt. Either of these
statements is valid in general; the latter ex-post as well. The third version specializes to
a constant real interest rate.

In general, real interest rate variation affects the present value of surpluses on the
right hand side of (11). Models with sticky prices imply variation in real interest rates.
I will argue below that such real interest rate variation is of first-order importance to
understand data, experience, and policy via the fiscal theory. However, the stability and
determinacy points are not affected by long-term debt or real rate variation, so I specify
constant real rates and short term debt to make basic points here, then generalize at a
cost in algebra below.

Multiplying and dividing (11) by Pt1 and taking innovations,

 
Bt1 Pt1 X
(Et Et1 ) = (Et Et1 ) j st+j . (12)
Pt1 Pt
j=0

In this simple setup, unexpected inflation is determined entirely by innovations to the

expected present value of surpluses.

Indeterminacy is the inability of the standard New Keynesian model to nail down
unexpected inflation with passive policy < 1, because we can always add any unex-
pected shock to the solution. Equation (12) shows that the fiscal theory of the price
level solves the indeterminacy problem. The present value of future expected surpluses
provides the anchoring of inflation expectations widely perceived by policy-makers.
 INFLATION AND INTEREST RATES 17

In sum, the fiscal theory of the price level merged with the new-Keynesian model
says that with an interest rate peg, or passive < 1 target, inflation is stable, and deter-
minate.

Fiscal theory does not mean the central bank is powerless. Far from it. In this sim-
ple frictionless model, the Fed, by setting in interest rate target it will set expected infla-
tion it = r + Et t+1 . The fiscal theory can then select which value of unexpected infla-
tion t+1 Et t+1 will occur. Unexpected inflation or disinflation revalues outstanding
government debt, which requires changes in discounted future surpluses. Monetary
policy setting of interest rate targets remains the central determinant of the path of
expected inflation.

More concretely, multiplying and dividing (11) by Pt1 and taking expectations,
 
Bt1 Pt1 X
Et1 = Et1 j st+j . (13)
Pt1 Pt
j=0

Together with
1 Pt
= Et1 ,
1 + it1 Pt 1
we see that the government can set a nominal interest rate target, and vary that tar-
get without varying future surpluses. Doing so controls expected inflation. Equation
(13) then describes the number of bonds that will be sold given the interest rate target
and expected surpluses. This is important confirmation that the action is possible: an
interest rate target will not lead to unbounded demand for bonds.

In the context of Figure 6, then, if we define monetary policy as changing nominal

interest rates with no change in the present value of future surpluses, the fiscal theory
gives us a completely money-free assurance that the Fed can target nominal interest
rates without controlling surpluses, and the fiscal theory plus new-Keynesian model
picks the = 0 solid line of Figure 6 as the uniquely determined response to a monetary
policy shock.

Figure 6 includes the case of less price stickiness, = 1 in place of = 1/2. Again,
dynamics happen more quickly. But in this case, dynamics smoothly approach the
frictionless limit, in which it = r + Et t+1 and expected inflation rises immediately to
18 COCHRANE

match the rise in nominal interest rate.

3.2 Language

The language used to describe dynamic properties of economic models varies, and I
can dispel some remaining confusion by being explicit about language.

I use the words stable and unstable in their classic engineering sense, to refer to
the underlying dynamic system. A scalar system zt+1 = Azt + t+1 is stable if |A| < 1
and unstable if |A| > 1. Authors often use stable to mean the opposite of volatile,
but they are distinct concepts. A stable system with large shocks can display lots of
volatility.

I use the word determinate to mean that an economic model only has one equi-
librium. Stability and determinacy are also distinct concepts, frequently confused.

A harder case concerns expectational models with roots greater than one, and a
variable that can jump, Et (zt+1 ) = Azt + vt , |A| > 1. I continue to use the word un-
stable to describe their dynamics. However, if one rules out explosions and solves
forward, zt = Et (j+1) v
P
j=0 A t+j , then one could justifiably call the equilibrium path

of {zt } stable since it always jumps just enough to forestall explosions, and samples
show future z expected to revert back after a shock. This behavior is sometimes called
saddle-path stable. I use the term stationary to describe this property of equilib-
rium paths, which is common to this case and to the stable case, using the word sta-
ble or unstable to describe the properties of the dynamic system, A. The stability
of a forward-looking system with a jump variable is qualitatively different from that of
a system with backward-looking dynamics and no jump variables.

The standard three-equation new-Keynesian model with passive < 1 monetary

policy has dynamics of the form Et (zt+1 ) = Azt + vt in which one eigenvalue of A is
greater than one and unstable, and the other is less than one and stable. One could call
such a model mixed, or develop additional language to describe the relative number
of unstable eigenvalues and expectational equations. However, in this case, we un-
controversially solve the unstable eigenvalue forward, determining one of output or
inflation from the other or equivalently determining one eigenvector. Then, the other
 INFLATION AND INTEREST RATES 19

eigenvector follows a scalar Et (zt+1 ) = Azt + vt with A < 1. I use the language stable
vs unstable and determinate vs indeterminate to describe the remaining eigen-
vector.

The data speak fundamentally about these dynamic properties of models, not the
model ingredients directly. A large class of old-Keynesian and monetarist models with
adaptive expectations and passive < 1 interest rate targets produce unstable, deter-
minate dynamics. A large class of new-Keynesian models with rational expectations
and passive interest rate targets produce stable, indeterminate dynamics, and adding
the fiscal theory produces stable, determinate dynamics. But I do not claim that all
models in these categories produce such dynamics.

4 Michelson-Morley and Occam

The simplest interpretation of the data presented in section 2 is that we are living at
something approximating an interest rate peg, or at least a prolonged period of in-
sensitivity to inflation < 1, and that inflation is stable and has low volatility in this
state. The data also suggest that reserves and short-term treasuries are essentially per-
fect substitutes when they pay the same interest rate, so an arbitrarily large balance
sheet causes no inflation.

If we accept from this experience that our models should allow inflation to be stable
and determinate at an interest rate peg, then only the new-Keynesian model with fiscal
theory of the price level is left standing.

Theories fail publicly when they predict nothing and big things happen. This was
the case in the 1970s, when unexpected stagflation broke out, and again in the early
1980s, when inflation dropped suddenly as well. Theories fail no less when they predict
movements that do not happen. That is the case now. Its just less public

Of course, one can interpret economic data and apply models in many ways.
20 COCHRANE

4.1 Offsetting instabilities?

Perhaps the economy really is unstable as in old-Keynesian/monetarist models, but

inflationary quantitative easing and deftly-timed forward guidance just offset a zero-
bound deflationary spiral, leading to the false impression of stability. An old-Keynesian
Taylor rule Fed was already thought to stabilize an unstable economy. Perhaps the new
QE and talk tools were even more effective at quashing unstable deflation than low
interest rates, ruled out by the zero bound, would have been.

Perhaps. But when the night at sea was quiet Occam suggests that the weather was
calm, rather than believe that the captain deftly steered the ship just close enough to
the hurricane of hyperinflation to avoid the great whirlpool of deflation.

4.2 Really slow unstable dynamics?

Perhaps the economy really is unstable as in old-Keynesian/monetarist models, but

the dynamics are much slower than we previously thought, and observed in previous
episodes (under this interpretation) such as the 1970s, either due to new mechanisms
or different parameters. In the extremely simple model I presented above, lower
stickier prices and wages and lower output less affected by real interest rate vari-
ation both slow down dynamics. If so, the deflation spiral is still waiting to break
out any day, even in Japan. Likewise, perhaps the long-run stability of velocity, even
at low interest differentials, is much longer-run than previously thought; velocity will
finally recover some day soon and inflation will break out.

Perhaps. But perhaps not. First, this speculation is all pretty clearly ex-post ra-
tionalization. The broad consensus of people using old-Keynesian policy models was
and remains that a deflation spiral was a danger. Many monetarists did clearly expect
quantitative easing to lead to runaway inflation.

This observation is praise, not criticism. The models clearly made those predictions.
People should commended for offering the advice that their models present. The mod-
els were also broadly consistent with data, at least outside of Japan, before interest rates
hit zero. In an unstable old-Keynesian model, when the Fed follows a > 1 Taylor rule,
equilibrium interest rates and inflation comove positively, driven by external shocks,
 INFLATION AND INTEREST RATES 21

just as the seals nose and ball move together across the pool. Data from a period with
varying interest rates dont distinguish stability vs. instability well. Similarly, MV=PY
may describe how PY is determined from M control, or it may describe how M demand
follows from P, Y, and V determined elsewhere. Until you push, its hard to tell the two
stories apart. Thats why the zero bound / QE era is so revealing.

But now that we have pushed, and the models fail, we all understand the dangers of
patching a model every time it fails.

Second, these sorts of patches to old-Keynesian models such as very sticky wages
are either not worked out analytically, and compared with macroeconomic or microe-
conomic data large job churn more broadly. Old-Keynesian models dominate policy
thinking but have vanished from academic journals, so the patches remains a sugges-
tion, not a result. The large literature on wage stckiness in new-Keynesian models does
not apply at all as I hope is clear by now, new-Keynesian models are utterly different
starting from basic properties such as stability and determinacy.

So, Occam suggests, perhaps not: perhaps the economy is stable at an interest rate
peg.

4.3 Not really a peg? Selection from future actions

Our interest rates are not exactly zero and not a true peg. I do not argue that we are
at a peg only that models which can credibly explain our experience must also have
the property that inflation is stable and determinate at a true peg. But perhaps our
experience is sufficiently different from a true peg that models which predict instability
or indeterminacy at a true peg can survive.

Zero bound is not accurate, which is why I have avoided using the term. The Fed-
eral Funds rate has remained a few basis points above zero, and European interest rates
have dipped slightly below zero. However, the main point for the models is whether the
Federal Funds rate responds more than one for one to inflation or deflation, and that
point does not require an exact zero bound.

The long period of immobile interest rates, and forward guidance that rates would
stay low for a long time, is not a formal peg. There is really no such thing as a true peg
22 COCHRANE

even apparently permanent commitments such as a gold standard are suspended, de-
valued, or revoked in some circumstances. Our central bankers have long claimed that
interest-rate stasis is one-sided, and they will raise interest rates if output and inflation
rise though so far Europe, Japan, and Swedens attempts to raise rates were quickly
abandoned. So, perhaps the difference between actual policy and an interest rate peg
is large enough that our models need not display stability or determinacy at a peg in
order to be consistent with experience.

How? In an active literature such as Werning (2012) or Eggertsson and Mehrotra

(2014), expectations of possible future active, destabilizing, Taylor rules take the place
of current Taylor principle responses to select equilibria while interest rates are zero in
new-Keynesian models. In these papers, eventually, either deterministically or stochas-
tically, we will leave the zero interest rate range. Modelers specify the destabilizing ver-
sion of the Taylor rule to select a unique locally-bounded equilibrium around the in-
flation target in that future state. Modelers tie multiple equilibria during the zero-rate
period to the following equilibria, and thereby eliminate indeterminacies during our
apparent peg.

Here is a concrete example. From time t = 0 to t = T , there is a negative natural rate

shock, vtr = 2%. At time t = T the natural rate shock passes so vtr = 0. The Fed follows
a constrained active Taylor rule, so together interest rates and inflation follow

it = max [ + (t ), 0] (14)

t = (1 + )Et t+1 (it vtr ) (15)

Figure 7 shows the corresponding equilibria. The thick line in the middle is the
selected equilibrium. It fairs in to the inflation target = 2% at t = T . The alternative
equilibria are selected even in the stable region, for t < T , by the fact that they diverge
from the inflation target for t > T .

(In this simulation, I use the active = 2 but zero-bound constrained form of the
Taylor rule (14) throughout, even for t < T . As a result, equilibria which start with
high enough inflation to free interest rates from the zero bound are already unstable at
t < T . If one specifies a pure peg until t = T , then these equilibria are stable util t = T ,
 INFLATION AND INTEREST RATES 23

and explode only afterwards. Werning (2012) and Cochrane (2014c) analyze this case.)

This equilibrium-selection scheme has many troubles. Anchoring of inflation ex-

pectations does not occur because the Fed is expected to stabilize inflation around the
inflation target, but because the Fed is expected to destabilize inflation should it di-
verge from the target. More deeply, equilibria in which inflation udershoots the target
return back to zero inflation and zero interest rates. They are locally unstable around
the target and thus locally determinate, but not globally unstable and thus globally
indeterminate. The rationale for ruling them out is tenuous.

Figure 7 also illustrates a predictive failure of this model, highlighted by Werning

(2012) and Cochrane (2014c). It predicts a jump to deflation at t = 0 when the shock
hits, which then rapidly improves. This did not happen.

Figure 7 likewise illustrates some of the policy paradoxes highlighted by Werning

(2012), Wieland (2015), Cochrane (2014c). Forward stable means backward unstable,
so small change in the inflation target on the right move the left end of equilibrium
inflation around a lot. Forward guidance has large effects. Moreover, the further in the
future the end of the shock and promise occurs, the larger the effect at time 0, and as
price stickiness is reduced, the dynamics happen faster. Less price stickiness implies
larger deflation and greater effect of such promises.

Here too, this anchoring story is also an ex-post patch to a theory, given pretty uni-
form expectation and analysis that the zero bound and passive policy must result in
additional sunspot volatility, before it didnt.

Do people really pay that much attention to promises by Federal Reserve officials
and distinguish them from the routinely broken promises of other government func-
tionaries Treasury secretaries who routinely promise to end deficits one year after
their presidents term of office? Does all concrete action of monetary policy really van-
ish, leaving only expectations of far-future off-equilibrium threats behind? Can mon-
etary policy really speak loudly with no stick? Did Japan really avoid deflation in 2001
because people expected some sort of explosive promises around a 2% inflation target
to emerge and select in equilibria, maybe sometime in 2025 when Japan finally exits
zero rates?
24 COCHRANE

Inflation
10

8 vr = -2 vr = 0
6

2
percent

-2

-4

-6

-8

-10
0 5 10 15
time
Interest rate
10

8 vr = -2 vr = 0
7

5
percent

-1
0 5 10 15
time

Figure 7: Selection by future and contingent Taylor rules. Top: Inflation. Bottom: Inter-
est rates. Solid line is the selected equilibrium, dashed lines are alternative equilibria,
indexed by inflation at time t = T . There is a natural rate shock v r = 2% from time
t = 0 to t = T = 10. The Fed follows a rule it = max [ + (t ), 0] and the simple
new-Keynesian model reduces to t = (1 + )Et t+1 (it vtr ). = 1, = 1/2,
= 2.
 INFLATION AND INTEREST RATES 25

Even Janet Yellen Yellen (2016) expresses a deep distrust that promises of future Tay-
lor rules anchor inflation today:

...how does this anchoring process occur? Does a central bank have to
keep actual inflation near the target rate for many years before inflation ex-
pectations completely conform? ...Or does ...a change in expectations re-
quir[e] some combination of clear communications about policymakers in-
flation goal, concrete policy actions,..., and at least some success in moving
actual inflation toward its desired level ...?

Moreover, her language clearly states that anchoring results because a Taylor rule will,
in the future, stabilize inflation around the target, in the old-Keynesian tradition, not
destabilize inflation to produce determinacy as shown in Figure 7. If she doesnt believe
the dynamics of Figure 7, why should we?

This search for far-future off-equilibrium expectations-anchoring might be more

plausible if a much simpler solution were not at hand. The Fiscal theory picks one
equilibrium of the stable new-Keynesian model directly. Expectations are anchored by
the present value of future surpluses, not by what people believe the Fed might do to
select equilibria after a jump to a higher-interest rate regime. Though stable, since it is
unique, the fiscal-theory plus new-Keynesian model does not display policy paradoxes.
Each equilibrium corresponds to an innovation in the present value of surpluses, and
moving to a larger inflation at time t requires a larger change in the present value of
surpluses at time t. The model also has a smooth frictionless limit. Occams suggestion
is pretty clear!

4.4 Sunspot volatility?

Perhaps the new-Keynesian prediction of higher inflation volatility under passive pol-
icy can be patched. The solution to the model with > 1 is not the same as the so-
lution with < 1, plus some extra sunspot shocks. By (10), the variance of expecta-
tional shocks is not zero under active policy. Rather, there is a jump each period,
just enough to offset that periods shocks. Therefore, it is possible for the determinate
solutions of the model with > 1 to exceed the volatility of the t = 0 solutions of
26 COCHRANE

the model with < 1, so small sunspot shocks could still leave the model less volatile
under passive than active policy.

As a concrete example, Appendix section 19.1 works out the volatility of inflation
with AR(1) natural rate shocks vtr = vt1
r + rt , in the active, forward-looking specifica-
tion (9), and with = 0 and t = 0 in the passive, backward-looking specification (8).
For = 1, the ratio of the two inflation variances is

backward 2 (2 + ) [2 (1 ) + ( 1)]2
= . (16)
forward 2 (2 ) 3

For = 1.5, & 0.5 is sufficient for the ratio to be less than one. In the limit of very
persistent shocks,
backward 2
lim 2
= ( 1)2
1 forward

This forms a very simple counterexample: Passive policy an interest rate peg can
display less inflation volatility than active policy, if shocks are persistent, the active pa-
rameter is not too large, and if sunspot shocks t are not too large.

In this specification, higher values of the active parameter result in less inflation
volatility. Optimal-policy more generally exercises point in that direction (Woodford
(2003)). To the extent that the Federal Reserve followed good high- policy in the past,
the chance of seeing lower volatility at the = 0 bound is reduced. And even then,
sunspots being ephemeral, we dont have much of a reason to limit their volatility. On
the other hand, sunspots being ephemeral, there werent any sunspot shocks is an
irrefutable ex-post explanation of quiet. Phlogiston comes and goes as it pleases.

In any case, this is a novel possibility which as far as I know has not yet been ad-
vanced to rescue the prediction of higher inflation volatility at the zero bound. The
new-Keynesian literature, summarizing larger and more realistic models, clearly warns
that passive < 1 monetary policy causes inflation volatility. That proposition is one of
the models most central empirical claims, explaining the greater volatility of the 1970s
vs. the 1980s. If we throw out the prediction of higher volatility under passive policy
in the 2010s, we are hard pressed not also to throw out the central prediction of higher
volatility under passive policy in the 1970s.

For example, Clarida, Gal, and Gertler (2000), who found < 1 in the 1970s, > 1
 INFLATION AND INTEREST RATES 27

in the 1980s, attribute the reduction of inflation volatility to that fact, writing (p. 149)

...the pre-Volcker rule leaves open the possibility of bursts of inflation

and output that result from self-fulfilling changes in expectations. ... On
the other hand, self-fulfilling fluctuations cannot occur under the estimated
rule for the Volcker-Greenspan era since, within this regime, the Federal Re-
serve adjusts interest rates sufficiently to stabilize any changes in expected
inflation.

Again on p. 177, they write

Finally, we have argued that the pre-Volcker rule may have contained
the seeds of macroeconomic instability that seemed to characterize the late
sixties and seventies. In particular, in the context of a calibrated sticky price
model, the pre-Volcker rule leaves open the possibility of bursts of inflation
and output that result from self-fulfilling changes in expectations.

Benhabib, Schmitt-Grohe, and Uribe (2001) likewise write (p. 167)

Perhaps the best-known result in this literature is that if fiscal solvency

is preserved under all circumstances,[ i.e. passive fiscal policy] ... a passive
monetary policy, that is, a policy that underreacts to inflation by raising the
nominal interest rate by less than the observed increase in inflation, desta-
bilizes the economy by giving rise to expectations-driven fluctuations.

They write again in Benhabib, Schmitt-Grohe, and Uribe (2002), summarizing a grow-
ing body of theoretical work,

Taylor rules contribute to aggregate stability because they guarantee the

uniqueness of the rational expectations equilibrium, whereas interest rate
feedback rules with an inflation coefficient of less than unity, also referred
to as passive rules, are destabilizing because they render the equilibrium
indeterminate, thus allowing for expectations-driven fluctuations.
28 COCHRANE

(Both sets of authors use stability to mean low volatility, not as I have used the
term.) Benhabib, Schmitt-Grohe, and Uribe (2002) survey many other similar opin-
ions, along with policy prescriptions to avoid the zero inflation state, all motivated by
the prediction of extra volatility at that state.

Indeed, without extra sunspot volatility, low inflation is welfare-improving in this

model. It is both Friedman-optimal, and reduces pricing distortions. Sunspot volatility
is the main reason all of these authors have for the great effort to find policies that avoid
the zero bound and return us to permanently higher inflation.

Occam says, lets take this wide reading of the model seriously. If, indeed, this model
says that passive < 1 policy gave rise to sunspot fluctuations and greater inflation
volatility in the 1970s, while active > 1 policy stabilized inflation by ruling out sunspot
equilibria in the 1980s, if indeed Benhabib, Schmitt-Grohe, and Uribe (2002) and the
rest of the iceberg of which they are the tip has spent 20 years since Japanese rates hit
zero looking for ways out of an otherwise desirable circumstance because the models,
broadly construed, predict additional inflation and output volatility at an interest rate
peg, then we should take that prediction, and its failure, seriously.

The alternative, I suppose, is to patch up the model by finding some other feature
of current policy that is eliminating sunspot shocks, but did not eliminate those shocks
in the 1970s. Indeed, that is the feeling of much discussion that expectations are an-
chored. But anchored by what? As above, the feeling that sooner or later interest rates
must rise to strongly positive values, then the Fed will undertake active policy, and
somehow avoid the theoretical problems of active policy, and people believe all this,
seems strained. Just why didnt that work in the 1970s? The 1970s did not lack from
forward-guidance promises by Federal Reserve officials! We even had those cute little
WIN (whip inflation now) buttons. If anchoring was going to work this time, just why
did researchers not know that fact, and loudly opine not to worry about the zero bound?

4.5 Occam summary

In sum, ex-post elaborations and modifications may be correct. We should learn, and
adapt models as we gain more data. But that work is suspiciously complex, and clearly
 INFLATION AND INTEREST RATES 29

ex-post patches. One sniffs ether drag and epicycles, not Bayesian updates.

Occam suggests, perhaps not. Given that we have a simple, straightforward model,
which neatly accounts for the facts without patching, and which is much cleaner theo-
retically no paradoxical policies, and a smooth frictionless limit perhaps it is instead
the right answer. The only reason its controversial, really, is just how deep the doctrines
are that this model overturns: inflation can be stable at an interest rate peg.

The remaining doubt may be the plausibility of fiscal anchoring given large debts
and deficits. That discussion comes below. The point here is theoretical simplicity.

5 Quantitative easing and monetarism

Quantitative easing and monetarism matter here for three reasons. First, is it possible
that inflation really is unstable at a stuck interest rate, but the Feds inflationary QE
operations deftly offset a deflation spiral? (Im not aware of a parallel argument that
MV=PY somehow addressed indeterminacies, so Ill leave that out.)

Second, if so, is hyperinflation around the corner? To control inflation once it starts
rising and interest rates leave the zero bound, must the Fed quickly return to the pre-
2007 policy configuration with a much smaller balance sheet? Must it go further, and
eliminate interest on reserves, returning to interest-rate management by rationing re-
serves rather than by varying the interest paid on reserves?

Third, the question has become a deep one: What is the fundamental determinant
of inflation? Is inflation determined, fundamentally, from yesterdays inflation, an in-
terest rate target and a Phillips curve (old-Keynesian)? Is inflation determined, funda-
mentally, from > 1 and Fed off-equilibrium selection threats (new-Keynesian?) Is
inflation determined, fundamentally from the fact that the government requires peo-
ple to pay taxes with money (fiscal theory)? Is inflation determined, fundamentally,
from MV=PY, a restricted supply of a special asset needed to make transactions? The
last classic view needs an airing.

Quantitative easing has two parts: The Fed buys bonds or other assets, and issues
reserves. Here I consider the question whether larger reserve supply is inflationary. In
30 COCHRANE

traditional monetarist thought, the M in MV=PY drives output Y and inflation P, and
whether the M came from buying short-term Treasuries, long-term Treasuries, mort-
gage backed securities, or from buying nothing from helicopter drops makes no
difference. Commenters who refer to QE not as drying up bond markets, but as in-
jecting liquidity echo monetarist thought.

Much of the current QE discussion takes on a diametrically opposite view: The lia-
bilities (reserves) are irrelevant, but QE works by affecting segmented markets for the
assets. Its the bonds, not the money. Since the economic issues are smaller whether
bond purchases affect interest rates by a few tens of basis points, not the sign and sta-
bility of inflation I postpone that question. Complex QE buying mortgage backed
securities, long term bonds, or stocks is equivalent to an open market operation of
short-term debt for reserves plus a twist of long term debt, MBS, etc. for short term
debt. I think here just about the traditional swap of short term debt for reserves, and
leave the composition of the assets on the Feds balance sheet for later.

Monetarist thought took a back seat during the interest-rate targeting period start-
ing in 1982. However, when Japan hit zero interest rates in the 1990s, the idea came
back quickly. Ben Bernanke advocated the view most prominently, among other alter-
native policies (see the review and fascinating discussion in Ball (2016)).

In simple terms, monetarists think about interest rate targets as just another way to
control money supply. Facing M d (i), if you set i and let M d determine M , or if you set
M s and let M s = M d (i) doesnt really matter. An interest rate at zero is not particularly
meaningful. That case means the Fed can no longer control money supply via an inter-
est rate target. But nothing stops the Fed from printing up money directly, and letting
MV=PY do its magic.

The behavior of velocity or money demand at zero interest rates is the stumbling
block to this line of thought. Monetarist thought emphasizes the idea that velocity is
stable, at least in the long run. Even at zero interest rates or our current situation
that reserves pay even more than Treasuries even if velocity V decreases somewhat, it
will soon bounce back and more M will lead to more PY.

The contrary view is that at zero interest rates, or zero interest costs due to inter-
est on reserves, money and short-term bonds become perfect substitutes. Velocity be-
 INFLATION AND INTEREST RATES 31

comes a correspondence, an upper bound, not a function of interest rates. MV = PY

becomes V = PY/M. People are perfectly happy to hold reserves, directly or indirectly
via bank accounts, in place of short-term treasuries. Open market operations have no
more effect on spending than change operations, in which the Fed gives everyone two
$10 bills for each $20.

Absent data, there really was no way to tell these views apart. Now there is, and the
experiment is nearly as decisive as the stability of an interest rate peg.

20

10

3 2007-2016
Reserves M/PY (Percent)

1 1984-2000

0.5

0.3

0.2

0.1
fit 2000-2007
2000-2007
0.05
0 2 4 6 8 10 12
Fed Funds Rate - IOER

Figure 8: Reserves vs opportunity cost. Y axis: total reserves (Fred series WRESBAL) at
end of quarter divided by nominal GDP; log scale. X axis: effective Federal funds rate
(Fred FF) less interest on excess reserves (Federal Reserve website policy rates). Sample
1984:1-2016:2. Line fit by OLS 2000:1-2006:4

Figure 8 presents reserves, scaled by nominal GDP, as a function of their oppor-

tunity cost, the difference between the effective Federal Funds rate (the rate at which
banks can lend out reserves) and the interest on excess reserves at the Fed. You see a
steady decline in reserves from 1980 to 2000, as fewer bank liabilities required reserves
and banks became better at avoiding excess reserves. You also see a negatively sloped
32 COCHRANE

curve in the periodic recessions. Following tradition, Ill just call this plot a demand
curve without further ado. The dashed line is fit to the 2000-2007 period, and gives a
conventional semi-elasticity log(reserves/PY) = constant - 0.094 (interest rate).

What does reserve demand do at zero opportunity cost? As Figure 8, shows, we have
now run this out-of-sample experiment, on a grand scale note the numbers on the log
scale y axis. Reserves have increased by two orders of magnitude from 0.1% of GDP to
15% of GDP with no effect on inflation or nominal GDP.

The answer seems unavoidable: Reserve demand is a correspondence when re-

serves pay market interest rates; reserves and short-term debt are perfect substitutes;
there is no tendency for velocity to revert to some stable previous value; there is no
upper bound on the demand for liquidity at zero opportunity cost; arbitrary quanti-
ties of zero-cost reserves do not cause inflation.
18 70

17

65
16

2007-2016
15 M1, 1978-2000
60
14
M2 / PY, Percent
M1 / PY, Percent

13 2007-2016 M2, 1978-2000

55
12

11
50
2000-2007
10
2000-2007

45
0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16
3 Month Treasury Rate 3 Month Treasury Rate
80

70 2007-2016

60
MZM / PY, Percent

50
2000-2007
45

40

35

MZM, 1978-2000
30

25
0 2 4 6 8 10 12 14 16
3 Month Treasury Rate

Figure 9: M/PY vs three month treasury bill rate, using M1, M2, and MZM.
 INFLATION AND INTEREST RATES 33

One may object that reserves are not the relevant M in MV=PY. Figure 9 presents M1,
M2, and MZM (M2 less small-denomination time deposits plus institutional money
funds), as percentages of nominal GDP PY, versus the three-month Treasury bill rate.
(Since many components of these aggregates pay interest, the three-month Treasury
bill rate is not a good measure of their opportunity costs, but Im both following tra-
dition and keeping it simple.) Each aggregate has increased since 2007, but less, pro-
portionally, than reserves have increased. M1 has increased from about 9% of GDP to
almost 18%, a bit less than doubling but not rising by a factor of 100. M2 has increased
from 50% of GDP to almost 70% of GDP, only a 60% rise. Depending on what starting
point you choose, MZM has risen the least, by 10 to 20 percentage points of GPD, or 20
to 40 percent overall.

These are still substantial increases, which if velocity were stable should result in
equiproportionate rises in nominal income, and eventually the price level.

But even making these plots grants too much. How much inflation a monetarist
model predicts for the recent period is beside the point. The point is whether arbitrary
amounts of reserves, exchanged for short-term treasuries, cause any inflation. Even if
one believes that M2 V = PY (say), and claims that a monetarist view does not predict
inflation in the current period because reserves did not leak in to M2, that fact only
emphasizes that arbitrary quantities of reserves are not inflationary, precisely because
they did not leak into M2. That leakage is, in this view, a central part of the transmission
mechanism. When banks are holding trillions of dollars of excess reserves, the money
multiplier ceases to operate. Reserve requirements are an inequality constraint, not
an equality constraint. So, to argue there is no inflation because M2 did not rise is
precisely to admit that arbitrary quantities of interest-bearing reserves, corresponding
to arbitrarily lower quantities of interest-bearing treasuries, are not inflationary.

Figure 8 really only makes a secondary point: What would happen if reserves were
to leak to larger increases in M1, M2, or MZM? Figure 8 suggests that these aggregates
display the same behavior as reserves, only on a smaller (so far) scale they happily
crawl up the vertical axis. There is nothing in their behavior so far to suggest that this
correspondence could not reach the astonishing level that banks willingness to hold
reserves at the expense of treasuries have reached.
34 COCHRANE

Another view admits that the conventional money multiplier is inoperative re-
serves are so hugely beyond required reserves that bank money creation is uncon-
strained by reserves. However, in this view regulatory leverage and equity constraints
still bind. Banks would create more money by lending out more, if only they werent
held to capital requirements or leverage requirements. This view requires some sort of
upward sloping supply of capital, otherwise banks would retain earnings and undo the
constraint. Nonetheless, the secondary interpretation of Figure 8 suggests that even
then, were capital or leverage requirements loosened allowing banks to create more
money, that creation would have little impact on inflation through the MV=PY chan-
nel.

Looking back at an 80 year controversy, one wonders why the stability of veloc-
ity, even at zero interest cost, and the perfect substitutability of treasuries for reserves,
was so controversial. One answer may be that for most of that period, there was no
coherent, simple, economic theory of the price level that could hold in that circum-
stance. In a monetarist world, strike MV=PY and nothing ties down P. Keynesian and
new-Keynesian economics eventually had a Phillips curve to describe price dynamics,
but no simple satisfactory alternative model of price level determination, at least to an
economist longing for the transparency of MV=PY. So, it is natural to cling to the idea
that velocity must be stable, as otherwise the price level would be indeterminate. But
now we have an equally simple theory the fiscal theory that ties down the price level
when money and bonds are perfect substitutes, and a long period of apparent empir-
ical validation. Economic beliefs survive for their usefulness as well as, or sometimes
despite lack of, their theoretical consistency and empirical validation. This one has
now lost the former, as well as latter, impulses.

My conclusion that abundant interest-bearing reserves will not cause inflation does
not address many objections to the Feds large balance sheet. The Feds liabilities cause
no problems for inflation, but its assets may cause problems. One may object to the
Feds purchases of long-term bonds, of mortgage-backed securities, and of other cen-
tral banks purchases of corporate bonds (ECB) and even stocks (BOJ), both on grounds
that independent central banks should not try to influence directly risky asset prices,
or on political economy grounds that such policies constitute credit allocation bet-
 INFLATION AND INTEREST RATES 35

ter done (if at all) by politically-accountable Treasuries. In this analysis, the Treasury
could just as well supply fixed-value floating-rate electronically-transferable debt, i.e.
reserves, and allow people and businesses, not just banks, to hold them. (For details
on this proposal, see Cochrane (2015)). If the Treasury does so, the Fed itself would no
longer need to act as the worlds largest money market fund, transforming longer-term
government debt into floating-rate government debt.

6 Do higher interest rates raise or lower inflation?

If we grant that inflation can be stable under an interest rate peg, that observation sug-
gests that raising interest rates should sooner or later raise inflation, contrary to the
usually presumed sign. If so, central banks will partly cause the inflation they wish to
forestall though in the event will likely congratulate themselves for their prescience.
It also implies that current low inflation is in part due to pedal misapplication cen-
tral banks, by keeping interest rates low, partly caused the low inflation that they were
trying, wisely or not, to prevent.

This empirical suggestion needs theoretical validation, beyond the too-simplified

models of section 3. Do sensible models produce a long-run rise in inflation when in-
terest rates rise? Then, if raising interest rates eventually leads to higher inflation, per-
haps raising interest rates temporarily reduces inflation? What is the minimal simple
economic model that produces this classic belief? (The qualifiers simple and eco-
nomic are important.)

This section is rather long, because the important results are negative. It only takes
one model to show a positive result. The interesting negative result here is that a suite
of sensible modifications one might adduce to provide the desired sign do not work
sticky prices, monetary distortions and even backward-looking Phillips curves. One
simply cannot say, for example, that sure, the Fisher relation means that raising in-
terest rates raises inflation, but sticky prices overturn that result. They dont. A novel
fiscal theory argument with long-term debt produces the desired negative sign, though
rather deeply changes ones views of just what monetary policy is and how it works. The
alternative possibility is that monetary policy necessarily relies on complex or non-
36 COCHRANE

economic ingredients, or that raising interest rates really does raise inflation, even in
the short run.

6.1 Pure Fisher

The simplest starting point is a frictionless model incorporating a Fisher relation

it = rt + Et t+1

and the government debt valuation equation

Bt1 X
= Et j st+j (17)
Pt
j=0

(One can flesh these out as the two important equilibrium conditions of a complete
general equilibrium model with a constant endowment. See Cochrane (2005).) This
model is simple, economic, and stable under an interest rate peg.

Figure 10 presents the impulse-response function for this case. In this model, rais-
ing the interest rate produces an immediate and complete rise in expected inflation.

The impulse-response function measures the path of {Et t+j } on the date t that the
interest rate rise is announced not necessarily the same as the day that interest rates
actually rise. Thus, the Fisher equation alone allows in the impulse-response function
an arbitrary one-period jump in inflation, upwards or downwards, coincident with the
announcement of the policy change. The fact that it only ties down expected inflation
has no impact on the impulse-response function which plots the path of expected
inflation at other dates.

The top panel of Figure 10 shows some possible impulse-response functions when
the interest rate rise is announced at the same time as it occurs. This is the usual as-
sumption of VAR analysis. However, many real-world interest rate rises are announced
long in advance, and we want to know the effect of such anticipated monetary policy as
well. The bottom panel of Figure 10 shows possible impulse-response functions when
the rise is announced three periods before it occurs.
 INFLATION AND INTEREST RATES 37

1.5

interest rate i
1

0.5
Percent response

inflation
0

-0.5 , with fiscal shocks

-1

-1.5
-3 -2 -1 0 1 2 3
Time
1.5

, with fiscal shocks interest rate i

1

0.5
Percent response

inflation

-0.5

-1

-1.5
-5 -4 -3 -2 -1 0 1 2 3
Time

Figure 10: Response of inflation to a permanent interest rate increase. Frictionless

model it = r + Et t+1 . Top: the rise is announced and implemented at time t = 0.
Bottom: The rise is announced at t = 3. The main inflation line assumes one-period
debt and no change in fiscal policy coincident with the monetary shock. The dashed
with fiscal shocks lines add a change in primary surpluses coincident with the an-
nouncement of a higher interest rates.
38 COCHRANE

Now, any unexpected inflation, which revalues government debt, must correspond
to a revision in expectations of future surpluses. In (12), we took innovations of (17),

Bt1 Pt1 X
(Et Et1 ) = (Et Et1 ) j st+j .
Pt1 Pt
j=0

I think of monetary policy as a change in interest rates with no change in fiscal

policy. This additional definition picks the central, = 0 equilibrium as the effect of
monetary policy, and it is to raise inflation throughout, with no difference between ex-
pected and unexpected policy.

How can we generate a temporary negative response? I start with the obvious eco-
nomic ingredients sticky prices and money. I then consider fiscal responses, to see if
we can choose one of the negative paths.

7 Sticky prices

It is natural to hope that by adding sticky prices, the simple frictionless Fisherian model
will display a temporary negative inflation response.

I use the standard optimizing sticky-price model,

xt = Et xt+1 (it Et t+1 ) (18)

t = Et t+1 + xt (19)

where xt denotes the output gap, it is the nominal interest rate, and t is inflation.

The solution of this model for a given equilibrium interest rate path is derived in the
Appendix. Inflation and output are two-sided geometrically-weighted distributed lags
of the interest rate path,



j j2 Et+1 it+j j
X X X
t+1 = it + 1 itj + + 1 t+1j (20)
1 2
j=1 j=1 j=0
 INFLATION AND INTEREST RATES 39

j j2 Et+1 it+j

X
X
xt+1 = 1 1
1
1
1 itj + 1 2
1 2
j=0 j=1

j
X
+ (1 1
1 ) 1 t+1j , (21)
j=0

where
q
(1 + + ) + (1 + + )2 4
1 = >1 (22)
q2
(1 + + ) (1 + + )2 4
2 = <1 (23)
2

Here, t+1 , with Et t+1 = 0, is an expectational shock indexing multiple equilibria. (In
the frictionless and some simple sticky price models, t+1 = t+1 Et t+1 . Here, since
inflation t+1 already responds to interest rate shocks at time t + 1 in the t+1 = 0
solution, t+1 represents an additional iid disturbance.)

7.1 Basic impulse-response function

Figure 11 presents the response of inflation and the output gap to a step function rise in
the interest rate, using (20)-(21), and choosing the basic solution 0 = 0. Throughout,
unless otherwise noted, I use parameters

= 0.97, = 0.2, = 1. (24)

The solid lines of Figure 11 plot the responses to a pre-announced interest rate rise.
The dashed line plots the response to an unexpected interest rate rise. Announced and
surprise interest rate paths are the same after the announcement day. More generally,
the response to this policy announced at any time before zero jumps up to match the
anticipated-policy reaction on the day of announcement.

Inflation rises throughout the episode. Mathematically, that is a result of a two-

sided moving average with positive weights in (20).

Output declines around the interest rate rise. When the nominal interest rate is
40 COCHRANE

1 i
interest rate i
x
0.8
inflation
0.6

0.4
Percent response

0.2

-0.2

output gap x
-0.4

-0.6

-0.8

-1
-4 -2 0 2 4 6
Time

Figure 11: Response of inflation and output to a step function interest rate change in the
standard IS - Phillips curve new-Keynesian model. The solid lines show the response
to an expected change. The dashed lines show the response to an unexpected change.
Paramters = 0.97, = 0.2, = 1.

higher than the inflation rate, the real rate is high. Output is low when current and
future real interest rates are high via intertemporal substitution. Equivalently, the for-
ward looking Phillips curve (19) says that output is low when inflation is low relative to
future inflation, i.e. when inflation is increasing.

Output eventually rises slightly, as the steady state of the Phillips curve (19) with
< 1 gives a slight increase in the level of output when inflation increases permanently.
Using = 1, there is no permanent output effect, and all graphs are otherwise visually
indistinguishable. The positive inflation effect does not require a permanent output
effect.

(Solving the Phillips curve (19) forward, one obtains inflation as a positive function
of future output gaps. One might conjecture therefore that higher inflation requires
positive output gaps in the long run. However, as rises, and long-run output gaps
 INFLATION AND INTEREST RATES 41

decline, the j Et t+j term becomes more important. The solutions smoothly approach
the case that long-run output gaps are zero, approaching from below, and expected
inflation is one.)

In sum, this simple standard model gives a smoothed Fisherian inflation response
to interest rate changes. One might have hoped that price stickiness would deliver the
traditional view of a temporary decline in inflation. It does not.

The model does, however, generate the output decline that conventional intuition
and most empirical work associates with monetary policy tightening. It therefore sug-
gests a novel picture of monetary policy. Raising interest rates to cool off a booming
economy, and lowering interest rates to stimulate a slow economy may make sense.
Doing so just has a different effect on inflation than we might have thought. It paints
a picture, not unlike recent experience and Fed statements such as Yellen (2016), in
which monetary policy is primarily concerned with manipulating output, not inflation.

The sign is not affected, and magnitudes not greatly affected, by changes in the pa-
rameters. There isnt much you can do to an S shape. The parameters and enter
together in the inflation response. Larger values speed up the dynamics, smoothly ap-
proaching the step function of the frictionless model as their product rises. Larger val-
ues of the parameter slightly slow down the dynamics. Larger gives larger output
effects with the same pattern.

Expected and unexpected policy have similar responses because the interest rate
shock it Et1 it does not appear as a separate right hand variable in the models solu-
tions (20)-(21), as it does in information-based Phillips curves such as Lucas (1972). As
a result, in this class of models, expected monetary policy matters.

VARs often focus on the responses to unexpected policies, both out of historical
tradition based on models such as Lucas in which only unexpected monetary policy
matters, and in order to try to identify the small exogenous movements in monetary
policy. But our Fed telegraphs its intentions, often far in advance. So for policy and
historical analysis, it is important to ask of models what is the effect of an expected
policy change.

Output and inflation move ahead of the expected policy change. This fact reminds
42 COCHRANE

us that forward guidance matters, and that outcomes are affected by expectations,
even when those expectations do not bear out.

7.2 Mean-reverting and stairstep rates

Empirical impulse-response functions usually find that the response of interest rates to
an interest rate shock is mean-reverting, not a pure random walk as is the conceptual
experiment of Figure 11. To think about that case, Figure 12 plots responses to an AR(1)
interest rate shock.

1 i

x

0.8

0.6
Percent response

0.4

0.2

-0.2

-0.4

-0.6

-4 -2 0 2 4 6
Time

Figure 12: Response of inflation and output to a mean-reverting interest-rate path.

Dashed lines are the response to an unexpected change. Solid lines are the response
to an expected change.

One might have hoped that, since an expected rise in interest rates raises inflation,
the expected declines in interest rates set off by the initial shock might have a contrary
effect, depressing inflation or maybe even giving rise to a negative movement. Alas,
that hope does not bear out. The responses in Figure 12 are similar to those of Figure
 INFLATION AND INTEREST RATES 43

11 in the short run, with a long-run return to zero.

Figure 12 serves as an important reminder though: VARs that estimate transitory

responses of interest rates to interest rates do not give us evidence on the long-run
Fisher hypothesis. The zero bound experience tells us something that we could not
observe in the transitory interest rate changes typical of the previous era.

1.2 i

x

0.8
Percent response

0.6

0.4

0.2

-0.2

-0.4

-4 -2 0 2 4 6 8 10 12 14 16
Time

Figure 13: Response of inflation and output to a stairstep interest-rate path

Federal Reserve tightening typically takes the form of a well-anticipated steady set
of stair step interest rate rises. Figure 13 presents the effects of such a policy. The re-
sult is qualitatively predictable from the other figures, though the smoothness of the
inflation and output effects is noteworthy.

7.3 Taylor rules disclaimer

By solving for inflation and output given the equilibrium interest rate sequence {it }
I appear to assume that the Fed follows a time-varying peg without Taylor-rule re-
44 COCHRANE

sponses. This is not the case.

The series {it } represents a conjectured path for the equilibrium interest rate. Equa-
tions (20)-(21) tell us that if an equilibrium has an interest rate sequence {it }, then
its inflation and output paths {t , xt } must follow (20)-(21). The impulse response
functions are really the response of equilibrium inflation and equilibrium output to a
policy change that also produces a step function rise in equilibrium interest rates. In
it = t + vti , the disturbance {vti } is not the same thing as the interest rate {it }. In fact,
with > 1 they can even have a different sign as we will see shortly.

To match VARs, and to understand how inflation and output would respond if the
Fed engineers a step function path of the nominal interest rate, this is exactly the ques-
tion one wants to answer. Just how the Fed engineers the equilibrium interest rate path
is not important, so long as it can do so.

Yes, the Fed can follow a time-varying or state-varying peg, and simply set the nom-
inal interest rate path. But the Fed can also follow a Taylor rule with an active inflation
response. For example, if the Fed follows a policy rule it = it + (t t ) + vti , if the
Fed changes monetary policy by shocks to {it }, {t }, {vti }, and if, on solving the the
model, the equilibrium expected interest rates it follow the paths shown in Figures 11
- 13, then output and inflation also follow the paths shown in those figures. (Simply
picking shocks t equal to the plotted inflation response will do.)

Werning (2012) innovated this clever idea of first finding equilibrium inflation and
output given equilibrium interest rate paths, and then constructing the underlying Tay-
lor rule. It simplifies the calculation, and makes it more general, as multiple underlying
policies may give the same equilibrium interest rate path.

Below, I exhibit some underlying Taylor rule assumptions behind stairstep equilib-
rium interest rates. The important point here is that by calculating and plotting the
response to a given interest rate path I do not make any assumptions about active vs.
passive policy, and in particular I do not assume a time varying peg with = 0. The
calculations apply as much to active, > 1 policy specifications as they do to passive
specifications.

As previewed in Figure 10, there also remains the possibility of alternative equilibria,
 INFLATION AND INTEREST RATES 45

indexed by a one-time unexpected shocks on the date on which the new policy is an-
nounced, and each with a different fiscal consequence. Active Taylor rules and passive
fiscal policy may act by selecting one of those equilibria, and then inducing the required
change in fiscal surpluses. I return to this issue below also. Again, the quest is to see
if monetary economics lowers inflation, not just because a monetary change induces a
fiscal tightening. So, I first continue with the definition that monetary policy means
changing interest rates and not changing or inducing changes in fiscal surpluses, and
thus (to a good approximation, outlined below) picking the = 0 equilibrium.

8 A fiscal-theoretic model with long term debt

Adding long-term debt produces a stable model in which a rise in interest rates can pro-
duce a temporary decline in inflation, with no change in fiscal surpluses. Sims (2011)
makes this point in the context of a detailed continuous time model. Cochrane (2016b)
shows how to solve Sims model and boils it down to this central point.

Start with a frictionless model and hence use a constant real interest rate r and
1/(1+r) to discount surpluses. In the presence of long-term debt, the government debt
valuation equation (11) becomes

P (j) (j)
j=0 Qt Bt1
X
= Et j st+j , (25)
Pt
j=0

(j)
where Bt1 is the amount of zero-coupon debt that matures at time t + j outstanding at
(j)
the end of time t 1 and thus at the beginning of time t, and Qt is the time t nominal
price of a j period discount bond.

When the Fed unexpectedly raises interest rates it , it lowers long-term bond prices
(j) (j)
Qt . Debt Bt1 is predetermined. By assumption, primary surpluses dont change.
Hence, the price level Pt must jump down by the same proportional amount as the de-
cline in the nominal market value of the debt. The sense that monetary policy works
by affecting long-term bond prices is correct here, though the mechanism is utterly
different from a Keynesian investment demand channel.
46 COCHRANE

By raising nominal interest rates, the Fed still raises expected inflation uniformly,
it rt + Et t+1 still applies. However, the price level on the date of the announcement
can jump downwards, as Figure 10 showed for innovations to fiscal surpluses coinci-
dent with the interest rate announcement, giving us the temporary decline in inflation
that were looking for.

This price level jump results in models such as this frictionless one, and the forward-
looking new-Keynesian Phillips curve, in which the price level is a jump variable. In a
model such as Sims (2011) with costs to swiftly changing prices, this jump can result in
a smeared out period of disinflation. The jump in these simple models is a guide to the
cumulative value of the disinflationary period.

The proportional change in market value of the debt, and hence the disinflation,
is larger for interest rate changes that are expected to last longer, and so have a greater
effect on long-term bond prices. The disinflation is also larger when there is more long-
term debt around. Therefore, this channel predicts an interesting relationship between
coefficients the persistence of shock and the size of its effect and an interesting
state-dependence monetary policy has larger effects when the maturity structure is
longer.

Just how large a disinflation does this mechanism produce? Is it quantitatively sig-
nificant, and hence a candidate to understand the apparent patterns in the data, or to
guide policy?

Building up to the usual stair-step impulse-response function, suppose the interest

(j)
rate i = it+j is expected to last forever. The bond price is then Qt = 1/(1 + i)j . Con-
(j)
sider a geometric maturity structure, Bt1 = j Bt1 , so = 1 is a perpetuity and = 0
is one-period debt, and a constant surplus st = s. Now, the government debt valuation
equation reads

X j Bt1 1 + i Bt1 s
= = (26)
(1 + i)j Pt 1 + i Pt 1
j=0

(j)
The continuous time analogue is prettier. With maturity Bt = ej
Z
Bt Bt s
eij ej dj = = . (27)
j=0 Pt i + Pt r
 INFLATION AND INTEREST RATES 47

Here = 0 is the perpetuity and = is instantaneous debt.

Now, suppose interest rates rise permanently and unexpectedly at time t. Denote
by i the post-shock interest rate, and Pt the post-shock price level. Then, dividing (26)
for the starred by the nonstarred case,

Pt 1 + i 1 + i
= .
Pt 1 + i 1 + i

The prettier continuous time analogue is

Pt i+
= . (28)
Pt i +

Now we can get a back of the envelope idea of the size of this effect and its crucial
determinants. The longer the maturity, the stronger the effect. In the most extreme
case, pairing this permanent interest rate rise with perpetual debt = 0 then the
continuous-time formula gives Pt /Pt = i/i . A jump in interest rates from 2% to 3%
causes the price level to drop to 2/3 of its previous value, a 33% decline!

However, the US doesnt issue that much long-term debt. Debt out to a 20 year
maturity follows a geometric pattern with 0.2. In this case, a one percentage point
interest rate rise implies Pt /Pt = (0.2/0.21) = 0.95, a 5% disinflation.

Shorter-lived interest rate rises, and announcements of future rate rises have less
effect still. In the Appendix, I show that an interest rate rise from i to i that only lasts
M years yields in place of (28),

Pt
 
  i+
1 1 eM 1 (29)
P i +

For example, an interest rate rise that lasts 2 years M = 2 has only 1 e2 = 1
e0.22 1/3 as large an effect, about 2% price level reduction.

An announcement of a future interest rate rise will lower long-term bond prices
today, and cause the disinflation to get going before the actual interest rate rise. How-
ever, it has smaller effects because it only affects bonds of maturity longer than the
announcement delay. An announcement that interest rates will rise in 3 years has no
effect on 3 year bonds, and no effect on the price level if the maturity structure is lim-
48 COCHRANE

ited to 3 year bonds. In this same simple geometric-maturity structure model, an an-
nouncement of an interest rate rise from i to i that starts in M years yields in place of
(28),
Pt
 
i+
1 eM 1 . (30)
Pt i +

Thus, an interest rate rise that is announced two years ahead of time has a e(0.22) or
about 2/3 as much effect, or about 3% rather than 5% disinflation.

To give a more serious quantitative evaluation of this disinflationary effect, I use

the 2014 zero-coupon U.S. Treasury debt outstanding provided by Hall and Sargent
(j)
(2015) for Bt1 , and the 2014 zero coupon yield curve {Y (j) } from Gurkaynak,
Sack,
(j) j
 
(j)
and Wright (2007) for bond prices Qt = 1/Yt . I calculate the nominal market
value of the debt as
X (j) (j)
X 1 (j)
Qt Bt1 = Bt1 .
(j) j
 
j=0 j=0 Yt

A permanent one percentage point increase in interest rates will increase yields at all
maturities, so yields after the change are Y (j) = Y (j) + 0.01 . The downward price

level jump is then given by

P 1 (j)
Pt j=0 Y (j) j Bt1
( )
=P (j)
.
Pt 1
j=0 Y (j) +0.01 j Bt1
( )

To calculate the effects of a rate rise that lasts only M years, I first find the zero
(j) (j) (j+1)
coupon forward curve before the change Ft = Qt /Qt . The effect of a M-year,
one percentage point rise in interest rates is then
  
P QM 1 1 Qj1 1 (j)
Bt1
Pt j=0 k=0 F (j) +0.01
t
k=M F (j)
t
=  
Pt P Qj1 1 (j)
j=0 k=0 F (j) Bt1
t

Similarly, the effect of a permanent one percentage point interest rate rise, announced
 INFLATION AND INTEREST RATES 49

M years in advance, is
  
P QM 1 1 Qj1 1 (j)
Bt1
Pt j=0 k=0 F (j)
t
k=M F (j) +0.01
t
=  
Pt P Qj1 1 (j)
j=0 k=0 F (j) Bt1
t

Figure 14 presents response functions, pairing these price level jumps with the fric-
tionless Fisherian model it = r + Et t+1 . In the top panel, a surprise permanent 1
percentage point increase in interest rates drives a 5% disinflation, consistent with the
back of the envelope calculation above. The long term debt response behaves exactly as
a fiscal tightening of Figure 10, i.e. an announcement of 5% higher primary surpluses
coincident with the monetary tightening. If the rate rise only lasts 3 years, in the the
long end of the ballpark found in most VAR studies, long term bonds are less affected,
so the disinflation is only about 2.5%.

The bottom panel of Figure 14 shows the response to an interest rate rise announced
three years ahead of time. Though VARs look only at interest rate surprises, the bulk of
actual interest rate changes are expected, so this response is most important to under-
stand history, episodes, and policy. Like the fiscal shock of Figure 10, this disinflation-
ary effect happens only on the announcement of the interest rate change, not on the
day of the actual change. This is a vital point to remember when evaluating the plausi-
bility of this mechanism relative to experience. The responses are smaller yet 3% for
the announced permanent rate increase, and 1.2% for the announced, three-year rate
increase.

8.1 Long term debt and sticky prices

The same analysis applies to the sticky-price model, except that real interest variation
changes the present value of surpluses. This consideration reduces further the size of
the disinflationary effect. Allowing real interest rate variation and long-term debt, the
valuation formula becomes

(j) j1
!
X (j) Bt1
X u0 (ct+j ) X Y 1
Qt = Et j 0 st+j = Et st+j . (31)
Pt u (ct ) 1 + rt+k
j=0 j=0 j=0 k=0
50 COCHRANE

interest rate i
1

inflation
0

-1
Percent response

-2
, long-term debt, 3 year i

-3

-4

, long-term debt
-5
-3 -2 -1 0 1 2 3
Time
interest rate i
1

inflation
0

-1
Percent response

, long-term debt, 3 year i

-2

, long-term debt
-3

-4

-5
-5 -4 -3 -2 -1 0 1 2 3
Time

Figure 14: Response of inflation to interest rate changes with long-term debt. Top: re-
sponse to interest rate change on the date of announcement. Bottom: response to a
preannounced interest rate change. The responses are calibrated to the term structure
of U.S. debt outstanding in 2014. Dashed lines give the response to a three-year rise in
interest rates.
 INFLATION AND INTEREST RATES 51

The first equality is the general formula; the second is an approximation reflecting the
linearized nature of the new-Keynesian model we are working with, in which risk pre-
miums do not vary over time.

We can always write nominal bond prices in terms of forward rates, and ignoring
risk premiums in the term structure also in terms of expected nominal interest rates

j1 j1
(j)
Y 1 Y 1
Qt = = Et .
1 + ft+k 1 + it+k
k=0 k=0

This expression helps us to compare left and right hand sides of the present value for-
mula (31). The Fed still controls nominal interest rates, and hence bond prices at all
maturities. Real rate variation and sticky prices make no difference to the left hand side
side. Typically, when prices are sticky, higher nominal rates translate more into higher
real rates and less into inflation. But such higher real rates reduce the real present value
of surpluses on the right hand side of (31), to some extent matching the reduction in
nominal present value of government debt on the left hand side, and thus requiring a
lesser decline in price level.

For example, suppose inflation is perfectly sticky. Then, the rise in real rates on the
right hand side of (31) exactly match the rise in nominal rates on the left, and there is
no deflationary pressure at all.

To calculate the response function merging the standard new-Keynesian sticky price
model with the fiscal theory and long-term debt, I again suppose interest rates start at
their 2014 values, and I compute the market value of the debt. I then suppose nomi-
nal interest rates all rise by the interest rate response function, and I calculate the new
nominal market value of the debt. I calculate the present value of an unchanged sur-
plus using the government debt valuation formula (31). That consideration chooses
a single value of Pt , equivalently of the multiple equilibria t , on the announcement
date. Equations are in the Appendix.

Figure 15 presents the response of inflation and output to an unexpected and per-
manent interest rate increase, shown as the top line. The solid line marked Inflation
and s = 0.00 is the inflation path. In order to require no change in future surpluses,
the devaluation of long-term debt now requires a 1.2% disinflation. Inflation is stable,
52 COCHRANE

Unexpected permanent rate rise

1

s=-4.47 Inflation
0
Percent response

-1
s=0.00
Output gap x
-2

-3

Inflation, no r effect

-4

-5
-3 -2 -1 0 1 2 3 4 5 6 7
Time
Expected permanent rate rise
1

Inflation
s=-2.68
0

Output gap x
Percent response

-1 s=0.00

-2

-3

-4

-5
-4 -2 0 2 4 6
Time

Figure 15: Response to permanent interest rate rises with long-term debt. I use the 2014
maturity structure of the debt to find the jump in price level that implies no change in
primary surpluses. Thin dashed lines present the case with no additional shock, = 0.
The line inflation, no r effect in the first panel ignores the effect of rising real rates
in devaluing future surpluses. s = gives the percent permanent change in primary
surpluses associated with each inflation path.
 INFLATION AND INTEREST RATES 53

Unexpected AR(1) rate rise

1 Interest rate

0.5

s=-1.71
Percent response

-0.5 s=0.00
Output gap x

-1

-1.5

-3 -2 -1 0 1 2 3 4 5 6 7
Time
Expected AR(1) rate rise

1 Interest rate

0.5
Percent response

s=-0.97
0

s=0.00
-0.5

Output gap x

-1

-1.5

-4 -2 0 2 4 6
Time

Figure 16: Response to transitory interest rate rises with long-term debt. I use the 2014
maturity structure of the debt to find the jump in price level that implies no change in
primary surpluses. Thin dashed lines present the case with no additional shock, = 0.
s = gives the percent permanent change in primary surpluses associated with each
inflation path. The interest rate reverts with a 0.7 AR(1) coefficient.
54 COCHRANE

so eventually rises to meet the long-term interest rate. This is the most hopeful graph
in this paper for an economically based model that gives the desired response function
for monetary policy changes.

The upper thin dashed line in Figure 15 presents the previously calculated inflation
path from Figure 11, with no extra shock 0 = 0. I show below that this is nearly the
same as the path with no change in primary surpluses in the presence of short-term
debt. That line is marked s = 4.47 in this case because, in the presence of long
term debt which is devauled by higher interest rates, surpluses would have to decline
by 4.47% permanently to boost inflation to this level.

The contrast between the dashed and solid inflation lines shows the effect of adding
long-term debt to the model and it is to give the long-sought period of disinflation.

The output gap lines in Figure 15 show that adding long-term debt, and thus pro-
ducing a downward jump in the price level, increases the output effects of the interest-
rate rise. With a forward-looking Phillips curve, the unexpected downward jump in the
price level has no output effect at all. However, the expectation of a strong increase in
inflation (from a lower level) drives output down.

The line marked Inflation, no r effect ignores the change in real interest rates on
the right hand side of (31), to compare the pure devaluation of debt from higher nomi-
nal rates with the devaluation of surpluses from higher real rates. The higher real rates
substantially lower the present value of surpluses, and make a big moderating differ-
ence to the initial disinflation. Models with this mechanism thus produce less disinfla-
tion from nominal interest rate rises when they have more sticky prices.

The bottom panel of Figure 15 shows the response when the interest rise is an-
nounced three years in advance. Again, these are models in which expected interest
rate rises have effects, and most experience in the data and most policy interventions
involve substantial expectations of future interest rate changes.

The higher interest rates now only affect bonds with three year or higher maturity.
Thus, the downward inflation rate jump is smaller, only 1%. Output suffers a much less
severe contraction, bottoming out at 2% not 4%.

As in the frictionless model, fiscal effects happen only on the day of announcement.
 INFLATION AND INTEREST RATES 55

This is an important consideration in evaluating this channel. It will not rescue the
old-Keynesian view that the interest rate rise itself sets off the disinflation. Nor does it
rescue old-Keynesian instability.

The effects get uniformly smaller as the interest rate rise is expected further and
further in the future. When the interest rate rise is expected after the maturity of the
longest bond, the disinflationary effect vanishes entirely. Thus, this fiscal channel sen-
sibly predicts smaller effects of expectations further in the future, and does not suffer
from the forward-guidance puzzle.

Figure 16 presents the response to an unexpected (top) and expected (bottom) AR(1)
rate rise. The unexpected transitory rate rise, though potentially unimportant for policy
and for most historical events, is the slice of variation that is potentially recovered by
VARs. (Though VARs dont attempt to orthogonalize monetary and fiscal policy shocks,
s = 0 as I do here.) When comparing to Figure 15, note that Figure 16 has a larger
vertical scale. The disinflation effect is now smaller still, less than 0.5% in both cases,
down from 2% - 4%. The AR(1) interest rate rise has less effect on longer term bonds
than a permanent rate rise. The expected AR(1) interest rate rise only really affects the
value of bonds in the middle maturity range.

The big picture is that long-lived interest rate rises can produce disinflations on the
same order of magnitude as the interest rate rises, and thus has the potential to explain
the perceived effects of monetary policy.

8.2 Critical assumptions and intuition

To think about this channel more deeply, return to the constant real rate frictionless
case,
P (j) (j)
j=0 Qt Bt1
X
= Et j st+j
Pt
j=0

(j)
Using the bond price, Qt = j Et Pt /Pt+j ,

 
X
j (j) 1 X
Bt1 Et = Et j st+j (32)
Pt+j
j=0 j=0
56 COCHRANE

When the Fed chooses higher nominal interest rates, and hence higher inflation and a
(j)
higher future price level, it thereby devalues the long-dated coupons Bt1 are divided
by larger Pt+j . This is great news for the Treasury it doesnt have to raise as many real
(j)
dollars st+j to pay off coupons. Rather than raise surpluses st+j = Bt1 /Pt , the Treasury
(j)
could cut surpluses to st+j = Bt1 /Pt and higher Pt > Pt means less surpluses.

But in this exercise, the Treasury stubbornly refuses the gift: The Fed says, you can
pay off the $1 coupons with (say) half as many real resources. But the Treasury says,
no, were going to insist on paying off the coupons with exactly the same real resources.
Thus, the Treasurys refusal of the Feds gift has the same effect as a fiscal contraction, a
promise to raise future surpluses. Such a fiscal contraction causes government debt to
be more valuable in real terms, and thus causes the price level to decline as people buy
less goods and services to hold more government debt.

Having stated that intuition, you see how very important the fiscal reaction to mon-
etary policy is in determining the outcome. Why does the Treasury stubbornly refuse
to reduce surpluses when the Fed wants to inflate away long-dated coupons? Why does
the Treasury not reduce future surpluses instead? If it is expected to do so, then we lose
the disinflationary effect.

A second intuition is also valuable. Equation (32) acts like a budget constraint
on the government. With no change in expected surpluses, any increase in expected
price level at one date must be matched by a decrease in expected price level at an-
other date. With long-term debt, expected surpluses control the moving average of
current and expected future prices on the left side of (32). Within that moving average,
the government can trade a higher price level at some days for a lower level of prices
at other days, by varying the path of future debt sales, or equivalently by controlling
nominal interest rates. Thus, when the government raises nominal interest rates, and
thereby raises the price level at future dates relative to the current price level and de-
values future coupons, it must also see a lower price level at date 0, and raise the value
of near-term coupons.

As in the one-period debt case, (13), the Fed adjusts price levels at one day versus
another by buying and selling debt with no changes in surpluses. To implement the
interest rate target, and thus to raise, say, Pt+j while lowering Pt+1 , the Fed will end
 INFLATION AND INTEREST RATES 57

up buying one-period debt and selling j-period debt. (Cochrane (2001) goes through
the algebra.) Conversely, the Fed could lower say, Pt+j while raising Pt+1 , by buying
j-period debt and selling one-period debt, all with no change in surpluses.

Therefore, in this operation, monetary policy and quantitative easing (QE) opera-
tions operate in much the same way. Integrating QE and interest-rate policy in a single
framework, requiring no market segmentation or other frictions, is attractive. It also
sheds light on the puzzle why QE has had so little, or actually negative, effects on long-
term inflation expectations.

This model, like any specific model, ties its explanation of a temporary decline in in-
flation to other predictions. Whether this is in fact the model of temporary disinflation
that we seek depends on those other predictions as well. In this model, monetary pol-
icy only reduces inflation if it changes long term bond prices and the nominal market
value of the debt.

This model does not easily produce a traditional stabilization. The decline in short-
term inflation is inextricably linked to a rise in long-term inflation. If the Fed were
to raise short-term rates, and people expected inflation to decline permanently, with
the Fed eventually lowering short-term rates in the future the traditional pattern of
a monetary stabilization then the value of nominal debt would rise, and this model
would not produce the initial decline of inflation.

Sims (2011), elaborating on this models mechanism, called it stepping on a rake.

He views the model as a description of the failed monetary stabilizations of the 1970s,
in which interest rate increases produced temporary reductions of inflation that only
came back more strongly later. To produce a successful inflation stabilization, a model
of the 1980s, one needs something else. A natural candidate is to view the 1980s as a
joint monetary-fiscal stabilization. The interest rate increases of the early 1980s were
paired with fiscal reforms such as the 1986 tax act, and subsequent strong economic
growth. Surpluses surged, which permanently reduce inflation. The natural policy for
a permanent inflation reduction in this model remains to lower interest rates, along
with some measure that really ensures fiscal expectations are unchanged; then wait
out the temporary increase in inflation that results.

Furthermore, as I emphasized in the figures, this mechanism gives a disinflation

58 COCHRANE

when the interest rate rise is expected, not when it actually happens.

Finally, this mechanism for a short-run negative inflation effect does not immedi-
ately rationalize traditional policy. It is not necessarily wise or possible for the Fed to
try to control inflation by exploiting this short-run negative sign. The negative sign only
appears for unexpected policy changes, at the time of the news. Systematic policy, such
as the t in it = t , will not have any negative inflation effect. And getting the timing
and dynamics just right are likely to be a challenge. Since the long-run effect is positive,
there is a good case that to control inflation, the Fed should steady interest rates based
on its long-run inflation goal and real-rate assessment, and not try to micromanage the
path of inflation. with activist policy exploiting the transitory negative sign.

Most deeply, this model does not revive the instability of the old-Keynesian model
behind both traditional activist policy advice and stabilizations. A transitory negative
sign for unexpected interest rate rises, with a long-run positive, stable relationship be-
tween interest rates and inflation, is a very different animal from a transitory negative
sign for all interest rate rises, with a long-run negative, unstable relationship. The latter
does demand a Fed that actively exploits the negative sign. The former does not.

If we accept this direction, two important next steps follow. First, changes in in-
terest rates with fixed surpluses are a useful textbook, problem-set sort of assumption.
They are worth working out to understand mechanisms. But fixed or exogenous sur-
pluses are not necessary for the theory. And fixed or exogenous surpluses are a terri-
ble assumption for policy, econometric or historical analysis. Just how will the Treasury
respond to inflation? Surely not zero, as I specified here. Yet it is the crucial assumption
to understand how interest rates affect inflation. More deeply, both fiscal (surpluses)
and monetary (interest rates) policy react to the same underlying sets of events. Any
historical episode represents a set of monetary and fiscal responses to other events. A
movement in interest rates with no change in expected surpluses just doesnt happen.
So the first step is to think much harder about the path of surpluses, how surpluses re-
act to economic events, and how fiscal policy reacts to the same underlying economic
shocks that motivate the change in interest rates.

Second, of course, one must move beyond the extremely simple model presented
here to more detailed models capable of matching dynamics. Sims (2011) is a good
 INFLATION AND INTEREST RATES 59

example, adding a preference for smooth consumption, a monetary policy rule with
output and price reactions and inertia, and a fiscal policy rule that raises surpluses
in good times. He produces a hump-shaped inflation response curve in place of my
downward jump followed by rise.

In sum, this mechanism produces a model in which inflation is stable, but has the
desired short-run negative inflation response. The response is quantitatively impor-
tant and could potentially describe empirical work. It describes interest rate policy and
quantitative easing in one breath. However, comes with some unusual additional pre-
dictions it does not give the conventional description of a long-run stabilization by
monetary policy alone, does not describe any negative effect of expected rate changes,
and does not offer an easy justification that the Fed should exploit the negative sign in
regular policy making. Most of all, it relies on a fundamentally unconventional mecha-
nism. Though flight from or to government debt feels like aggregate demand to those
living in such an economy, the mechanism is entirely fiscal-theoretic it acts exactly
as a change in expected surpluses with one-period debt acts. Completely absent from
the story are conventional mechanisms such as IS curves, the effect of interest rates on
investment demand, monetary distortions and even pricing frictions.

Now that we see in detail the car we are driving, before taking this fork in the road,
let us make sure that standard monetary economics cannot be saved.

9 Money

Perhaps monetary distortions, in addition to pricing distortions, will give us the tradi-
tional result. Perhaps when interest rate increases were accomplished by reducing the
supply of non-interest-bearing reserves, that reduction in money produced a tempo-
rary decline in inflation that simply raising the interest rate on excess reserves will not
produce. Such a finding would, however, suggest that raising interest rates by simply
raising the rate paid on abundant excess reserves will not have the same temporary
disinflationary effect as past history suggests.

I introduce money in the utility function, nonseparable from consumption, so that

changes in money, induced by interest rate changes, affect the marginal utility of con-
60 COCHRANE

sumption, and thus the intertemporal-substitution equation.

Woodford (2003) (p. 111) begins an analysis of this specification. But Woodford
quickly abandons money to produce a theory that is independent of monetary fric-
tions, and does not work out the effects of monetary policy with money. If theory
following that choice now does not produce the desired outcome, perhaps we should
revisit the decision to drop money from the analysis.

The detailed presentation is in the Appendix. The bottom line is a generalization of

the intertemporal-substitution condition (18), to:

m
it+1 im m


xt = Et xt+1 + ( ) Et t+1 (it it ) (it Et t+1 ) . (33)
c

The presence of money in the utility function has no effect on firm pricing decisions
and hence on the Phillips curve (19). Here, is the interest-elasticity of money de-
mand. Evidence such as Figure 8 and literature surveyed in the Appendix suggests
d log(m)/d log(i) = 0.1. The value m/c is the steady state ratio of real money hold-
ings to consumption. The larger this value, the more important monetary distortions.
The quantity im
t is the interest rate paid on money.

Equation (33) differs from its standard counterpart (18) by the middle, change-in-
interest rate term. Equation (33) reverts to (18) if utility is separable between money
and consumption ( ) = 0, if m/c goes to zero, or if money pays the same interest
rate as bonds i = im .

The expression m/c (it im

t ) represents the proportional interest costs of holding

money. The middle term following ( ) represents the expected change in those
proportional interest costs. An expected increase in interest costs of holding money,
induces the consumer to shift consumption from the future, when holding the money
needed to purchase consumption goods will be relatively expensive, towards the present.
It acts just like a lower real interest rate to induce an intertemporal reallocation of con-
sumption.

The presence of expected changes in interest rates brings to the model a mechanism
that one can detect in verbal commentary: the sense that changes in interest rates affect
the economy as well as the level of interest rates.
 INFLATION AND INTEREST RATES 61

However, monetary distortions only matter in this model if there is an expected

change in future interest rate differentials. Expected, change, and future are all cru-
cial modifiers. A higher or lower steady state level of the interest cost of holding money
does not raise or depress todays consumption relative to future consumption. An un-
expected change in interest costs has no monetary effect at all, since Et (it+1 it ) = 0
throughout.

The model solution is essentially unchanged. The extra term in the intertemporal
substitution equation (33) amounts to a slightly more complex forcing process involv-
ing expected changes in interest rates as well as the level of interest rates. One simply
replaces it in (20)-(21) with zt defined by
 
m 
Et it+1 im m


zt it t+1 (it it ) .
c

The slight subtlety is that this forcing process is the change in expected interest differen-
tials. Lag operators must apply to the Et as well as whats inside. Inflation depends on
past expectations of interest rate changes, not to past interest rate changes themselves.

9.1 Impulse-response functions

I start with the traditional specification that the interest on money im

t = 0, so that in-

creases in the nominal interest rate are synonymous with monetary distortions. Figure
17 plots the response function to our expected and unexpected interest rate step with
money distortions m/c = 0, 2, 4.

For the unexpected interest rate rise, shown in dashed lines, the presence of money
makes no difference at all. The dashed lines are the same for all values of m/c, and
all the same as previously, and the model remains stubbornly Fisherian. This is an
important negative result. Money can only affect the response to expected interest rate
changes.

The response to an expected interest rate rise, shown in solid lines, is affected by the
monetary distortion. As we increase the size of the monetary distortion m/c, inflation
is lower in the short run. For m/c = 4, we get the classic shape of the impulse response
function. The announced interest rate rise produces a temporary decline in inflation,
62 COCHRANE

Inflation

i
1 Un, expected
, expected

0.8

0.6
Percent response

0.4
m/c=0

0.2

0
m/c=2

-0.2

-0.4 m/c=4
-4 -2 0 2 4 6
Time
Output
1
i
x, unexpected
x, expected

0.5
m/c=4

0 m/c=2
Percent response

m/c=0
m/c=0
-0.5

-1

-1.5
m/c=4

-2

-4 -2 0 2 4 6
Time

Figure 17: Response of inflation and output to an interest rate rise; model with money.
The three cases are m/c = 0, 2, 4. Solid lines are an expected interest rate rise, dashed
lines are an unexpected rise.
 INFLATION AND INTEREST RATES 63

and then eventually the Fisher effect takes over and inflation increases.

The only time-difference in interest costs comes at time 0. Larger m/c induces the
consumer to shift consumption to times before 0, to consume when the interest costs
of holding the necessary money are lower. Output is high when inflation is decreasing,
and vice versa, so this pattern of output corresponds to lower inflation before time 0
and higher inflation afterward.

The m/c = 4 curve seems like a great success, until one ponders the size of the
monetary distortion non-interest bearing money holdings equal, on average, to four
years of output. This model is not carefully calibrated, but m/c = 4 is still an order of
magnitude or more too large.

Equation (33) suggests that raising , which multiplies m/c, may substitute for a
large m/c, by magnifying the effect on consumption of a given monetary distortion.
Now, higher also magnifies the last term, which induces Fisherian dynamics. But in
our response functions, the middle term multiplies a one-time shock, where the last
term multiplies the entire higher step. Thus, raising can raise the relative importance
of the one-time shock in the dynamics of inflation.

Figure 18 investigates the effect of changing the intertemporal substitution elastic-

ity . Since an unexpected interest rate rise again has no monetary effect, I present only
the case of an expected interest rate rise.

The left two panels, labeled mc = 0 in Figure 18, show the effect of varying in the
model without money. In this model, and enter symmetrically in the determina-
tion of inflation. Raising increases the speed of the dynamics, pulling the S shaped
response closer to the step that holds in a frictionless model. Raising the speed of the
dynamics lowers inflation in the early period, a step in the direction of the conventional
belief. But raising without money can never produce a negative effect on inflation.

The right two panels of Figure 18 with m/c > 0 show how increasing can work
together with a monetary friction. At m/c = 1, increasing from = 1 to = 3
produces a slight decline in inflation before the inevitable rise. The subsequent rise
is quicker; the main effect here has been to borrow inflation from the future. To get
a substantial negative effect, one must increase either or m/c even more. The line
64 COCHRANE

Inflation, no money Inflation, with money

1 1
Percent response

Percent response
0.8
0.5
0.6 =1, mc=1

0.4
=3, mc=0 0 =3, mc=1
=1, mc=0
0.2 i i

0 -0.5 =4, mc=2

-4 -2 0 2 4 6 -4 -2 0 2 4 6
Time Time

Output, no money Output, with money

1
1

0.5
Percent response

Percent response
0

0 -1
=1, mc=1
-0.5 -2
=3, mc=1
-1 =1, mc=0 -3
i i
x x
-1.5 -4
=3, mc=0 =4, mc=2
-4 -2 0 2 4 6 -4 -2 0 2 4 6
Time Time

Figure 18: Response of output and inflation to an expected interest rate step; model
with money and varying intertemporal substitution elasticity .

= 4, m/c = 2 produces about the same inflation decline as = 1, m/c = 4 produced

in Figure 17.

So, higher can help to produce a temporary dip in inflation, largely by speeding up
dynamics. Alas, = 1 was already above most estimates and calibrations. A coefficient
= 3 implies that a one percentage point increase in the real interest rate induces
a three percentage point increase in consumption growth, which is well beyond most
estimates. And m/c = 1 is already at least twice as big as one can reasonably defend.

In sum, these calculations show what it takes to produce the standard view: For an
anticipated interest rate rise only, money in the model can induce lower inflation than
a model without monetary frictions produces. If we either have very large money hold-
 INFLATION AND INTEREST RATES 65

ings subject to the distortion, or a very large intertemporal substitution elasticity, the
effect can be large enough to produce a short-run decline in inflation. Adding money
to the model in this way has absolutely no effect on responses to an unexpected per-
manent interest rate rise.

9.2 Money and transitory rate shocks

Since expected changes in interest rates are the crucial mechanism in this model, per-
haps putting in more reasonable interest rate dynamics can revive the desired inflation
dynamics.

Figure 19 shows the response function to an AR(1) interest rates shock. Money does
affect the response functions. And, that effect is uniformly to raise inflation. The ex-
pected decline in interest costs posed by the AR(1) reversion after the shock shifts con-
sumption from the present to the future, and inflation rises when output is low.

With expected interest rate dynamics, the unanticipated rate rise now has monetary
effects, shown in the dashed lines. These too are uniformly in the direction of higher
inflation as monetary frictions increase.

9.3 Interest spread policy

The Federal Reserve is contemplating varying the interest it pays on reserves as sepa-
rate policy tool. By changing the interest on reserves, the Fed can affect money demand
without changing the nominal rate. Thus, it can focus on the monetary effects on de-
mand without the direct intertemporal substitution effects.

Figure 20 presents a calculation. Here, the Fed raises the interest on reserves im by
one percentage point, with no change in the nominal interest rate i.

One can debate whether we should call such a rise in interest on reserves expan-
sionary or contractionary policy a priori. Raising interest on reserves is often con-
sidered contractionary, as it encourages banks to sit on reserves rather than to pursue
lending. On the other hand, raising interest on reserves lowers the spread between re-
serves and other instruments, and so encourages the accumulation of money, which
66 COCHRANE

Inflation

i
1 , unexpected
, expected

m/c=4
0.8

m/c=2
0.6
Percent response

0.4 m/c=0

0.2

m/c=0

0 m/c=2

m/c=4

-0.2

-4 -2 0 2 4 6
Time
Output
1
i
x, unexpected
x, expected

0.5 m/c=4

0 m/c=2

m/c=0
Percent response

-0.5

-1

-1.5

-2

-2.5

-4 -2 0 2 4 6
Time

Figure 19: Response of inflation and output to a temporary rate rise, model with money.
Dashed lines are the response to an unexpected rise, solid lines are the response to an
expected rise.
 INFLATION AND INTEREST RATES 67

Inflation

im

1

0.8
Percent response

0.6

m/c=4

0.4

m/c=2
0.2

0 m/c=0
-4 -2 0 2 4 6
Time
Output

im
x

1
m/c=4

0.5 m/c=2
Percent response

0 m/c=0

-0.5

-1

-4 -2 0 2 4 6
Time

Figure 20: Response to a permanent rise in the rate paid on reserves, holding the nom-
inal interest rate constant.
68 COCHRANE

Inflation
m
i
1 expected
unexpected

0.8

0.6
m/c=4
Percent response

0.4

m/c=2
0.2

0 m/c=0 m/c=0

-0.2 m/c=2

-0.4
m/c=4
-4 -2 0 2 4 6
Time
Output

2 im
expected
unexpected

1.5

1
Percent response

0.5

0 m/c=0 m/c=0

m/c=2

m/c=4
m/c=2
-0.5

m/c=4
-4 -2 0 2 4 6
Time

Figure 21: Response to a transitory rise in the rate paid on reserves, holding the nominal
interest rate constant. Solid lines are the response to an expected change; dashed lines
are the response to an unexpected change.
 INFLATION AND INTEREST RATES 69

one might consider to be expansionary.

Again, the response to an unexpected rise in the interest on reserves is exactly zero.
The intertemporal substitution mechanism only operates when the expected future is
different from the present.

An expected rise in interest on reserves raises inflation throughout. Output declines

ahead of the change, and rises after the change. Money is cheaper to hold after the rise,
encouraging consumers to postpone consumption. With a forward looking Phillips
curve, lower output corresponds to rising inflation, and vice versa.

Figure 21 graphs the effects of a transitory increase in the interest on reserves, which
is expected to die out with an AR(1) pattern, again holding the nominal rate unchanged.

In this case, the unexpected change (dashed lines) lowers inflation. The shock it-
self has no effect, because it was unexpected. However once the shock as passed,
consumers expect interest on reserves to decline, and expect interest costs of holding
money to rise. This change induces them to bring consumption forward to the periods
just after t = 0, as shown in the bottom panel. Higher output means declining inflation
to the forward-looking Phillips curve as shown in the upper panel.

One might celebrate the first unambiguous negative inflationary effect of a tighten-
ing, but the mechanism is far from traditional. The rise in interest on reserves, being
unexpected, has no effect at all. The rise allows for an expected decline in interest on
reserves, and it is this expected decline which does all the work. An expected decline
with no initial rise would have the same effect. And the actual interest rate is constant
throughout.

The expected case of Figure 21 adds the effects of the anticipated rise in interest
on reserves to this story. The expected rise induces consumers to postpone consump-
tion to the period just after the rise, doubly increasing output then, but also driving up
inflation.
70 COCHRANE

9.4 Money summary

The basic mechanism for demand in this, as in all new-Keynesian models, is in-
tertemporal substitution, changes in the margin of current versus future consumption.
Adding money to the model distorts the tradeoff of current vs. future consumption. In
making that tradeoff, the consumer thinks about the actual real interest rate, but also
the relative costs of holding money necessary to make purchases today vs. that cost in
the future. A higher interest cost of holding money in the future is, like a higher real
interest rate, an inducement to consume now.

Alas, this mechanism is quantitatively small. Relative to actual changes in real in-
terest rates, the distortions to intertemporal incentives from greater or lesser costs of
holding money are second-order. We just dont hold that much non-interest-bearing
money.

Also, this mechanism does not give rise to classic intuition. Interest costs of money
holdings only affect demand if people expect higher or lower interest costs in the fu-
ture than they experience today. The level of interest costs has no effect.

10 The Phillips curve

Empirically, lags seem important in Phillips curves. The forward-looking Phillips curve
(19) specifies that output is higher when inflation is high relative to future inflation,
i.e. when inflation is declining. Though all Phillips curves fit the data poorly, especially
recently, output is better related to high inflation relative to past inflation, i.e. when
inflation is rising (Mankiw and Reis (2002)).

Theoretically, the pure forward looking Phillips curve is not central. Though it does
some violence to the economic criterion for the simple baseline theory that we are
searching for, we should check if the short or long-run neo-Fisherian conclusions can
be escaped by adding past inflation to the Phillips curve.

The simplest approach is to consider a static Phillips curve. This specification is

the 0 limit of the three equation model (18)-(19). Kocherlakota (2016) provides
detailed micro-foundations for a static Phillips curve.
 INFLATION AND INTEREST RATES 71

So consider

xt = Et xt+1 (it Et t+1 ) (34)

t = xt . (35)

The equilibrium is simply

1
Et t+1 = t + it (36)
1 + 1 +

and hence
X 1 X 1
t = i tj + tj . (37)
(1 + )j (1 + )j
j=1 j=0

The dynamics are stable, and inflation responds positively to interest rates through-
out. In fact, they are exactly the same dynamics as we found in section 3 for a static IS
curve and fully forward-looking Phillips curve. Repeating the equations for clarity,

xt = (it Et t+1 ) (38)

t = Et t+1 + xt (39)

leads exactly to the same conditions (36)-(37).

Thus, Figure 6 already plots the response function for the static Phillips curve case
(34)-(35) - and inflation rises smoothly throughout.

Even eliminating both intertemporal terms,

xt = (it Et t+1 )

t = xt

produces
1
Et t+1 = t + it .

The model is stable or unstable depending on , but the sign on interest rates remains
positive.

A backwards-looking accelerationist Phillips curve is more realistic. This model

72 COCHRANE

also adds some reassurance that the Fisherian results do not result from a permanent
inflation-output tradeoff, present both in the static Philips curve and the new-Keynesian
model with < 1. Consider

e
t = t1 + xt

X
te = (1 ) j tj .
j=0

Substituting the output gap from the usual intertemporal IS curve (34) ,

e
(t t1 ) = (Et t+1 te ) (it Et t+1 )

(1 + )Et t+1 = t + te t1
e
+ it

X
(1 + )Et t+1 = t + (1 ) j tj + it
j=0

Figure 22 plots the impulse-response function for this model. It is Fisherian through-
out. Now the tightening increases the output gap. In some sense by giving the right
(old-Keynesian) sign of the relationship between output and inflation, since a rise in
interest rates gives the wrong (positive) effect on inflation, it also gives the wrong
(positive) effect on output.

This response function is the same for expected as for unexpected monetary policy,
which at least should give some comfort to the traditional view that the effects of policy
have to await some action by the Federal Reserve.

Figure 22 plots the = 0 equilibrium. One can add a negative jump with fiscal
tightening consequence to deliver a negative inflation response on the announcement
day, but that is a different mechanism which I study in detail elsewhere.

The Appendix considers a model with both forward and lagged terms in the Phillips
curve.

X
X
t = xt + Et j xt+j + j xtj .
j=1 j=1
 INFLATION AND INTEREST RATES 73

Figure 22: Impluse-response with P backward-looking Phillips curve. Model xt =

Et xt+1 (it Et t+1 ), t = (1 ) j
j=0 tj + xt . = = 1, = 0.9.

This model also does not produce the desired temporary inflation decline.

The lesson from these exercises seems to be, that to generate a negative sign, substi-
tution of adaptive te = t1 rather than rational te = Et t+1 throughout is the crucial
ingredient. Neither forward-looking IS nor forward-looking Phillips curves per se are
essential. However, when we do that, we return to the unstable old-Keynesian model
of the first section, which does not deliver long-run stability or the long-run Fisher re-
lationship, and is therefore inconsistent with stability at the zero bound. Nothing so far
delivers a temporary negative sign and a long-run positive sign, other than the long-
term-debt fiscal-theoretic model.
74 COCHRANE

11 Other equilibria

There are multiple equilibria, indexed by the expectational shock {t }. As first displayed
in Figure 10, one might recover a short-run negative inflation response by pairing the
announcement of a rate increase with a negative multiple-equilibrium shock .

11.1 Multiple equilibria in the impulse-response function

Returning to the standard new-Keynesian model of equations (18)-(19) and response

function displayed for 0 = 0 in Figure 11, Figure 23 plots a range of multiple equilib-
rium responses to the unanticipated step function in interest rates. Each equilibrium
is generated by a different choice of the expectational shock 0 that coincides with the
monetary policy shock at date zero.

(For an impulse-response function, equilibrium choice affects only the possibility

of a single shock t on the date t of the announcement. An impulse-response function
studies the path of Et t+j , j = 0, 1, 2, 3... to an announcement made at time t. There-
fore, values of tj , j > 0 in the solution (20) do not matter. They are the same pre- and
post-announcement. Values of t+j , j > 0 for an impulse-response function are zero
after the announcement, Et t+j = 0, j > 0. )

Equilibrium A has a positive additional inflation shock, 0 = 1%. Equilibrium B

chooses 0 to produce 1% inflation at time 0, 0 = 1%. Equilibrium C chooses 0 to
have no fiscal consequences, explained below. Between C and D lies the original fun-
damental equilibrium, with 0 = 0. Equilibrium D chooses 0 to produce no inflation
at time 0, 0 = 0. Equilibrium E chooses 0 = 1%.

The figure shows graphically that the model may have too many equilibria, but all of
them are stable, and all of them are Fisherian in the long run, with inflation converging
to the higher nominal interest rate.

Equilibrium E verifies that the model can produce a temporary decline in inflation
in response to the interest rate rise. Equilibrium E achieves that result by pairing a neg-
ative expectational or sunspot shock with the positive interest rate or expected inflation
shock. We will pay a lot of attention to the plausibility of equilibrium E.
 INFLATION AND INTEREST RATES 75

Inflation
1.2 A,s=-1.53

1 B,s=-1.00

0.8

0.6 C,s=-0.00
Percent response

0.4

=0,s=0.67
0.2

0 D,s=1.66

-0.2

-0.4

E,s=2.99
-4 -2 0 2 4 6
Time
Output
1

0.5 A

B
0
Percent response

-0.5
C

-1
=0

-1.5

D
-2

-2.5
E
-4 -2 0 2 4 6
Time

Figure 23: Multiple equilibrium responses to an unexpected interest rate rise. The solid
green line gives the interest rate path. Letters identify different equilibria for discussion.
The original case is 0 = 0.
76 COCHRANE

The other possibilities are informative as well. In equilibrium B inflation jumps in-
stantly to the full increase in nominal interest rates, and stays there throughout. Output
also jumps immediately to the steady-state value. Thus, despite price stickiness, the
model can produce a super-neutral or super-Fisherian response, in which an interest
rate rise instantly implies inflation with no output change!

Equilibrium A shows that even more inflation is possible. With a sufficiently large
expectational shock, inflation can actually increase by more than the interest rate change,
and then settle down, and output can increase as well.

Equilibrium D adds a small negative expectational shock 0 , so that the initial infla-
tion response is precisely zero. One may be troubled by inflation jumps, since inflation
seems to have inertia in the data. It can be inertial in the model as well. (In continuous
time, the no-jump equilibrium D is the same as the = 0 equilibrium.)

11.2 Choosing equilibria

Is there is a convincing argument to prefer equilibria such as E, and to view this result as
an embodiment of the conventional belief that raising interest rates temporarily lowers
inflation?

The issue is not what shock t we will see on a particular date. The question is what
shock t we will expect to see on average in response to announcements at date t of an
interest rate rise.

One could make an empirical argument. But the point of this paper is to find eco-
nomics for an inflation decline, not to fit the most central prediction of monetary eco-
nomics through a free parameter, the correlation of expected and unexpected inflation
shocks.

The rest of this section examines arguments for particular equilibrium choices.

11.3 Anticipated movements and backwards stability

The behavior of the different equilibrium choices with regard to anticipated move-
ments is an important consideration in this choice. Most monetary policy tightenings
 INFLATION AND INTEREST RATES 77

are expected, and this kind of model describes responses to expected monetary policy
changes.

Figure 10 showed in the frictionless model that mulitiple equilibria show up in the
impulse-response functions of the frictionless model as a jump in inflation on the date
of the announcement only. Figure 24 likewise shows multiple equilibrium responses to
monetary policy change announced at t = 3 in the model with pricing frictions. All
responses except equilibrium C are the same as in Figure 23 for t 0. Equilibrium C is
recalculated to give zero fiscal effect at time t = 3 rather than t = 0.

Figure 24 cautions us on the apparent success of equilibrium E to capture a tempo-

rary inflation decline when interest rates rise. That success relies crucially on matching
the announcement of the interest rate change with the actual change. Figure 24 tells
us that if a tightening is expected, as tightenings usually are, then inflation and output
should both drop the most on the announcement of a tightening, not later when rates
actually rise. (A well specified VAR should throw out such events, but analysis of policy
and episodes must include them.) Classic intuition might allow small announcement
effects, but the bulk of inflation and output reactions should at least coincide with if
not follow actual interest rate rises.

Equilibrium D makes this announcement behavior more apparent: Inflation jumps

down on the announcement but (by construction) is zero on the day of the actual rate
rise, and positive thereafter.

This is an instance of a more general problematic behavior of many equilibrium

choices. Most of the alternative equilibrium choices explode backwards; they imply
bigger inflation shocks on the day of the announcement than at t = 0 when inter-
est rates change. Equivalently, they have the property that news about events further
in the future has larger effects today. The original t = 0 equilibrium choice has the
backwards-stable property that it does not explode backwards, and that news about
events further in the future has less and less effect today.
78 COCHRANE

Inflation
A, s=-7.91
3

B, s=-5.96
2

1
Percent response

=0,s=0.14
C, s=0.00
0

-1
D, s=3.77

-2

-3
E, s=8.63
-6 -4 -2 0 2 4 6 8 10
Time
Output
A

4 B

2
Percent response

=0
0
C

-2

-4

-6
E
-6 -4 -2 0 2 4 6 8 10
Time

Figure 24: Multiple equilibrium responses to an anticipated interest rate change. The
numbers s = give the percent change in steady state surpluses required to achieve
each equilibrium.
 INFLATION AND INTEREST RATES 79

11.4 Fiscal index

Each equilibrium choice has a fiscal policy consequence. Unexpected inflation deval-
ues outstanding nominal debt, and thus lowers the long-run financing costs of the debt.
Higher real interest rates raise financing costs. For each equilibrium choice, then, I cal-
culate the percentage amount by which long-run real primary surpluses must rise or
fall for that equilibrium to emerge. That number is presented alongside the initial in-
flation value of each equilibrium in Figure 23 and Figure 24.

Making this calculation requires no assumption whether fiscal policy is active or

passive. Even if the equilibrium choice is made by monetary policy or other off-
equilibrium threats, and fiscal policy is completely passive, we can still calculate the
surpluses that the passive treasury must collect. For this reason, I call the calculation
a fiscal index it indexes equilibria even if it does not select equilibria.

To make this calculation, I start with the valuation equation for government debt,

0
Bt1 X u (Ct+j )
= Et j 0 st+j , (40)
Pt u (Ct )
j=0

where Bt1 denotes the face value of debt outstanding at the end of period t 1 and
beginning of period t, Pt is the price level and st is the real net primary surplus.

Starting from a steady state with constant surplus s, I calculate the fractional per-
manent change in surplus s, i. e. st = S s , that is required of the right hand side of
expression (40) for each response function. Linearizing, I obtain in the appendix

1 X j
s Et (t ) + Et (xt+j xt ) (41)

j=0

where Et Et Et1 and t is the date of the announcement of a new policy. (The
computations are nearly identical with a linearized or fully nonlinear valuation equa-
tion.)

The first term of (41) captures the fact that unexpected inflation devalues outstand-
ing government debt. In the second term, (xt+j xt )/ is the real interest rate between
time t and time t+j. So this term captures the fact that if real rates rise, the government
80 COCHRANE

must pay more interest on the debt.

This calculation is simplified in many ways. I specify one-period nominal debt.

Here, the objective is to focus on changes in surpluses, not the long-term debt effect
studied above. A more realistic calculation adds both effects, and gives similar devia-
tions from the responses with long-term debt plotted above. Second, in reality output
changes affect primary surpluses, as taxes rise more than spending in booms and fall
more than spending in recessions. Third, inflation also raises revenue due to a poorly
indexed tax code. But some of these effects may represent a change in timing of sur-
pluses borrowing during recessions that is repaid later during booms rather than
permanent changes that affect the real value of government debt. A serious calculation
of the fiscal impacts of monetary policy requires considerable detail on these lines. The
point here is not quantitative realism, but to capture some of the important effects and
to show how one can use fiscal considerations to evaluate different equilibrium possi-
bilities.

The super-neutral equilibrium B in which inflation rises instantly by 1%, also marked
s = 1.00 in Figure 23, corresponds to a 1% decline in long-run surpluses. The 1%
jump in inflation devalues outstanding nominal debt by 1%, and since output is con-
stant after the shock there is no real interest rate change. Equilibrium A, with a larger
inflation shock, corresponds to a larger than 1% decline in long-run surpluses.

Further down in Figure 23, equilibrium D has no change in inflation at time 0, and
so there is no devaluation of outstanding nominal debt. However, the rise in real inter-
est rates means that the government incurs greater financing costs. These costs require
a small permanent rise in surpluses.

In between, at equilibrium C, I find the shock 0 that requires no change in fiscal

policy at all, so s = 0 by construction. Here, the devaluation effect of an inflation
shock just matches the higher financing costs imposed by higher real interest rates.

However, at least in this back of the envelope calibration, the difference is not large.
Ignoring real interest rate effects, and discounting surpluses at a constant rate does not
make a first-order difference. One does quite well grafting the simple constant-interest-
rate FTPL formulas on to the new-Keynesian model.
 INFLATION AND INTEREST RATES 81

The original equilibrium with no expectational shock, 0 = 0, implies a small but

nonzero change in surpluses, to offset the real interest rate effect.

Equilibrium E, in which inflation temporarily declines half a percentage point af-

ter the interest rate shock, requires an 1.54% rise in permanent fiscal net-of-interest
surpluses. Disinflation raises the value of nominal debt, which must be paid.

Turning to the anticipated shocks of Figure 24, the larger inflation shocks at time
t = 3, and the longer periods of high real interest rates, mean that the fiscal changes
required to support most of the equilibria increase as we move the announcement back
in time. For example, the originally super-neutral equilibrium which required a 1% de-
cline in surpluses in Figure 23 now requires a 4.11% surplus decline, because of the
larger inflation shock. And equilibrium E, selected to generate a 1% decline in infla-
tion when interest rates rise 1%, now requires a 5.6% permanent rise in fiscal surpluses
rather than 1.54%.

The exceptions to this rule are the original equilibrium choice = 0, the equilibrium
choice C or s = 0 with no fiscal impact, and an equilibrium (not shown) that always
chooses no inflation on the announcement date. All of these equilibria have smaller
fiscal impacts as interest rates are announced earlier in time, they all converge to the
same point and they are all stable backward.

11.5 Using fiscal policy to choose equilibria

To produce the standard view that raising interest rates lowers inflation, we must ac-
company the rate rise with a fiscal tightening as in equilibrium E. Disinflation implies
an unexpected present to holders of nominal government debt. Higher real interest
rates also imply higher debt service payments. Fiscal authorities must be expected to
raise taxes or to cut spending to make those payments.

An event such as equilibrium E is therefore a joint fiscal-monetary tightening. It

provides useful guidance about what would happen if fiscal and monetary policy act
together. Historic successful disinflation programs have typically combined monetary
tightening with fiscal reform, which produces a rise in future (though often not cur-
rent) surpluses. And unexpected monetary policy changes are often responses to un-
82 COCHRANE

expected economic or political conditions, which trigger fiscal policy reactions as well.

So equilibrium E of Figure 23 or its counterpart in the announced tightening of

Figure 24, where inflation drops on the mere announcement of the policy change is
compelling when one wishes to match data, understand an episode, or analyze a policy
that potentially combines a monetary and a fiscal policy shock.

But our question is to evaluate the hypothetical effects of monetary policy alone.
For that question, and given the possibility of any of these equilibria, it is not com-
pelling that we should pair the monetary policy shock (rise in interest rates) with a sub-
stantial fiscal policy shock.

Monetary policy has fiscal implications, and can thereby affect inflation. Higher
real rates are an inflationary force. The differences between the s = 0 equilibrium C,
the = 0, and in Figure 23 equilibrium D with no change in inflation capture that fact.
Monetary policy raises real interest rates which impacts the budget. Equivalently, the
same surpluses are discounted at a higher rate at time 0. Therefore, If fiscal surpluses
st do not change as in C, the present value of surpluses declines, so a small inflation
surprise slightly devalues outstanding debt. Equivalently, in D, for there to be no deval-
uation of debt, surpluses must rise a bit to pay the larger real interest costs.

This interest-expense channel is a possible fiscal-theoretic channel for the impact

of monetary policy, stressed most recently by Sims (2016). However, the differences
between the the s = 0 equilibrium C, the = 0 equilibrium, and the no-inflation-
jump equilibrium D are small here, and get smaller for anticipated policy. In addition,
the interest-expense channel C raises inflation, so does not even directionally help in
the quest for a temporary negative response.

The fiscal calculation can serve as an equilibrium-selection mechanism. The fiscal

theory of the price level and the definition that a monetary policy shock means one
with no fiscal policy response s = 0, selects equilibrium C.

However, one can also keep a passive-fiscal view of these calculations. In that case,
the fiscal index merely reveals how much fiscal policy must passively adjust to whichever
equilibrium is selected by some other means. One can then decide if the required fiscal
adjustment is reasonable or not. If criterion x selects a path that requires the passive
 INFLATION AND INTEREST RATES 83

fiscal authority to raise 200% of GDP in taxes, its not going to happen.

In this vein, one might argue for equilibria that have limited or small fiscal require-
ments, rather than equilibria which require large changes in surpluses to be generated
by the passive fiscal authorities, or insist on equilibria with exactly zero fiscal impli-
cations. That argument, which we might call fiscal theory lite, puts us in a range around
the s = 0 equilibrium, and still limits our ability to produce disinflation.

Pushing the announcement date back as in Figure 24 enlarges these fiscal consid-
erations. The equilibria that are not backward-stable all have larger and larger fiscal
policy consequences as the announcement is pushed back. Conversely, = 0 equi-
librium, choice, the no fiscal impact s = 0 choice and the no inflation jump choice
converge as the announcement is pushed back. By an announcement t = 3 shown
in Figure 24, the = 0 equilibrium and the s = 0 equilibrium are already visually
indistinguishable.

Equilibria with no jump in inflation are also attractive. Equilibrium D in Figure 23

has this property, and one can construct an equilibrium with no change in inflation
upon announcement for the t = 3 shock of Figure 24. We do not see inflation jumps
in the data, and new-Keynesian models are often specified so that inflation must be
set one or more periods in advance. This choice also is stable as the announcement
horizon moves backward.

In sum, the principles of small fiscal requirements, sensible behavior as announce-

ments come earlier than actual rate changes, or limited jumps in inflation all push one
to the view that equilibria near the original = 0 equilibrium are sensible, and the
others less so. One does not need to be a fiscal theory of the price level zealot for this
choice.

The absence of the affirmative is more important here than the negative. I have
not found a strong economic reason that we should pair large negative expectational,
equilibrium-changing, or fiscal-policy-induced shocks 0 with announcements of in-
terest rate rises. I find no economic mechanism for producing a large unexpected in-
flation shock, except fiscal policy, which suggests no such shock when thinking about
monetary policy interventions. The essential definition of a central bank is that it can
rearrange government debt but cannot take fiscal action such as a helicopter drop. So,
84 COCHRANE

this discussion leads me to look away from paths such as equilibrium E as the device
to generate a temporary decline in inflation when interest rates rise, and to look else-
where.

12 Taylor rules

Taylor rules with active responses to inflation are usually invoked to prune equilibria,
and to deliver a short-run negative inflation response. Can writing policy in terms of a
Taylor rule help us to choose among the equilibria displayed in Figures 23 and 24?

12.1 Constructing Taylor rules

As previewed above, the solution method using equations (20)-(21) does not assume
a peg. We can construct a Taylor rule that supports any of the equilibria displayed in
Figures 23 and 24, as follows. Assume interest rate policy is

it = it + (t t ) (42)

where it is the step-function or other desired equilibrium interest rate path, t is the
equilibrium path of inflation, i.e. one of the choices displayed in Figure 23 or Figure 24,
and is arbitrary. If > 1, then the desired path {it , t , xt } is the unique locally-
bounded (nonexplosive) equilibrium. (Here, of course, I consider the case away from
the zero bound. At the zero bound, one must construct more complex promises of
future Taylor rules as discussed briefly in section 4.3.)

Traditionally, one solves this kind of model by adding to the IS and Phillips curves
(18)-(19) a monetary policy rule, say

it = it + t (43)

and then solving for equilibrium {it , t , xt } given shocks including it . To produce an
impulse-response function, as I have, one must find a monetary policy disturbance
sequence {it } that produces the desired response of equilibrium interest rates {it }. In
 INFLATION AND INTEREST RATES 85

general, since > 0, the disturbance sequence {it } is different from the interest-rate
response.

However, equations (42) and (43) are the same for

it = it . (44)

In this context, then, my procedure solving for output and inflation given the desired
equilibrium interest rate path, and then constructing monetary policy that supports
the desired equilibrium by (43), or by (44) amounts simply to a way to avoid the un-
pleasant search for the monetary policy shock disturbance {it } that produces the de-
sired equilibrium interest rate path. This clever approach and interpretation is due to
Werning (2012).

Expressing the Taylor rule as in (42) emphasizes that the active Taylor rule includes
two policy settings. The rule consists of an interest rate target, {it }, and an equilibrium-
selection rule, the choice of and {t } from the set of equilibrium {t } consistent with
the interest rate target. The interest rate target determines the path of equilibrium in-
terest rates. The selection rule specifies a set of off-equilibrium threats or beliefs, that
rules out all but the desired equilibrium path of inflation. (Many other equilibrium se-
lection schemes achieve the same purpose, for example see Atkeson, Chari, and Kehoe
(2010) and the discussion in the online appendix to Cochrane (2011).)

This construction (42) and its equivalence with (43) addresses the first question:
Does the assumption of a Taylor rule solve the equilibrium selection problem? No. Via
(42), all of the equilibria, any choice of 0 such as graphed in Figure 23 and Figure 24,
are consistent with an active Taylor rule, and equation (42) shows how to construct the
Taylor rule assumption, down to the specific shock sequence it that generates any
desired equilibrium. The fact of adding a Taylor rule, by itself, doesnt help us at all to
choose among equilibria.

12.2 Open-mouth policy

Furthermore, if the Fed can induce jumps t by equilibrium-selection policy imple-

mented by an active Taylor rule > 1, there is no reason for the Fed to bother with
86 COCHRANE

interest-rate policy. Why pair the equilibrium-selection policy that shifts unexpected
inflation downwards with a rise in interest rates that shifts expected inflation upwards?
If the Fed wants to induce temporarily lower inflation, all it need to is to announce a
new inflation target.

To be specific, suppose the Fed follows a Taylor rule

it = it + (t t ). (45)

Suppose the Fed, starting at it = 0, t = 0 for t < 0, leaves it alone, but shocks mone-
tary policy for t 0 to
t = 0 t
1 . (46)

Here, 0 is a constant indexing how large the monetary policy shock will be. This is a
pure, temporary, change in the Feds inflation target.

Equivalently, suppose the Fed follows a Taylor rule

it = it + t . (47)

Suppose that the Fed, starting at it = 0 for t < 0, shocks monetary policy for t 0 to

it = 0 t . (48)
1

This is a pure, temporary, monetary policy disturbance.

Figure 25 plots the responses of inflation and output to these monetary policy dis-
turbances. Inflation and output move, but interest rates are constant throughout the
episode.

Intuitively, in response to a shock i0 , and its expected subsequent values {it }, infla-
tion jumps down just enough so that the systematic component of policy in (47) exactly
offsets the shock, and the actual interest rate it does not change at all. In response to
the shock 0 , and to the expected subsequent values {t }, inflation jumps to 0 = 0 .
Via the Taylor rule (45), this change in inflation is is just enough so that actual interest
rates do not change.
 INFLATION AND INTEREST RATES 87

Figure 25: Response of inflation and output to a shift in inflation target with no shift in
interest rate target.
88 COCHRANE

Once we see how pure open-mouth / equilibrium-selection operations can induce

the shocks 0 , pairing shocks to inflation as in Figure 25 with shocks to interest rates is
arbitrary.

Now, perhaps this is our world. Monetary policy at the zero bound has seemed to
evolve into central banker statements accompanied by no actual changes in interest
rates or asset purchases. Reserve Bank of New Zealand Governor Donald Brash (Brash
(2002)) coined the term open-mouth operations observing that he seemed to be able
to move interest rates by simply talking, without conducting open market operations.
The open mouth operation described by Figure 25 is doubly removed from action, since
the central bank can apparently move inflation without even moving interest rates.

Perhaps inflation really has little to do with economics; supply and demand, in-
tertemporal substitution, money, and so forth. Perhaps inflation really is predomi-
nantly a multiple-equilibrium question. Perhaps monetary policy affects inflation
entirely by government officials making statements, with implicit never-observed off-
equilibrium threats, that cause jumps from one equilibrium to another, validated by
passive fiscal policy. Perhaps the analysis of monetary policy should go back where
it left off in the 1950s and 1960s, in which inflation was largely thought to comprise
wage-price spirals, and inflation policy centered on talk not action. Perhaps changes
to actual interest rates, though economically irrelevant and even counterproductive in
the long run, evolved as some sort of communication and signaling equilibrium to in-
dicate a policy shock.

If so, again, sufficient becomes necessary. The quest of this paper a simple, trans-
parent, baseline economic model of the effect of interest rates on inflation is over,
with a negative result and a disquieting implication for the status of monetary policy in
the arsenal of robust and well-understood phenomena.

Figure 25 includes the change in long-run surpluses needed to validate each equi-
librium. I include this number as a reminder that it is there. If one takes the fiscal-
passive view, these are the resources that the passive Treasury will need to come up
with to validate the Feds active equilibrium-selection policy.

If one takes the fiscal theory of the price level view, these calculations have a much
simpler interpretation. In this case, the indicated change in expected future surpluses
 INFLATION AND INTEREST RATES 89

results in one of these equilibria as the unique equilibrium. These are movements in in-
flation achievable by a pure change in fiscal policy, when monetary policy leaves nom-
inal interest rates unchanged. In the fiscal theory of the price level, the Fed still sets
expected inflation freely by setting nominal interest rates, while fiscal policy uniquely
chooses unexpected inflation, and hence 0 .

If one takes a passive-fiscal view, nonetheless the fiscal authority must passively
come up with these surpluses to validate monetary policy. The Fed announces a pol-
icy shock. Actual interest rates dont move. Inflation jumps. People expect the fis-
cal authority to loosen in response to the announced monetary policy shock. To an
economist who cannot see the shock announcement, the whole affair looks exactly like
a sunspot. If this story challenges believability, that is the point.

12.3 Reasonable disturbances

Perhaps, however, expressing policy in terms of a traditional single shock (43) it = it +

t , rather than in terms of a separate interest rate target it , inflation target t and an
equilibrium selection policy in (42) it = it + (t t ), via (44) it = it , will
indicate that one or another equilibrium results from a more sensible monetary policy
disturbance.

Figure 26 gives the monetary policy disturbance it in a Taylor rule it = it + 1.5t

needed to produce the step function rise in equilibrium interest rates and each of the
possible inflation outcomes (equivalently, equilibrium choices ) from Figures 23 (an
unexpected interest rate rise, solid) and 24 (an expected interest rate rise, dashed).

The equilibria with large positive inflation shocks such as A result from negative
monetary policy disturbances, and vice versa. When i and move together, > 1 in
it = it + t , means it must go the opposite way. Again we see the interesting prop-
erty of the standard explosive Taylor rule that interest rate rises correspond to negative
policy shocks.

All the disturbances end up at it = 0.5, since they all end up with it = 1 and t = 1,
and 1.0 = 0.5 + 1.5 1.0.

To produce the baseline 0 = 0 inflation pattern in the unanticipated (solid) case,

90 COCHRANE

E
4

2 D

1
Percent

0
C
=0
-1

-2

-3

B
-4

A
-6 -4 -2 0 2 4 6 8 10
Time

Figure 26: Monetary policy disturbance it = it t that produces each equilibrium

with Taylor parameter = 1.5. Dashed lines give values for the announcement at time
t = 3. Solid lines give values for the announcement at the same time as the rate rise
at time t = 0. Letters and = 0 correspond to the equilibria shown in previous figures.

the disturbance {t } follows a pattern with geometric decay. This pattern mirrors the
geometric rise of inflation, relative to the step function rise in observed interest rates.

To produce a transitory decline in inflation, one needs the larger disturbance of

equilibrium E. E is a strong positive disturbance that slowly melts away to negative in a
geometric pattern. Comparing the two {t } its hard to say one is a lot more reasonable
than the other. Both disturbances are first positive and then negative.

The dashed lines, showing the monetary policy disturbances necessary to produce
the responses to an anticipated rise in equilibrium interest rates are wilder. Viewed
through the lens of a Taylor rule, the Fed does not simply announce that rates will rise
in the future. All of these equilibria feature inflation that moves between the announce-
ment and the actual interest rate rise. Therefore, with it = it + t , if t moves and it
 INFLATION AND INTEREST RATES 91

does not move, the Fed must have a disturbance it that offsets inflation t . As in the
open-mouth operations, the Fed announces a monetary policy shock {it }. Inflation
moves so much, however, that the systematic component of monetary policy t ex-
actly offsets the monetary policy shock {it }, producing a change in inflation with no
change at all in the actual interest rate.

For the anticipated rate rise (dash), there is no equilibrium in which it does not
move ahead of the actual interest rate rise, as there is no equilibrium in which inflation
does not move ahead of an anticipated interest rate rise. But the baseline equilibrium
= 0 and the equilibrium C with no fiscal consequence s = 0 at least have dis-
turbances it that are small and that decline as the policy announcement moves back
in time. By contrast, the disturbance E is large and grows as the announcement time
moves back.

In sum, if one pursues reasonable specifications for the monetary policy distur-
bance, in the context of a Taylor rule of the form it = it + t , as an equilibrium se-
lection device, that path does not strongly suggest equilibria such as D and E in which
inflation declines temporarily. In fact, the view that the Fed makes big monetary policy
shocks that induce big changes in inflation, which through the systematic component
of policy t then just offset the monetary policy shock and produce no change in
interest rate, may seem the more far-fetched assumption.

12.4 The standard three-equation model

The open-mouth calculation of the last section is a special case of the standard analysis
of a monetary policy shock in the three-equation new-Keynesian model. Didnt that
model produce the desired sign, one might ask?

I repeat here the standard equations,

xt = Et xt+1 (it Et t+1 ) (49)

t = Et t+1 + xt (50)

it = t + vti (51)
i
vt+1 = vti + t+1 (52)
92 COCHRANE

Figure 27 plots the response of inflation and interest rates to an unexpected monetary
policy shock vti for this model.

=1 = 0.9
1 vi 1 vi
0.5 0.5
Percent (%)

Percent (%)
0
0
-0.5
-0.5
-1
-1
-1.5
-1.5 i
-2 i
-2.5 -2
0 2 4 6 8 10 0 2 4 6 8 10
Time Time
i= 0.50838 i = 0.3
1 v 1 v
0.8
0.6
Percent (%)

Percent (%)

0.4 0.5
0.2 i
0 i
-0.2 0
-0.4
-0.6
-0.8 -0.5
0 2 4 6 8 10 0 2 4 6 8 10
Time Time

Figure 27: Response of inflation and interest rates to an AR(1) monetary policy shock
v i with persistence in the standard three-equation new-Keynesian model. = 0.95,
= 1/2, = 1, = 1.5.

The top left panel plots the response to a permanent shock, as in the other figures.
The standard three equation model is perfectly Fisherian in this case! The positive
monetary policy shock v i produces a negative response of actual interest rates and in-
flation i and . The rule it = t + vti becomes 2 = 1.5(2) + 1. This example em-
phasizes the important difference between monetary policy shocks and equilibrium
interest rates. The shock is positive, but interest rates fall. One might opine that it is a
strange tightening by which actual interest rates fall one for one with inflation.

The top right panel plots the response to a very persistent = 0.9 shock. Here the
inflation and interest rate responses are easier to distinguish. Still, interest rates fall,
 INFLATION AND INTEREST RATES 93

and one sees a largely Fisherian result. The endogenous negative part of interest
rates still overwhelms the positive monetary policy shock.

The bottom left panel plots the response in the knife-edge case = 1/1 that the
monetary policy shock becomes an open-mouth operation. Here the endogenous
effect t just offsets the shock vti so inflation moves with no change at all in interest
rates.

Finally, the bottom right panel shows that for a sufficiently short-lived shock, =
0.3 < 1/1 , the shock v i exceeds the endogenous response , so a positive monetary
policy shock increases interest rates i and decreases inflation .

Is this, at last, the answer we are looking for? It is not a fundamentally different
mechanism than the cases analyzed so far. It combines a swiftly mean-reverting pro-
cess for the interest rate, as graphed in Figure 12, with a strong contemporaneous fiscal
contraction like case E of Figure 23. Repeating for convenience the formula (20) for the
general inflation solution,



j j2 Et+1 it+j + j
X X X
t+1 = it + 1 itj + 1 t+1j (53)
1 2
j=1 j=1 j=0

we can see the case. If interest rates i mean-revert quickly enough, the central terms will
be small. Then, if we add a large enough shock at time zero, we produce a negative
inflation response.

So the standard three equation model, with sufficiently transitory monetary policy
shock to produce the right sign, is really just a case of a large multiple-equilibrium
shock, with a strong passive fiscal tightening.

13 Other modeling directions; necessary vs. sufficient

assumptions

We have searched for a simple modern, rational economic model, consistent with the
observed stability and quiet behavior of inflation at the zero bound, that restores the
traditional view in which a rise in interest rates produces at least a temporary decline
94 COCHRANE

in inflation. The result is, so far, largely negative. Price stickiness, money, backward
looking Phillips curves, alternative equilibrium choices and active Taylor rules do not
provide a convincing basis to overturn the short-run Fisherian predictions of the fric-
tionless model. They do not begin to overturn its long-run Fisherian prediction.

The one model that works is the fiscal theory with long-term debt, in which an inter-
est rate rise amounts to a swap of current for future inflation. However, the temporary
decline in inflation in this model is a deeply fiscal-theoretic phenomenon, is a sensitive
result of particular expected fiscal policies, only applies to unexpected rate rises, and
does not rescue traditional policy advice, either on how to engineer a disinflation or
that the Fed can and should systematically raise interest rates in response to inflation.

The next directions one might go to reestablish the conventional view involve aban-
doning one of the qualifiers simple, modern, or economic.

In order to produce the standard signs, one might add ingredients to micro-found
the ad-hoc old-Keyensian models basic structure, to try to restore the modern and
economic adjectives, while also somehow repairing the models prediction that the
zero bound is unstable. One might add extensive borrowing or collateral constraints,
hand-to-mouth consumers, irrational expectations or other irrational behavior, a lend-
ing channel, or other frictions, continuing the 60-year old quest to undermine the per-
manent income hypothesis.

Gabaix (2016) is a concrete behavioral example: By severely downweighting Et xt+1

in the IS curve, (18), Gabaix produces traditional explosive dynamics and therefore a
negative sign. His model still needs patching to match the long run and the zero bound
episode the key observation that rules out explosive dynamics but it represents a
state-of-the-art example of the possibility. (See Cochrane (2016a).)

The models in this paper are also quite simple by the standards of calibrated or esti-
mated new-Keynesian models, such as Smets and Wouters (2003), Christiano, Eichen-
baum, and Evans (2005) and their descendants. Garca-Schmidt and Woodford (2015)
and Angeletos and Lian (2016) (see also the extensive literature review in the latter)
fundamentally change the nature of equilibrium and expectations. Perhaps adding the
range of complexity in such models habits, labor/leisure, production, capital, vari-
able capital utilization, adjustment costs, alternative models of price stickiness (despite
 INFLATION AND INTEREST RATES 95

my unsuccessful foray into ad-hoc Phillips curve lags) or informational, market, pay-
ments, monetary, and other frictions, or fundamentally different views of expectations
and equilibrium will perturb dynamics in the right way.

But following these paths abandons the qualifiers simple or economic. They
mean that more complex or non-economic ingredients are necessary as well as suffi-
cient to deliver the central result. Doing so admits that there is no simple, rational
economic model one can put on a blackboard, teach to undergraduates, summarize in
a few paragraphs, or refer to in policy discussions to explain at least the signs and rough
outlines of the operation of monetary policynothing like, say, the stirring and simple
description of monetary policy in Friedman (1968).

Such an intellectual outcome would be unusual in macroeconomics. The standard

new Keynesian approach views the complex models or even behavioral modfications as
refinements, building on (18)-(19). The refinements help to match the details of model
dynamics with those observed in the data, but the simple model captures capturing the
basic message, signs, stability and intuition. The standard real business cycle approach
views complex models as refinements, building on the stochastic growth model, but
that simple model can still capture the basic story. The large multi-equation Keynesian
models developed in the 1970s built on simple ISLM models to better match details of
the data, but modelers felt that the simple ISLM model captured the basic signs and
mechanisms.

This was all healthy. Economic models are quantitative parables, and one rightly
distrusts predictions that crucially rely on the specific form of frictions, especially fric-
tions that have little microeconomic validation.

So, if we go down the route, that much greater complexity or abandonment of sim-
ple economics, rational behavior and expectations, are necessary as well as sufficient
to generate the basic sign and stability properties of monetary policy, we have already
admitted an important defeat. The scientific basis on which we analyze policy, and its
advise its use to public officials and public at large, is clearly much more tenuous.
96 COCHRANE

14 VARs

If theory and experience point to a positive reaction of inflation to interest rates, per-
haps we should revisit the empirical evidence behind the standard view to the contrary.

The main formal evidence we have for the effects of monetary policy comes from
vector autoregressions (VARs). There are several problems with this evidence.

First, the VAR literature almost always pairs the announcement of a new policy with
the change in the policy instrument, i.e. an unexpected shock to interest rates. That
habit makes most sense in the context of models following Lucas (1972) in which only
unanticipated monetary policy has real effects, and in the context of regressions of out-
put on money, rather than interest rates, in which VARs developed (Sims (1980)).

But in the models presented here, anticipated monetary policy has strong effects. In
the world, most monetary policy changes are anticipated. VARs may still want to find
rare unexpected rate movements, as part of an identification strategy to find changes
in policy that are not driven by changes of the Feds expectations of future output and
inflation. But there is no reason, either theoretical, empirical, historical, or for policy-
relevant analysis, to focus all model analysis on a surprise exogenous rate change. Its
like evaluating a Ferrari by driving it around the parking lot. OK, it should drive cor-
rectly around the parking lot, but it should do a lot else as well. That mismatch drives
the less formal style of analysis in this paper matching model predictions with broad
facts about the data.

Furthermore, the models with money presented here, as in Figure 17, only had a
chance of delivering the standard inflation decline if the interest rate rise was antici-
pated. An empirical technique that isolates unexpected interest rate rises cannot find
or verify that theoretical prediction.

Second, the analysis of multiple equilibria in Figure 23 and Figure 24 found that in-
flation declines occur when an interest rate rise is paired with a fiscal policy tightening.
Now, all interest rate changes are responses to something. So it is plausible that what-
ever motivates the Fed to raise or lower interest rates also motivates fiscal authorities
to change course, even if those events are truly uncorrelated with expected output or
inflation and thus candidate VAR shocks. It is plausible that rate shocks in our data set
 INFLATION AND INTEREST RATES 97

are like equilibrium E of Figure 23. VARs have to date made no attempt to orthogo-
nalize monetary policy shocks with respect to fiscal policy, especially expected future
fiscal policy which is what matters here. But such a VAR estimate does not measure the
effects of a pure monetary policy shock, which is our question.

Third, VARs typically find that the interest rate responses to an interest rate shock
are transitory, as are those of Figure 12. As a result, they provide no evidence on the
long-run response of inflation to permanent interest rate increases.

Fourth, and most of all, the evidence for a negative sign is not strong, and one
can read much of the evidence as supporting a positive sign. From the beginning,
VARs have produced increases in inflation following increases in interest rates, a phe-
nomenon dubbed the price puzzle by Eichenbaum (1992). A great deal of effort has
been devoted to modifying the specification of VARs so that they can produce the de-
sired result, that a rise in interest rates lowers inflation.

Sims (1992), studying VARs in five countries notes that the responses of prices to in-
terest rate shocks show some consistency - they are all initially positive. He also spec-
ulates that the central banks may have information about future inflation, so the re-
sponse represents in fact reverse causality. Christiano, Eichenbaum, and Evans (1999)
took that suggestion. They put shocks in the order output (Y), GDP deflator (P), com-
modity prices (PCOM), federal funds rate (FF), total reserves (TR), nonborrowed re-
serves (NBR), and M1 (M) (p. 83). Their idea is that commodity prices capture infor-
mation about future inflation that the Fed may be reacting to, so including commodity
prices first isolates policy shocks that are not reactions to expected inflation. This or-
dering also assumes that policy cannot affect output, inflation, or commodity prices for
a quarter. With this specification (their Figure 2, top left), positive interest rate shocks
reduce output. But even with the carefully chosen ordering, interest rate increases have
no effect on inflation for a year and a half. The price level then gently declines, but re-
mains within the confidence interval of zero throughout. Their Figure 5, p.100, shows
nicely how sensitive even this much evidence is to the shock identification assump-
tions. If the monetary policy shock is ordered first, prices go up uniformly. The inflation
response in Christiano, Eichenbaum, and Evans (2005) also displays a short run price
puzzle, and is never more than two standard errors from zero.
98 COCHRANE

Even this much success remains controversial. Hanson (2004) points out that com-
modity prices which solve the price puzzle dont forecast inflation and vice versa. He
also finds that the ability of commodity prices to solve the price puzzle does not work
after 1979. Sims (1992) was already troubled that commodities are usually globally
traded, so while interest rate increases seem to lower commodity prices, its hard to
see how that could be the effect of monetary policy.

Ramey (2015) surveys and reproduces much of the exhaustive modern literature.
She finds that The pesky price puzzle keeps popping up. Of nine different identifica-
tion methods, only two present a statistically significant decline in inflation, and those
only after four or more years of no effect have passed. Four methods have essentially
no effect on inflation at all, and two show strong, statistically significant positive ef-
fects, which start without delay. Strong or reliable empirical evidence for a short-term
(within 4 years) negative inflation effect is absent in her survey.

The Christiano, Eichenbaum, and Evans (1999) procedure may seem fishy already,
in that so much of the identification choice was clearly made in order to produce the
desired answer, that higher interest rates lead to lower inflation. Nobody spent the
same amount of effort seeing if the output decline represented Fed reaction to news,
because the output decline fit priors so well. As Uhlig (2006) points out, however, that
procedure is defensible. If one has strong theoretical priors that positive interest rate
shocks cause inflation to decline, then it makes sense to impose that view as part of
shock identification, in order to better measure that and other responses. (Uhligs elo-
quent introduction is worth reading and contains an extensive literature review.)

But imposing the sign only makes sense when one has that strong theoretical prior;
when, as when these papers were written, existing theory uniformly specified a nega-
tive inflation response and nobody was even considering the opposite. In the context
of this paper, when theory specifies a positive response, when only quite novel theo-
ries produce the negative sign, and we are looking for empirical evidence on the sign,
following identification procedures that implicitly or explicitly throw out positive signs
does not make sense. And even imposing the sign prior, Uhlig like many others finds
that The GDP price deflator falls only slowly following a contractionary monetary pol-
icy shock.
 INFLATION AND INTEREST RATES 99

With less strong priors, positive signs are starting to show up. Belaygorod and Dueker
(2009) and Castelnuovo and Surico (2010) find that VAR estimates produce positive in-
flation responses in the periods of estimated indeterminacy. Belaygorod and Dueker
(2009) connect estimates to the robust facts one can see in simple plots: Through the
1970s and early 1980s, federal funds rates clearly lead inflation movements. Dueker
(2006) summarizes.

And of course all of this evidence comes from the period in which the Fed kept a very
small amount of non-interest-bearing reserves, the money multiplier plausibly bound,
and the Fed implemented interest-rate changes by changing the small quantity of re-
serves or promising to do so. To state that interest rates in this regime have the same
effect on inflation as they do in the current regime, in which the Fed will change interest
rates by changing the interest it pays on super-abundant reserves, with no open market
operations at all, requires theory.

15 Fiscal policy anchoring and warning

In several places I have advocated fiscal theory of the price level arguments. I collect
here quick answers to some of the most common objections to these uses of fiscal ar-
guments. Cochrane (2005) contains many more.

15.1 Exogenous surpluses?

The fiscal theory of the price level does not require that surpluses are exogenous, or
set independently of monetary policy or economic conditions. The government debt
valuation equation

X u0 (ct+j )
Bt1
= Et j 0 st+j (54)
Pt u (ct )
j=0

is an equilibrium condition; it works in exactly the same way as the standard asset val-
uation formula in which the price per share is the present value of dividends d,

X u0 (ct+j )
Valuet
pt = = j 0 dt+j ,
Sharest u (ct )
j=0
100 COCHRANE

or the permanent income formula in which consumption is the present value of future
income
X
ct = rkt + rEt j yt+j
j=0

In all cases, though it is often useful to think through the logic of models with a story
that price level, stock price, or consumption, are determined given the quantities
on the right-hand side, in fact all quantities are economically endogenous and co-
determined. Furthermore, this analogy should but to rest any concern that the first
equation is a budget constraint or that it involves scary off-equilibrium threats, any
more than do the latter two.

The government debt valuation equation holds in equilibrium in all regimes ac-
tive or passive. Therefore active and passive regimes are observationally equivalent.
There is no test for Granger causality, no regression of surpluses on debt or interest
rates on inflation, no relationship among time series drawn from the equilibrium of
an economy, that can distinguish fiscally active and passive, or monetary active and
passive (using the terminology in Leeper (1991)) regimes. That fact drives the style of
analysis in this paper only by looking across regimes, or from experiments such as the
zero bound episode, and asking for Occams razor simplicity can we tell theories apart.

15.2 On and off equilibria

If a government pays back its debts, then the regime must be passive and the fiscal
theory cannot operate, no? No.

The key to a passive regime is that the government raises surpluses to pay back
debts arising from any cause. For example, if a bubble or sunspot causes the price level
to decline 50%, the government will double taxes to repay that windfall to bondholders,
just as it would to repay debt incurred to finance a war.

Furthermore, the distinction between on equilibrium and off equilibrium responses

common in the new-Keynesian Taylor-principle view applies to the fiscal theory as well.
The government can commit to repaying any on-equilibrium debt, but defaulting or
inflating away off-equilibrium debt. Following King (2000), I wrote in (42) the Taylor
 INFLATION AND INTEREST RATES 101

rule in a new-Keynesian model as

it = it + (t t ) (55)

This observationally equivalent form allows us to separate how a central banks ob-
served interest rate it responds to observed equilibrium inflation t from how the the
central bank threatens to modify interest rates off-equilibrium (it it ) from off-equilibrium
inflation (t t ). This writing of the Taylor rule makes it clear that no data from an
equilibrium can enlighten us about the crucial off-equilibrium threat, how the central
bank would respond to a sunspot inflation to rule it out.

Precisely the same argument applies to the fiscal theory. If surpluses respond to
debt levels,
st = st + (Bt1 /Pt ) (56)

then, yes, fiscal policy is passive, surpluses adjust on the right hand side of (54) for
any price level, and the equation drops out of equilibrium determination. But if we
write the fiscal rule in form analogous to (55)

st = st + (Bt1 /Pt ) (57)

where denotes the governments desired equilibrium, then we see exactly the same
response in equilibrium the government pays off its debts but no response to off-
equilibrium price level bubbles or sunspots. As Taylor-rule off-equilibrium interest
rate rises select a unique equilibrium, so fiscal-rule off-equilibrium inaction selects a
unique equilibrium. And there is no way to tell apart (56) from (57) using data from a
given equilibrium.

Specification (57) is not unreasonable. A government following (57) pays off debts
it incurs, for example, to finance wars or fight recessions. If the government wants to
borrow real resources and keep a stable price level P , then such commitment is vital.
If the government wants inflation, it needs to commit not to pay off debts due to a
rising price level. But a government could and arguably should refuse to accommodate
bubbles and sunspots in the price level.
102 COCHRANE

Empirically, governments do not always rally surpluses to unexpected inflations

and deflations! For example, the US in the 1930s went off the gold standard and de-
faulted on the gold clause in its bonds after a sharp deflation. Governments can well
refuse to validate price levels they dont like, whether on or off equilibria.

15.3 What about Japan? And Europe? And the US?

Japans debt approaches 200% of GDP, and 20 years of deficit spending have not pro-
duced desired inflation. Doesnt the fiscal theory predict hyperinflation for Japan? Eu-
rope is north of 100% debt to GDP ratio, and the US fast approaching. Present value
of surpluses? What surpluses? The CBOs deficit forecasts have even primary deficits
exploding. How can one begin to write about fiscal policy anchoring inflation expecta-
tions?

If only the fiscal theory were as simple as deficits and debt equals inflation! It is not.

Despite awful projections, it is not unreasonable for bond markets to believe for
now that the western worlds debt problems will be solved successfully. The CBOs
deficit forecasts are if something doesnt change forecasts, and include straightfor-
ward policies that can turn surpluses around mild pro-growth economic policies,
mild entitlement reforms. The same angst over debt-fueled inflation emerged in the
late 1970s and early 1980s Sargent and Wallaces unpleasant monetarist arithmetic
being only the most famous example. (We remember the pathbreaking analysis of
fiscal-monetary interactions. We politely forget the forecast that the US would soon re-
turn to inflation in the 1980s.) A few regulatory and tax reforms later, we were worrying
in the late 1990s about the disappearance of government debt. The US largely repaid
greater debts after WWII, and the UK grew out of much larger debt in the 1800s. A debt
crisis leading to inflation will be a self-inflicted wound, not an economic necessity.

Much more importantly, in my view, the fiscal theory needs to digest the main les-
son of asset pricing discount rates vary a lot, and are vitally important to understand
valuations.

Real interest rates are zero or negative throughout the western world, even at very
long horizons. From this perspective, the fiscal theory puzzle is not the lack of inflation.
 INFLATION AND INTEREST RATES 103

The fiscal theory puzzle is the lack of deflation.

For a simplified approach to this observation, consider a constant discount rate r

and an economy growing at rate g and hence primary surpluses st = yt growing at g.
Z Z
Bt st
= Et erj st+j dj = Et e(gr)j djst = (58)
Pt j=0 j=0 rg

or, the ratio of surplus to real value of the debt is r g. The discount rate r matters as
much as g. If r declines, the value of the debt rises, so P declines.

So, to understand low current inflation, the salient fact is the extraordinarily low
expected real return on government debt. It is valuable despite poor fiscal prospects,
not because of great ones.

15.4 What about 2008? And cyclical inflation?

Discount rate variation is likely also to be crucial to understanding episodes such as

2008, and the cyclical correlation of inflation.

Fall 2008. Output falls sharply. Deficits expand into the trillions. The growth slow-
down and continuing entitlement problems make future deficits seem even worse. And
inflation... plummets. How is that consistent with the government debt valuation equa-
tion? In every recession, low inflation correlates with large deficits, and vice versa. Isnt
the sign wrong?

Again, it is a mistake to confuse current with expected future deficits. A govern-

ment fighting a war borrows today and simultaneously commits to higher future taxes
to pay off the debt precisely because it does not want to create too much inflation.
Our governments stimulus programs always include at least lip service to future deficit
reduction. The surplus should not be modeled as an AR(1), but as a series in which
a dip today portends a rise in the future. (Cochrane (2001) pursues this point in de-
tail.) Contrariwise, helicopter drop plans to deliberately create inflation come with
commitments not to repay the debt.

But future surpluses didnt get a lot better in 2008. The real answer is the discount
rate. Both nominal and real interest rates dropped sharply in 2008. A flight to quality
104 COCHRANE

further lowered the expected return of government bonds relative to corporate bonds.
People were trying to hold more government bonds and to hold less private assets
and to spend less to get government bonds.

For a bit more analytical perspective, we can write the government debt valuation
formula in a form first introduced in (11)

Bt1 X 1
= Et st+j (59)
Pt Rt,t+j
j=0

We can always discount any asset by its ex-post return. So, a lower expected return on
government debt, coming either from a lower real risk free rate or from a lower (or more
negative!) risk premium raises the value of government debt, and pushes the price level
down.

This mechanism helps as well to understand the general cyclical correlation of in-
flation. In any recession, output falls, deficits rise, and yet inflation falls. Why? In part,
people understand that current deficits correspond to larger future surpluses. But the
most important part of the effect is likely that people are willing to hold claims to the
same surplus at much lower rates of return in a recession than they are in a boom.

Why in a recession do people want to defer consumption, sending interest rates

down, and why do they become more risk averse, sending government bond returns
down relative to private assets? Why are real rates of return on government debt so
low? Though good ideas for answers are plentiful, that is a cause beyond the current
point: discount rate variation is plausible, and discount rates rather than expected sur-
pluses likely account for the otherwise puzzling correlation or lack of correlation be-
tween debt, deficits and inflation.

Once again, the government debt valuation formula holds in the equilibrium of any
model, active or passive. Using ex-post returns to discount, it is an identity, which
always holds. It is not a testable equation. So the issue, whether expected surplus vari-
ation or discount rate variation accounts for movements in the price level over time,
is not particular to the fiscal theory. In a passive regime, getting this wrong is less im-
portant to other parts of the model, but the analysis does not rest on active vs. passive
sources of price level and surplus variation.
 INFLATION AND INTEREST RATES 105

15.5 What about 1951?

And what about all the other failed interest rate pegs? Answer: Fiscal policy.

In the face of inflationary fiscal policy, interest rate pegs cannot hold back inflation.
For example, Woodford (2001) analyzes the US peg of the 1940s and early 1950s. He
credits fiscal-theoretic mechanisms for the surprising stability of the interest rate peg
it did last a decade. In his view, it fell apart when the Korean war undermined fiscal
policy. Other countries whose pegs fell apart after WWII motivating Friedman (1968)
(quotes below), were facing difficult fiscal problems. Most historic pegs were enacted
along with price controls and monetary controls as devices to reduce interest payments
on the debt (that was explicit in the US case) and to help otherwise difficult fiscal pol-
icy. In most historic pegs, central banks were trying to hold down rates that otherwise
wanted to rise, by lending out money to banks at low rates, and with financial repres-
sion to force people to hold government debt they did not want to hold. Our central
banks are taking in money from banks who cant find better opportunities elsewhere,
thus apparently holding interest rates above what they otherwise want to be, in the face
of overwhelming demand for government debt. Countries whose pegs fell apart had
problems financing current deficits. We have doomsaying forecasts of deficits decades
from now. The lessons of historical pegs under vastly different fiscal circumstances do
not necessarily apply to all pegs.

Interest rate pegs, like exchange rate pegs, live on top of solvent fiscal policy or not
at all. For this reason I have been careful to state the doctrine as an interest rate peg
can be stable.

To be sure, in none of these cases What about secular and cyclical low inflation
with high debts, what about past pegs that blew up, etc. do I attempt to provide a
solid answer. The point here is not to establish these speculations as definitive answers.
The point is that there exists a vaguely plausible reconciliation of the data with the gov-
ernment debt valuation equation; that the model passes the laugh test and cannot be
instantly dismissed with obvious facts.
106 COCHRANE

15.6 The uncomfortable future

So far, this paper has presented an unfashionably comfortable view of our future. Low
interest rates just drag inflation down to the Friedman rule. Blips in inflation will melt
away on their own accord under an interest rate peg. Neither instability nor indetermi-
nacy need worry us. The huge balance sheet will just sit there. If the Fed were to raise
rates, inflation will gently rise to meet the new higher rates, after a possible short small
inflation decline buried deep in fiscal theory mechanics of long term debt.

But the tenuous fiscal underpinnings of our anchoring could fall apart quickly.
Most debt is short term, rolled over much more quickly than the maturity of any plau-
sible return to surpluses. Government debt is valued because each investor thinks he or
she will be paid from the proceeds of a roll over. Short term debt and long term illiquid
assets (the present value of a governments taxing power) are prone to runs.

As the factors driving low rates of return on government debt, low r g and hence
high valuations for government debt, are a bit mysterious, that gives us little confidence
against their sudden evaporation. Higher real interest rates times a large outstanding
stock of debt will put a sharp dent into budgets. The debt sustainability literature often
takes a view of multiple equilibria low interest rates make debt sustainable, and result
in low interest rates. Higher interest rates make debt unsustainable and lead to higher
interest rates, via either default or inflation premiums. (Had our governments chosen
much longer-term financing, we would not be exposed to runs or bad news in this way.)

The simple Gordon growth formula for the fiscal theory, (58) offers some insight to
these questions. We are deep in the region where r is close to g, so small changes in r
can have large effects.

On the other hand, small increases in our desultory growth rates g can have equally
large disinflationary effects. Long-run GDP growth rates are the prime determinant of
long-run surpluses.

So, there is both good and bad r. Good r is a rise in interest rates that results from
a rise in the economic growth rate. From the basic first order condition,

r = + (g n)
 INFLATION AND INTEREST RATES 107

where is the rate of time preference, = 1/ is the inverse substitution elasticity, and
n is population growth so g n is per capita growth. If = 1, log utility, then an increase
in economic growth will increase real interest rates one for one, and we live in a happy
world in which the value of government debt is unaffected by growth or interest rate
variation. Typical calibrations put slightly above one, but primary surpluses also are
more than 1-1 sensitive to economic growth, especially when deficits are so strongly
the result of entitlement programs rather than discretionary spending.

Bad r is a rise in expected rate of return on government debt that comes without
a rise in economic growth, or with a decline in economic growth. If bond markets get
scared of a default, or inflation, the risk premium on government debt will rise, and we
get the full inflationary effect of a rising rate of return with no concurrent growth rate
effects.

The bottom line: some historic interest rate pegs, like exchange rate pegs and the
gold standard, lasted a surprisingly long time. Many interest rate pegs fell apart when
their fiscal foundations fell apart. With short term debt, that can happen in what feels
like speculative attack bubble or run to central bankers. Inflations resurgence can
happen without Phillips curve tightness, and can surprise central bankers of the 2020s
just as it did in the 1970s, and just as inflations decline surprised them in the 1980s,
and its stability surprised them in the 2010s.

16 Literature

I boil these points down to very simple models, but we should not lose sight of how
deep and widespread the doctrines of inflation instability or indeterminacy are, how
radical it is to suggest that those doctrines are mistaken.

Milton Friedman (1968) gives the classic statement that an interest rate peg is unsta-
ble, and that higher interest rates lead to temporarily lower inflation. Friedman writes
(p.5) that Monetary policy cannot peg interest rates for more than very limited peri-
ods... Friedmans prediction also comes clearly from adaptive expectations: (p. 5-6):

Let the higher rate of monetary growth produce rising prices, and let
108 COCHRANE

the public come to expect that prices will continue to rise. Borrowers will
then be willing to pay and lenders will then demand higher interest rates-as
Irving Fisher pointed out decades ago. This price expectation effect is slow
to develop and also slow to disappear.

Friedman stressed a money growth channel rather than an IS channel low inter-
est rates require money growth; money growth leads to more spending and eventually
more inflation, more inflation puts upward pressure on interest rates unless money
growth increased further, and so on. But the bottomline dynamics from interest rate to
inflation does not depend on this view of the mechanism. The very simple Keynesian
model of Figure 4 captures completely Friedmans description of inflation instability
and interest rate policy under adaptive expectations.

Friedman was heavily influenced by recent history of his time:

These views [ineffectiveness of monetary policies] produced a widespread

adoption of cheap money policies after the war. And they received a rude
shock when these policies failed in country after country, when central bank
after central bank was forced to give up the pretense that it could indefi-
nitely keep the rate of interest at a low level. In this country, the pub-
lic denouement came with the Federal Reserve-Treasury Accord in 1951,
although the policy of pegging government bond prices was not formally
abandoned until 1953. Inflation, stimulated by cheap money policies, not
the widely heralded postwar depression, turned out to be the order of the
day. The result was the beginning of a revival of belief in the potency of
monetary policy.

Above I noted that all countries including the US, were also managing fiscal pol-
icy stresses and large wartime debts, and that the bond-pegging policy worked for a
surprisingly long time for an unstable equilibrium.

Most of all, we do not know how Friedman, an intensely empirical economist, might
have adapted his views on the stability of an interest rate peg with our recent 8 years
of experience, or Japans 20 in mind, rather than the inflations of the immediate WWII
aftermath.
 INFLATION AND INTEREST RATES 109

Though the literature developed over 20 years, Taylor (1999) provides a clear state-
ment that old-Keynesian (backward-looking, adaptive expectations) models are unsta-
ble, and that Taylor rules induce stability.

The doctrine that inflation is indeterminate under an interest rate peg or passive
policy, under rational expectations, started with Sargent and Wallace (1975). The basic
point: fixing it with it = rt + Et t+1 leaves t+1 Et t+1 indeterminate. Though often
confused, their point is quite different from Friedmans. Indeterminacy is not the same
thing as instability, and neither is the same thing as volatility!

The fiscal theory of the price level goes back to Adam Smith:

A prince who should enact that a certain proportion of his taxes should
be paid in a paper money of a certain kind might thereby give a certain value
to this paper money Wealth of Nations, Book 2, Ch. II.

Monetary economists have always recognized the importance of monetary-fiscal inter-

actions. It did not escape notice that inflations since Diocletians stem from fiscal prob-
lems, and that printing up money to cover debts as in our revolutionary and civil wars
causes inflation. Much of the structure of our monetary arrangements comes down to
institutions that prevent inflationary finance.

The modern resurgence and deep elaboration owes much to Sargent (2013) (first
published in the early 1980s). Leeper (1991) showed how the fiscal theory can uniquely
determine the price level under passive monetary policy. Sims (1994) clearly states that
the fiscal theory and rational expectations overturn Friedmans doctrine of instability,
as well as Sargent and Wallaces indeterminacy:

A monetary policy that fixes the nominal interest rate, even if it holds the
interest rate constant regardless of the observed rate of inflation or money
growth rate, may deliver a uniquely determined price level.

The observation that interest rate pegs are stable in forward-looking new Keyne-
sian models also goes back a long way. Woodford (1995) discusses the issue. Woodford
(2001) is a clear statement, analyzing interest rate pegs such as the WWII US price sup-
port regime, showing they are stable so long as fiscal policy cooperates.
110 COCHRANE

Benhabib, Schmitt-Grohe, and Uribe (2002) is a classic treatment of the zero-rate

liquidity trap. They note that the zero bound means there must be an equilibrium with
a locally passive < 1 Taylor rule, with multiple stable equilibria. However, they view
this state as a pathology, not a realization of the optimal quantity of money and optimal
(low) level of markups, and devote the main point of their paper to escaping the trap
via fiscal policy.

Schmitt-Grohe and Uribe (2014) realize that inflation stability means the Fed could
raise the peg and therefore raise inflation. To them this is another possibility for escap-
ing a liquidity trap. They write

The paper... shows that raising the nominal interest rate to its intended
target for an extended period of time, rather than exacerbating the recession
as conventional wisdom would have it, can boost inflationary expectations
and thereby foster employment.

The simple model here disagrees that raising inflation raises employment. With a
forward-looking Phillips curve, rising inflation reduces employment here, as in Figure
11.

Belaygorod and Dueker (2009) and Castelnuovo and Surico (2010) estimate new-
Keynesian / DSGE models allowing for switches between determinacy and indetermi-
nacy. They find that the model displays the price puzzle interest rate shocks lead to
rising inflation, starting immediately in the indeterminacy region < 1, as I do.

The possibility long periods of low rates cause deflation, so raising interest rates
will raise inflation, has had a larger recent airing in speeches and the blogosphere.
(Williamson (2013), Cochrane (2013, 2014b), Smith (2014).) Minneapolis Federal Re-
serve Chairman Narayana Kocherlakota (2010) suggested it in a famous speech:

Long-run monetary neutrality is an uncontroversial, simple, but nonethe-

less profound proposition. In particular, it implies that if the FOMC main-
tains the fed funds rate at its current level of 0-25 basis points for too long,
both anticipated and actual inflation have to become negative. Why? Its
simple arithmetic. Lets say that the real rate of return on safe investments
is 1 percent and we need to add an amount of anticipated inflation that will
 INFLATION AND INTEREST RATES 111

result in a fed funds rate of 0.25 percent. The only way to get that is to add a
negative number in this case, 0.75 percent.
To sum up, over the long run, a low fed funds rate must lead to consistent
but lowlevels of deflation.

To be clear, Friedman (1968) disagrees. Friedman views the Fisher equation as a

steady state, but it is an unstable steady state. Kocherlakota, seeing recent experience,
views it as a stable one. Friedmans long-run neutrality is that a permanent rise in M
gives rise to a permanent rise in P , but a permanent rise in i will lead to explosive P
and . A permanent rise in M gives rise to a permenent rise in P and therefore of both i
and , but not the other way around . Kocherlakotas definition of long-run neutrality
is that a permanent rise in i will lead to a permanent rise in . The difference between
a stable and an unstable steady state is key.

Cochrane (2014a) works out a model with fiscal price determination, an interest rate
target, and simple k-period price stickiness. Higher interest rates raise inflation in the
short and long run, just as in this paper, but the k-period stickiness leads to unrealistic
dynamics.

Following Woodford (2003), many authors have also tried putting money back into
sticky-price models. Benchimol and Fourcans (2012) and Benchimol and Fourcans
(2015) study a CES money in the utility function specification as here, in a detailed
model applied to the Eurozone. They find that adding money makes small but impor-
tant differences to the estimated model dynamics.

Ireland (2004) also adds money in the utility function. In his model, money also
spills over into the Phillips curve. He writes, (p. 974)

... optimizing firms set prices on the basis of marginal costs; hence, the
measure of real economic activity that belongs in a forward-looking Phillips
curve ... is a measure of real marginal costs, rather than a measure of de-
trended output ... in this model, real marginal costs depend on real wages,
which are in turn linked to the optimizing households marginal rate of sub-
stitution between consumption and leisure. Once again, when utility is non-
separable, real balances affect this marginal rate of substitution; hence, in
112 COCHRANE

this case, they also appear in the Phillips curve.

However, he finds that maximum likelihood estimates lead to very small influences of
money, a very small if not zero cross partial derivative ucm .

Where Irelands Phillips curve comes from quadratic adjustment costs, Andres, Lopez-
Salido, and Valles (2006) find a similar result from a Calvo-style pricing model. Their
estimate also finds no effects of money on model dynamics.

Many have noted that the stability and quiet (lack of volatility) of inflation at steady
interest rates is a puzzle. Homburg (2015) points out Japans benign liquidity trap
and the deep puzzle it poses for standard models. He advocates a repair with pervasive
credit constraints.

17 Concluding comments

The observation that inflation has been stable or gently declining and quiet at the zero
is important evidence against the classical view that an interest rate peg must be un-
stable, and the new-Keynesian view that it leads to sunspot volatility. If an interest rate
peg is stable, then raising the interest rate peg should raise inflation, at least eventually.

Conventional new-Keynesian models predict that inflation is stable. Adding the fis-
cal theory of the price level, or related rules for selecting nearby equilibria, removes
indeterminacies and sunspots and leads to a very simple monetary model consistent
with our recent experience

Those models also predict that raising interest rates will raise inflation, both in the
long and short run. My attempts to escape this prediction by adding money, backward
looking Phillips curves, multiple equilibria or Taylor rules all fail.

The new-Keynesian model plus fiscal theory and long-term debt does produce a
temporary negative inflation response to unexpected interest rate increases. It is sim-
ple, and economic, but quite novel relative to standard monetary models. It does not
produce the standard, adaptive-expectations view of a permanent disinflation such as
the 1980s, nor does it justify policy exploitation of the negative sign, especially in sys-
tematic, Taylor-principle form.
 INFLATION AND INTEREST RATES 113

This paper was also a search for a simple model that captures the effects of mone-
tary policy, but overcomes the critiques of active and instability-inducing Taylor rules
in forward-looking models. The new-Keynesian plus fiscal theory model in this paper
satisfies that criterion.

A review of the empirical evidence finds very weak support for the standard the-
oretical view that raising interest rates lowers inflation, and much of that evidence is
colored by the imposition of strong priors of that sign.

I conclude that a positive reaction of inflation to interest rate changes is a possibility

we, and central bankers, ought to begin to take seriously.

Both the fact of quiet inflation and the theory here rehabilitate interest rate pegs as
at least a possible policy. We can live the Friedman rule of low interest, low inflation,
and enormous reserves. Real policies may choose a time-varying peg if central banks
think they can offset shocks (vti here may react to vtr ), and may desire headroom of
higher inflation to do that. But there is no need to fear catastrophe of inflation from the
former policy configuration.

However, that quiet depends on fiscal foundations. The large demand for govern-
ment debt which provides the fiscal foundations of this quiet (under either active or
passive views), driven by low discount rates not strong fiscal policies, could evaporate
as unexpectedly as it arrived.
114 COCHRANE

References
Andres, Javier, J. David Lopez-Salido, and Javier Valles. 2006. Money in an Estimated Business
Cycle Model of the Euro Area. The Economic Journal 116 (511):457477. URL http://dx.doi.
org/10.1111/j.1468-0297.2006.01088.x.

Angeletos, George-Marios and Chen Lian. 2016. Forward Guidance without Common Knowl-
edge. Manuscript URL http://economics.mit.edu/files/12277.

Atkeson, Andrew, Vardarajan V. Chari, and Patrick J. Kehoe. 2010. Sophisticated Monetary Poli-
cies. Quarterly Journal of Economics 125 (1):4789.

Ball, Laurence. 2016. Ben Bernanke And The Zero Bound. Contemporary Economic Policy,
Western Economic Association International 34:720.

Belaygorod, Anatoliy and Michael Dueker. 2009. Indeterminacy, Change Points and the Price
Puzzle in an Estimated DSGE Model. Journal of Economic Dynamics and Control 33 (3):624
648.

Benchimol, Jonathan and Andre Fourcans. 2012. Money and risk in a DSGE framework: A
Bayesian application to the Eurozone. Journal of Macroeconomics 34:95111.

. 2015. Money and monetary policy in the Eurozone: an empirical analysis during
crises. Macroeconomic Dynamics Forthcoming.

Benhabib, Jess, Stephanie Schmitt-Grohe, and Martin Uribe. 2001. Monetary policy and mul-
tiple equilibria. American Economic Review :167186.

Benhabib, Jess, Stephanie Schmitt-Grohe, and Martn Uribe. 2002. Avoiding Liquidity Traps.
Journal of Political Economy 110 (3):535563.

Brash, Donald T. 2002. Inflation Targeting 14 Years On. URL http://www.bis.org/review/

r020121b.pdf. Speech given at 2002 American Economics Association Conference.

Castelnuovo, Efrem and Paolo Surico. 2010. Monetary Policy, Inflation Expectations and The
Price Puzzle. The Economic Journal 120:12621283. URL doi:10.1111/j.1468-0297.2010.
02368.x.

Christiano, Lawrence, Martin Eichenbaum, and Charles Evans. 1999. Monetary policy shocks:
what have we learned and to what end? In Handbook of Macroeconomics, edited by Michael
Woodford and John B. Taylor. Amsterdam: North-Holland, 65148.
 INFLATION AND INTEREST RATES 115

. 2005. Nominal Rigidities and the Dynamic Effects of a Shock to Montary Policy. Jour-
nal of Political Economy 113, (145).

Clarida, Richard, Jordi Gal, and Mark Gertler. 2000. Monetary Policy Rules and Macroeco-
nomic Stability: Evidence and Some Theory,. Quarterly Journal of Economics 115:147180.

Cochrane, John H. 2001. Long term debt and optimal policy in the fiscal theory of the price
level. Econometrica 69:69116.

. 2005. Money as Stock. Journal of Monetary Economics 52:501528.

. 2011. Determinacy and Identification with Taylor Rules. Journal of Political Economy
119:565615.

. 2013. What if we got the sign wrong on monetary policy? URL http:
//johnhcochrane.blogspot.com/2013/12/what-if-we-got-sign-wrong-on-monetary.html.
Grumpy Economist Blog Post.

. 2014a. Monetary Policy with Interest on Reserves. Journal of Economic Dynamics and
Control 49:74108.

. 2014b. The neo-Fisherian Question. URL http://johnhcochrane.blogspot.com/2014/

11/the-neo-fisherian-question.html. Grumpy Economist Blog Post.

. 2014c. The New-Keynesian Liquidity Trap. Manuscript URL http://faculty.

chicagobooth.edu/john.cochrane/research/papers/zero bound 2.pdf.

. 2015. A New Structure For U. S. Federal Debt. In The $13 Trillion Question: Managing
the U.S. Government?s Debt, edited by David Wessel. Brookings Institution Press, 91146.

. 2016a. Comments on A Behavioral New-Keynesian Model by Xavier Gabaix.

Manuscript URL http://faculty.chicagobooth.edu/john.cochrane/research/papers/gabaix
comments.pdf.

. 2016b. Stepping on a Rake: Replication and Diagnosis. manuscript URL http:

//faculty.chicagobooth.edu/john.cochrane/research/papers/sims replication.pdf.

Dueker, Michael J. 2006. The Price Puzzle: An Update and a Lesson. Economic Synopses 4:1.
URL https://research.stlouisfed.org/publications/es/06/ES0624.pdf.

Eggertsson, Gauti and Neil Mehrotra. 2014. A Model of Secular Stagnation. Manuscript .

Eichenbaum, Martin. 1992. Comment on Interpreting the macroeconomic time series facts:
the effects of monetary policy. European Economic Review, 36 (5):10011011.
116 COCHRANE

Friedman, Milton. 1968. The Role of Monetary Policy. The American Economic Review 58:1
17.

Gabaix, Xavier. 2016. Behavioral Macroeconomics Via Sparse Dynamic Programming.

Manuscript .

Garca-Schmidt, Mariana and Michael Woodford. 2015. Are Low Interest Rates Deflationary? A
Paradox of Perfect-Foresight Analysis. Manuscript.

Gurkaynak, Refet S., Brian Sack, and Jonathan H. Wright. 2007. The U.S. Treasury yield curve:
1961 to the present. Journal of Monetary Economics 54 (8):2291 2304. Data at https://www.
federalreserve.gov/econresdata/researchdata/feds200628.xls.

Hall, George J. and Thomas J. Sargent. 2015. A History of U.S. Debt Limits. Manuscript URL
http://people.brandeis.edu/ghall/papers/debt limit 12 08 2015.pdf.

Hanson, Michael. 2004. The Price Puzzle Reconsidered. Journal of Monetary Economics 51.
URL http://www.sciencedirect.com/science/article/pii/S0304393204000959.

Homburg, Stefan. 2015. Understanding Benign Liquidity Traps: The Case of Japan.
German Economic Review URL http://diskussionspapiere.wiwi.uni-hannover.de/pdf bib/
dp-557.pdf.

Ireland, Peter. 2004. Moneys Role in the Monetary Business Cycle. Journal of Money, Credit,
and Banking 36 (6):969983. URL http://irelandp.com/pubs/role.pdf.

King, Robert G. 2000. The New IS-LM Model: Language, Logic, and Limits. Federal Reserve
Bank of Richmond Economic Quarterly 86 (3):45104.

Kocherlakota, Narayana. 2010. Inside the FOMC. URL https://www.minneapolisfed.org/

news-and-events/presidents-speeches/inside-the-fomc. Speech.

Kocherlakota, Narayana R. 2016. Fragility of Purely Real Macroeconomic Models. Manuscript

.

Leeper, Eric M. 1991. Equilibria Under Active and Passive Monetary and Fiscal Policies.
Journal of Monetary Economics 27:129147.

Lucas, Robert E. 1972. Expectations and the Neutrality of Money. Journal of Economic Theory
4:103124.

Mankiw, N. Gregory and Ricardo Reis. 2002. Sticky Information versus Sticky Prices: A Proposal
to Replace the New Keynesian Phillips Curve. Quarterly Journal of Economics 117:1295
1328.
 INFLATION AND INTEREST RATES 117

Ramey, Valerie. 2015. Macroeconomic Shocks and Their Propagation. Manuscript.

Sargent, Thomas J. 2013. Rational Expectations and Inflation (Third Edition). Princeton Uni-
versity Press. URL http://www.jstor.org/stable/j.ctt2jc97n.

Sargent, Thomas J. and Neil Wallace. 1975. Rational Expectations, the Optimal Monetary In-
strument, and the Optimal Money Supply Rule. Journal of Political Economy 83:241254.

Schmitt-Grohe, Stephanie and Martn Uribe. 2014. Liquidity Traps: An Interest-Rate-Based

Exit Strategy. The Manchester School 81 (S1):114. URL doi:10.1111/manc.12065.

Sims, Christopher A. 1980. Macroeconomics and reality. Econometrica :148.

Sims, Christopher A. 1992. Interpreting the macroeconomic time series facts: The effects of
monetary policy. European Economic Review 36 (5):9751000.

. 1994. A simple model for study of the determination of the price level and the interac-
tion of monetary and fiscal policy. Economic Theory 4 (3):381399. URL http://dx.doi.org/
10.1007/BF01215378.

. 2011. Stepping on a Rake: The role of fiscal policy in the inflation of the 1970s. Euro-
pean Economic Review 55:4856. URL doi:10.1016/j.euroecorev.2010.11.010.

. 2016. Fiscal Policy, Monetary Policy and Central Bank Independence. Jack-
son Hole URL https://www.kansascityfed.org//media/files/publicat/sympos/2016/
econsymposium-sims-paper.pdf.

Smets, Frank and Raf Wouters. 2003. An Estimated Dynamic Stochastic General Equilibrium
Model of the Euro Area,. Journal of the European Economic Association 1 (5):11231175.

Smith, Noah. 2014. The Neo-Fisherite Rebellion. URL http://noahpinionblog.blogspot.com/

2014/04/the-neo-fisherite-rebellion.html. Noahpinion Blog Post.

Taylor, John B. 1999. The Robustness and Efficiency of Monetary Policy Rules as Guidelines for
Interest Rate Setting by the European Central Bank. Journal of Monetary Economics 43:655
679.

Uhlig, Harald. 2006. What are the effects of monetary policy? Results from an agnostic identi-
fication procedure. Journal of Monetary Economics 52 (2):381419.

Werning, Ivan. 2012. Managing a Liquidity Trap: Monetary and Fiscal Policy. Unpublished
URL https://dl.dropboxusercontent.com/u/125966/zero bound 2011.pdf. Manuscript, MIT.
118 COCHRANE

Wieland, Johannes. 2015. Are Negative Supply Shocks Expansionary at the Zero Lower Bound?
Manuscript .

Williamson, Steven. 2013. Phillips Curves and Fisher Relations. URL http://newmonetarism.
blogspot.com/2013/12/phillips-curves-and-fisher-relations.html. Blog Post.

Woodford, Michael. 1995. Price-Level Determinacy Without Control of a Monetary Aggregate.

Carnegie-Rochester Conference Series on Public Policy 43:146.

. 2001. Fiscal Requirements for Price Stability. Journal of Money Credit and Banking
33 (669-728).

. 2003. Interest and prices. Princeton N. J.: Princeton University Press.

Yellen, Janet. 2016. Macroeconomic Research After the Crisis. URL http://www.federalreserve.
gov/newsevents/speech/yellen20161014a.htm. Speech given at The Elusive Great Recov-
ery: Causes and Implications for Future Business Cycle Dynamics 60th annual economic
conference sponsored by the Federal Reserve Bank of Boston, Boston, Massachusetts.
 INFLATION AND INTEREST RATES 119

18 Appendix. A Phillips curve with lags

The usual Phillips curve (19) is forward looking:

t = Et t+1 + xt (60)

or equivalently

X
t = j Et xt+j . (61)
j=0

To incorporate lagged inflation, I write instead

X
X
t = xt + j Et xt+j + j xtj (62)
j=1 j=1

or, in autoregressive form,

(1 )
t = Et t+1 + t1 + xt .
1 + 1 + (1 + )

So that the sum of coefficients on the right hand side of (62) is the same as it is
in (61), and hence so that the steady state relationship between output and inflation
remains unchanged, I constrain the backward and forward looking coefficients and
to satisfy
(1 )(1 )
= (1 ) . (63)
(1 )

Repeating the model solution, inflation is again a two-sided moving average of in-
terest rates, and in the presence of money of expected changes in interest rates,


(1 ) 3
j3 Et zt+j +
X
t+1 = 1

1

Et zt +

1 3 1 1 3 2 j=1

1 1 1 1 1 1



2 3 1 3
j j
1
X 2
X
+ 1 Etj ztj 2 Etj ztj (64)

1 1
1
1 1
1



2 j=1 2 j=0
120 COCHRANE

where 1 , 2 , 3 are the roots of the polynomial in

(1 + (1 + )) 1 (1 + + + (1 )) 2
+ L L + L3

= L1 1 L1 1 L1 1




1 2 3

The expression (64) now has two backward looking moving averages as well as one for-
ward looking term. The long-run response of inflation to interest rates remains one.

18.1 Roots, moving average, and response

0.9

0.8

0.7

0.6

0.5

0.4
-1
1

0.3 -1
2
3

0.2

0.1

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
, backward-looking AR

Figure 28: Roots of the impulse response function, and forward looking Phillips curve
parameter , for each choice of the backward looking Phillips curve parameter .

Figure 28 displays for each value of the backward looking weight in the Phillips
curve with lagged inflation (62), the forward looking weight given by restriction (63),
and the three roots 1 1
1 , 2 and 3 that govern the moving averages (64).
 INFLATION AND INTEREST RATES 121

Phillips curve t = M A(L)xt Equilibrium t+1 = M A(L)it

= 0.000, = 0.970 = 0.000, = 0.970
1 0.3
0.8
0.2
0.6
0.1
0.4
0
0.2
-0.1
0
-30 -20 -10 0 10 20 30 -30 -20 -10 0 10 20 30

= 0.700, = 0.968 = 0.700, = 0.968

1 0.3
0.8
0.2
0.6
0.1
0.4
0
0.2
-0.1
0
-30 -20 -10 0 10 20 30 -30 -20 -10 0 10 20 30

= 0.942, = 0.942 = 0.942, = 0.942

1 0.3
0.8
0.2
0.6
0.1
0.4
0
0.2
-0.1
0
-30 -20 -10 0 10 20 30 -30 -20 -10 0 10 20 30

= 0.968, = 0.700 = 0.968, = 0.700

1 0.3
0.8
0.2
0.6
0.1
0.4
0
0.2
-0.1
0
-30 -20 -10 0 10 20 30 -30 -20 -10 0 10 20 30

= 0.970, = 0.000 = 0.970, = 0.000

1 0.3
0.8
0.2
0.6
0.1
0.4
0
0.2
-0.1
0
-30 -20 -10 0 10 20 30 -30 -20 -10 0 10 20 30
past; future past; future

Figure 29: Moving average representation of the two-sided Phillips curve, and corre-
sponding moving-average response of inflation to interest rates.
122 COCHRANE

The left hand column of Figure 29 presents the moving average representation of the
Phillips curve (62), and the right hand column presents the moving average coefficients
of the solution (64), for specific choices of and the consequent . These specific values
of , are represented as dark circles in Figure 28.

Starting at the left of Figure 28, and the top of Figure 29, we have the previous case
= 0, = of a purely forward looking Phillips curve. Figure 28 shows the nearly equal
forward and backward looking roots 1
2 and 3 , and = . The top left element of
Figure 29 shows the purely forward looking weights of the Phillips curve, while the top
right element shows the nearly equal forward and backward moving average weights of
the models solution for inflation as a function of interest rates.

As we raise the backward looking coefficient , Figure 28 shows that the forward
looking coefficient , the forward looking root in the solution 3 and the original back-
ward looking root 1
2 change little. We bring in a second backward looking root, roughly
equal to itself. Around = 0.55, the two backward looking roots become complex.
Their magnitude is still less than one, but the complex nature will generate a damped
sinusoidal response function.

The second row of Figure 29 shows the Phillips weights and response function for
a backward Phillips curve coefficient = 0.7. The forward looking coefficient is
still large, so this case captures a small amount of backward looking behavior, and helps
us to assess if a small amount of such behavior can substantially change results. The
moving average solution in the right column is still basically two-sided and positive.
It begins to weight the past more than the future. The small change in this moving
average previews the result below that this small amount of backward looking behavior
in the Phillips curve will not materially affect the neo-Fisherian response to an interest
rate rise.

Continuing to the right in Figure 28, the backward looking roots continue to grow
nearly linearly with the backward looking Phillips parameter . The forward looking
coefficient and the forward looking root 3 remain nearly unaffected however, for
even very large values of .

The third row of Figure 29 shows the case = that the Phillips curve is equally
backward- and forward looking. The right column shows that the response function is
 INFLATION AND INTEREST RATES 123

now weighted more to the past than the future. In addition, negative coefficients are
starting to show up, giving us some hope that higher interest rates can result in lower
inflation at some point along the dynamic path.

The fourth and fifth rows of Figure 29 show cases in which the Phillips curve be-
comes more and more backward looking. The fourth row shows a forward weight re-
duced to = 0.7, and the fifth row shows the purely backward looking case = 0,
= . Figure 28 shows that only for very large values of the backward looking coeffi-
cient near = do and the forward looking root 3 substantially decline. At that
limit, both forward looking terms disappear, and the two complex backward looking
roots remain. Figure 29 shows that the moving average solution becomes more and
more weighted to past values, with larger sinusoidal movements.

Figure 30 presents the response of inflation, on the left, and output, on the right,
to the standard step function interest rate path, for the same choices of forward and
backward looking Phillips curve parameters as in the last two figures. The dashed line
in each case is the unexpected case, verifying that once again expected and unexpected
paths are the same for dates after the announcement.

Starting at the top of Figure 30, we have the purely forward looking case = 0, =
, and the same result as before. Inflation rises smoothly to meet the higher interest
rate, and the Phillips curve produces a small output reduction on the whole path.

In the second row, a little bit of backward looking behavior = 0.7 produces a plot
that is almost visually indistinguishable. Inflation rises throughout, and output is still
depressed until the long-run inflation-output tradeoff of this model takes hold. So, the
basic result is robust to adding backward looking behavior.

As we go to more and more backward looking behavior in the remaining rows of

Figure 30, inflation and output cease to respond ahead of the funds rate rise. Backward
looking Phillips curves mean that forward guidance has less and less effect.

However, in none of the cases does a rise in interest rates provoke a decline in infla-
tion. We can see the reason in the moving average coefficients of Figure 29. Though that
figure does have some negative coefficients, in which past interest rates lower current
inflation, the coefficients at low lags are always positive, and outweigh the negative
124 COCHRANE

Inflation response to rate rise Output response to rate rise

= 0.000, = 0.970 = 0.000, = 0.970
1 1

0.8
0.5
0.6
0
0.4
-0.5
0.2

0 -1
-10 0 10 20 -10 0 10 20

= 0.700, = 0.968 = 0.700, = 0.968

1 1

0.8 0.5
0.6
0
0.4
-0.5
0.2

0 -1
-10 0 10 20 -10 0 10 20

= 0.942, = 0.942 = 0.942, = 0.942

1
1

0.8 0.5

0.6 0
0.4
-0.5
0.2
-1
0
-10 0 10 20 -10 0 10 20

= 0.968, = 0.700 = 0.968, = 0.700

1.2
1
1
0.8
0.8 0.6
0.6 0.4
0.4 0.2
0
0.2
-0.2
0
-10 0 10 20 -10 0 10 20

= 0.970, = 0.000 = 0.970, = 0.000

1.5

1 1

0.5
0.5
0

0 -0.5
-10 0 10 20 -10 0 10 20

Figure 30: Response of inflation and output to a step-function rise in interest rates,
with lagged inflation in the Phillips curve. Solid lines are the response to an expected
change, dashed lines are the response to an unexpected change. The backward-looking
Phillips parameter is , and is the forward-looking parameter.
 INFLATION AND INTEREST RATES 125

coefficients further in the past. Integrating, we see the overshooting behavior in the
bottom of Figure 30.

As the Phillips curve becomes more backward looking, the output decline with an
interest rate rise weakens, and eventually becomes an output rise. While the forward
looking Phillips curve gives higher output when inflation is declining, the backward
looking Phillips curve gives higher output when inflation is rising. In this experiment
inflation does rise, so output rises as well. Moving to a backward looking Phillips curve,
we did turn around a sign: the right decline in output turned into a wrong rise in
output, leaving the wrong rise in inflation alone.

Together, then, the backward looking Phillips curve and the neo-Fisherian behav-
ior of inflation mean that in interest rate rise looks much like what is conventionally
expected of a monetary expansion, not a contraction, plus some interesting slow sinu-
soidal dynamics.

Figure 31 adds both money and a purely backward looking Phillips curve. Com-
pare this result to Figure 17 for money with a forward looking Phillips curve, and to the
bottom row of Figure 30 for a backward looking Phillips curve without money.

Figure 31 produces something like the standard intuition, for inflation, at last. The
unanticipated rate rise still does not interact with money at all, so it produces the same
response for all values of money m/c. But the anticipated rate rise now benefits from
the postponement of its response, from the backward looking Phillips curve, and also
the reductions in near-term inflation from adding money. Now there is a temporary re-
duction in inflation before inflation rises to join the interest rate, for any positive value
of money m/c.

Output now also falls, before rising with inflation. The backward looking Phillips
curve generates low output when inflation is decreasing, and high output when infla-
tion is increasing.

However, comparing the results to Figure 17 with a forward looking Phillips curve,
the benefit is small. That figure already showed a temporary inflation decline, and no
oscillating dynamics. The inflation decline is larger now, and smoother, but not dra-
matically different. We still need substantial monetary distortions m/c > 1 to obtain a
126 COCHRANE

Inflation

1.5 i
Un, expected
, expected

1
Percent response

0.5
m/c=0

m/c=1

0
m/c=2

-0.5

m/c=4

-4 -2 0 2 4 6
Time
Output

i
x, unexpected
x, expected
2

1
Percent response

0 m/c=0

m/c=1
-1

m/c=2

-2

-3 m/c=4
-4 -2 0 2 4 6
Time

Figure 31: Response to expected and unexpected interest rate rise, with money and a
purely backward looking Phillips curve = , = 0.
 INFLATION AND INTEREST RATES 127

quantitatively interesting response, or large > 1 (not shown).

The conventional sign of the short-run output response along with lower short-run
inflation is perhaps the greatest benefit.

But the cost is throwing out all of the forward looking optimizing microfoundations
of the forward looking Phillips curve. Anything much less that purely backward looking
behavior (not shown) does not produce significant improvements.

Also, since unanticipated interest rate rises have no interaction with money, this
modification does not help to match VAR evidence and intuition that focuses on unan-
ticipated changes in interest rates.

Finally, the long-run responses still defy conventional intuition, losing the smooth
decline present in the simple model of Figure 11. The disinflation and output cooling
are borrowed from future inflation and an output boom.

18.2 Backward-looking Phillips algebra

The Phillips curve with lagged inflation is

X
X
t = xt + Et j xt+j + j xtj
j=1 j=1

L1
 
L
= Et 1 + + xt
1 L1 1 L

or, in autoregressive form

 
1
t = Et xt (65)
(1 L1 ) (1 L)

Et 1 + L1 L t = (1 ) xt .

(1 )
t = Et t+1 + t1 + xt .
1 + 1 + (1 + )

To maintain the same steady state relationship between output and inflation, I con-
strain and to satisfy
(1 ) 1
= .
(1 )(1 ) (1 )
128 COCHRANE

So, for each choice of the weight on past inflation, I use a weight on forward
inflation given by

= . (66)
1 + 2
The case = occurs where

= .
2

Now, to solve the model. To keep the algebra simple I find the perfect foresight
solution and put the Et back in at the end. The IS curve is

xt = xt+1 + t+1 zt

(1 L1 )xt = L1 t zt

Forward-differencing (65) and substituting,

 
1 1
1 L1 xt


(1 L )t =
(1 L1 ) (1 L)

 
1 1
(1 L )t = (L1 t zt )
(1 L1 ) (1 L)

(1 L1 ) 1 L1 (1 L) (1 ) L1 t = (1 ) zt



L + (1 + (1 + )) (1 + + + (1 )) L1 + L2 t = (1 ) zt
 

 
1 + (1 + ) 1 1 + + + (1 ) 2 3
+ L L +L Lt = (1 ) zt

Denoting the three roots of the lag polynomial 1

i ,

Et L1 1 L1 1 L1 1




1 2 3 Lt = (1 ) zt .

I find these roots numerically. Nonetheless we can characterize them somewhat.

Evaluating the lag polynomial at L = 0 we have

1 2 3 = . (67)

 INFLATION AND INTEREST RATES 129

while L = 1 gives
 
1 + (1 + ) 1 + + + (1 )
1 1 1




1 1 1 2 1 3 = + +1

(1 )
1 1 1 1 1 1




1 2 3 =

We will have 1 , 2 > 1, 3 < 1 so its convenient to write the result as

1 1 1 1 1



 1
3 1 L 1 2 L 1 3 L L t = (1 ) zt

1
t+1 = 3 (1 ) z
1 t
1 1 L 1 1
1



2 L (1 3 L )

I use the partial fractions identity

3
==
3

1 1 1
1 3 1



1 1
1 2 L (1 3 L ) 1 3 1
1 L 2

!
1 1 1 1 1 1



3 L1 1 3 L 1 3 L
1+ + 1 1 2 1
1
2 1 2
(1 3 L1 ) 1 1 1 1
1



2 1 1 L 1 2 1 2 L

This takes pages of algebra to derive. Its easier just to check it. Thus, and reinserting
the Et
(1 )
t+1 = 1

3

1 3 1 1 3 1
2
" ! #
1 1 1 1 1 1



3 L1 1 3 L 1 3 L
Et+1 1+ + 1 1 2 1
1
2 1 2
Et zt
(1 3 L1 ) 1 1 1 1
1



2 1 1 L 1 2 1 2 L
(68)

The long-run response (L = 1) is

1
t+1 = 3 (1 ) 1 Et zt
1 1



1 1 2 (1 3 )
1
= (1 ) 1

Et zt
1 1 1 1



1 1 2 3
1
= (1 ) Et zt
(1 )
= 1Et zt
130 COCHRANE

The case = 0, = is the conventional forward looking curve. The case = 0,

= , is a purely backward looking curve. In this case, the solution is

(1 + ) L1 + (1 + ) L2 Lt = zt
 

 
(1 + ) 1 2
L +L (1 + ) Lt = zt
(1 + ) (1 + )

L 1 L 1
 1
1

1 2 (1 + ) Lt = (1 ) zt

q
(1 + ) (1 + )2 4 (1 + )
1
1 =
2 (1 + )
q
(1 + ) (1 )2 4
1
2 =
2 (1 + )
3 = 0

1 1 1 1 1



1 L 2 L (1 + ) L t = zt

1
t+1 =
zt
(1 + ) 1 1 1


1 L 1 2 L
!!
1 1
1 L
1
2 L

t+1 = 1 + 1 1
1 1
2 zt
(1 + ) 1 1
2
1
1 1 L 1 1
2 L

19 Algebra Appendix

19.1 Variance of inflation in forward vs. backward looking models

This appendix works out the variance of inflation under the forward-looking vs. backward-
looking solutions, equation (16).

The forward-looking solution is (9)


1 + j+1
X 
i r
t = Et (vt+j vt+j ).
1 + 1 +
j=0
 INFLATION AND INTEREST RATES 131

With v i = 0 and vtr = vt1

r + rt ,


1 + j+1 j r
X 
t = vt .
1 + 1 +
j=0

 
1
t =   vtr .
1 + 1 1+

1+

t = vr .
1 + (1 + ) t

t = vr . (69)
(1 + ) (1 ) + ( 1) t
for = 1,
1
t = vr .
2 (1 ) + ( 1) t
1 1
2 (t ) = 2
2 1 2 r .
[2 (1 ) + ( 1)]

The backward-looking solution is (8)

1 + r
t+1 = t v + t+1
1 + 1 + t

With t = 0, = 0,
1 r
t+1 = t v
1 + 1 + t
Working out the variance of an AR(1) with AR(1) shock,

k
(1 aL)t+1 = r
1 L t
1
a ; k
1 + 1 +

1 1
t+1 = kr
1 aL 1 L t
 
1 a
t+1 = krt
a 1 aL 1 L
1 
(a )t + (a2 2 )t1 + (a3 3 )t2 + ... krt

t+1 =
a
132 COCHRANE

 2
1
2t+1 (a )2 + (a2 2 )2 + (a3 3 )2 + ... k 2 2rt
 
=
a
 2
1
2t+1
 2
(a + 2 2a) + a4 + 4 2a2 2 ) + (a6 + 6 2a3 3 ) + ... k 2 2rt


=
a
2 
a2 2
 
1 a
2t+1 = 2
+ 2
2 k 2 2rt
a 1a 1 1 a
(1 + a) 1
2t+1 = k 2 2rt
(1 a) (1 a ) (1 2 )
2

Note the naive answer is wrong

1 1 1
2t+1 6= 2
v2r = k 2 2rt .
1a 1 a 1 2
2

Plugging in a and k,
 
1
1+ 1+ 1


2
2t+1 =  2  2vt
1 1
 1 +
1+
1
1 1+ (1 2 )

(1 + + ) ()2
2t+1 = 2r
(1 + ) (1 + )2 1 (1 2 ) t
 

For = 1,
(2 + ) 1
2t+1 = 2r
(2 ) 3 (1 2 ) t

The ratio of the two solutions is:

backward (2 + ) [2 (1 ) + ( 1)]2
= .
forward (2 ) 3

19.2 Fiscal theory formulas for delayed and temporary rate rises

Here I work out the algebra for impulse response functions of the fiscal theory with
long term debt model, with an announcement M years ahead of the interest rate rise,
and an interest rate rise that only lasts M years, in both continuous and discrete time,
Equations (29)-(30).
 INFLATION AND INTEREST RATES 133

An interest rate rise from i to i that only lasts M years, continuous time:
 Z M Z 
i j j i M i(jM ) j Bt s
e e dj + e e dj =
0 M Pt r
"
#
e(i +)M 1 e(i +)M Bt s
+ =
i+ i + Pt r

! 
Pt e(i +)M 1 e(i +)M

1
= + /
Pt i+ i + i+

Pt
    i + 

 i+
1 = 1 e(i +)M 1 1eM
1
Pt i + i +

An announcement of an interest rate rise from i to i that starts in M years, contin-

uous time:  Z M Z 
ij j iM i (jM ) j Bt s
e e dj + e e dj =
0 M Pt r
" #
e(i+)M 1 e(i+)M Bt s
+ =
i + i+ Pt r
! 
Pt e(i+)M 1 e(i+)M

1
= + /
Pt i + i+ i+

Pt
   
i+ i+
1 = e(i+)M 1 e M
1
P i + i +

An interest rate rise from i to i that only lasts M years, discrete time:

M 1
X j X M (jM ) Bt1 = s
+
(1 + i )j (1 + i )M (1 + i)(jM ) Pt 1
j=0 j=M

 M  M

1 1+i 1+i
Bt1 s
+ =

P 1
1 1+i 1 1+i

t

Thus,  M  M

1
!
Pt 1+i 1+i 1
1= + / 1

Pt
1 1+i 1 1+i 1 1+i
134 COCHRANE

1

M !
Pt

1 1+i
1= 1 1
Pt 1 + i 1

1 1+i

M ! 
Pt 1 + i 1 + i
 

1= 1 1
Pt 1 + i 1 + it 1 + i

Pt
 
1+i
1 1 M


1
Pt 1 + i

An interest rate rise from i to i that starts in M years, discrete time:

M 1
X j X M (jM ) Bt1 = s
+
(1 + i) j M
(1 + i) (1 + i ) (jM ) Pt 1
j=0 j=M

 M  M

1 1+i 1+i Bt1 s
+ =

P 1
1 1+i 1 1+i

t

Thus,  M
 M


!
Pt 1 1+i 1+i 1
1= + / 1

Pt
1 1+i 1 1+i 1 1+i

M ! !
Pt

1 1 1
1=

Pt 1+i 1 1+i 1 1+i 1 1+i
M 
Pt 1 + i 1 + i
 

1= 1
Pt 1+i 1 + i 1 + i
Pt
 
M 1+i
1 1
Pt 1 + i

19.3 Sticky-price model solution

Here I derive the explicit solutions (20)-(21), for inflation and output given the equilib-
rium path of interest rates. The simple model (18)-(19) is

xt = Et xt+1 (it Et t+1 )

t = Et t+1 + xt .
 INFLATION AND INTEREST RATES 135

The model with money generalizes the IS equation only, to (33)

m
it+1 im m


xt = Et xt+1 + ( ) Et t+1 (it it ) (it Et t+1 ) .
c

We can treat the two cases simultaneously by defining

 
m 
Et it+1 im m


zt it t+1 (it it )
c

and writing the IS equation as

xt = Et xt+1 (zt Et t+1 ) .

One must be careful that lags of zt are lags of expected interest rate changes, not lags of
actual interest rate changes.

Expressing the model in lag operator notation,

Et (1 L1 )xt = Et L1 t zt

Et (1 L1 )t = xt

Forward-differencing the second equation,

Et (1 L1 )(1 L1 )t = Et (1 L1 )xt

Then substituting,

Et (1 L1 ) 1 L1 t = Et L1 t zt

Et (1 L1 ) 1 L1 L1 t = zt



Et 1 (1 + + ) L1 + L2 t = zt .
 

Factor the lag polynomial

Et (1 1 L1 )(1 2 L1 )t = zt
136 COCHRANE

where q
(1 + + ) (1 + + )2 4
i = .
2
Since 1 > 1 and 2 < 1, reexpress the result as

Et (1 1 1 1
 
1 L)(1 2 L )1 L t = zt

Et (1 1 1 1
 
1 L)(1 2 L )t+1 = 1 zt

The bounded solutions are

1
1 1
t+1 = Et+1 1 zt + t+1
1
(1 1 L)(1 2 L ) (1 1
1 L)

where t+1 is a sequence of unpredictable random variables, Et t+1 = 0. I follow the

usual practice and I rule out solutions that explode in the forward direction.

Using a partial fractions decomposition to break up the right hand side,

1 1 2 L1
 
1 1 L
1

1 = 1+ + .
1 1 L (1 2 L1 ) 1 2 1 1
1 L
1 2 L1

So,

1 2 L1
 
1 1 L 1
t+1 = Et+1 1 + 1 + zt + t+1
1 2 1 1 L 1 2 L1
(1 1
1 L)

or in sum notation,


1
j j2 Et+1 zt+j + j
X X X
t+1 = zt + 1 ztj + 1 t+1j . (70)
1 2
j=1 j=1 j=0

We can show directly that the long-run impulse-response function is 1:

1
=
(1 1
1 )(1 2 ) 1 (1 1 )(1 2 )

= = = 1.
(1 (1 + 2 ) + 1 2 ) (1 (1 + + ) + )
 INFLATION AND INTEREST RATES 137

Having found the path of t , we can find output by

xt = t Et t+1 .

In lag operator notation, and shifting forward one period,

xt+1 = Et+1 (1 L1 )t+1

 

1 2 L1 (1 L1 )
   
1 1 L
xt+1 = Et+1 (1 L ) 1 + + z t +Et+1 t+1 .
1 2 1 1
1 L
1 2 L1 (1 1
1 L)

We can rewrite the polynomials to give

"
#
1 1
1 1 1 1
 
1 (1 2 ) 2 L 1
xt+1 = Et+1 + zt + Et+1 t+1 .
1 2 1 1
1 L
1 2 L1 1 1
1 L

(In the second term, I use Et L1 t+1 = 0) or, in sum notation,

 

j j2 Et+1 zt+j +
X X
1 1 1



xt+1 = 1 1 ztj + 1 2
1 2
j=0 j=1

j

X
+ 1 1
1 1 t+1j .
j=0

19.4 Model and solutions in continuous time

It is convenient to work both in discrete and continuous time. To keep the math simple,
I consider the perfect-foresight continuous-time specification, and treat the impulse
response function as a once and for all unexpected shock. The continuous time version
of the model is

dxt
= (it t ) (71)
dt
dt
= t xt (72)
dt
138 COCHRANE

The solution is
Z t Z 
2 t 2 (ts) 1 (st)
t = 0 e + p e is ds + e is ds (73)
2 + 4 s=0 s=t
 Z t Z 
2 t 2 (ts) 1 (st)
xt = 1 0 e +p 1 e is ds 2 e is ds (74)
2 + 4 s=0 s=t

where

1 p 2 
1 = + 4 +
2
1 p 2 
2 = + 4
2

Here I have defined the roots with the square root term first, so both are positive num-
bers. That choice clarifies the notation for continuous time expressions. However,
it means that 1 + 2 shows up where corresponding discrete time expressions have
1 2 , which can cause some confusion.

To derive the solution we proceed as in discrete time. Difference (71),

d2 t dt
2
= + t it
dt dt

d2 t
 
dt
2
t = it .
dt dt
We seek roots of the form
  
d d
1 + 2 t = it (75)
dt dt

in which case
d2 t dt
2
+ (2 1 ) 1 2 t = it . (76)
dt dt
Matching coefficients,

1 2 =

1 2 = .
 INFLATION AND INTEREST RATES 139

Solving,

1 =
1
21 1 = 0,

and hence,

1 p 2 
1 = + 4 +
2
1 p 2 
2 = 1 = + 4 .
2

The solution to (76) is derived in Cochrane (2014c),

Z t Z 
2 t 2 (ts) 1 (st)
t = 0 e + e is ds + e is ds
2 + 1 s=0 s=t

It straightforward to check. The derivatives are

 Z t Z 
dt
= 2 0 e2 t + 2 e2 (ts) is ds + 1 e1 (st) is ds
dt 2 + 1 s=0 s=t

Z t Z
d2 t
 
2 2 t 2 2 (ts) 2 1 (st)
= 2 0 e + (1 + 2 )it + 2 e is ds + 1 e is ds
dt2 2 + 1 s=0 s=t

Plugging these derivatives in to the differential equation (75), we have

 Z t Z 

22 0 e2 t
+ (1 + 2 )it + 2 2
e 2 (ts)
is ds + 1 2
e 1 (st)
is ds
2 + 1 s=0 s=t
 Z t Z 
2 t 2 (ts) 1 (st)
+ 2 0 e 2 e is ds + 1 e is ds
2 + 1 s=0 s=t
Z t Z
()2

2 t 2 (ts) 1 (st)
0 e e is ds + e is ds + it =?0.
2 + 1 s=0 s=t

Collecting terms, the equation holds if

22 + 2 = 0

21 1 = 0
140 COCHRANE

which follow by substitution.

We can find the output response from

dt
xt = t =
dt Z t Z 
2 t 2 (ts) 1 (st)
0 e + e is ds + e is ds
2 + 1 s=0 s=t
 Z t Z 

+ 2 0 e2 t 2 e2 (ts) is ds + 1 e1 (st) is ds
1 + 2 s=0 s=t

Collecting terms
 Z t Z 
2 t 2 (ts) 1 (st)
xt = 1 0 e + ( + 2 ) e is ds + ( 1 ) e is ds
2 + 1 s=0 s=t

or,
 Z t Z 
2 t 2 (ts) 1 (st)
xt = 1 0 e +p 1 e is ds 2 e is ds .
2 + 4 s=0 s=t

19.5 Impulse response function explicit solution

The solution (70) is

j j
j
X X X
t+1 = it + i
1 tj + E t+1 i
2 t+j
+ 1 t+1j
1 2
j=1 j=1 j=0

p
(1 + + ) + (1 + + )2 4
1 =
2
p
(1 + + ) (1 + + )2 4
1 =
2

While it is straightforward to calculate and simulate the solution for a given path of
interest rates, it is useful also to have a formula for the response to a step function. We
want to find the impulse-response function to it = 0, t < 0, and it = i, t = 0, 1, 2, ... The
 INFLATION AND INTEREST RATES 141

interest rate rise is announced at time M , so only M 6= 0. That response is,

t < (M + 1) : t+1 = 0
t
 
2 (t+1+M )
(M + 1) t 0 : t+1 = + 1
M
1 2 1 2
1 t 

1 1 (1 1 ) (t+1+M )
0 < t : t+1 = + + 1 M
1 2 1 2 1 1
1

Proceeding in the same way, the solution for x is

j j
j
X X X
xt+1 = (1 1
1 ) i
1 tj + (1 1
2 )Et+1 i
2 t+j
+(11
1 ) 1 t+1j
1 2
j=0 j=1 j=0

so the impulse-response function to it = 0, t < 0, and it = i, t = 0, 1, 2, ..announced at

time M , is,

t < (M + 1) : xt+1 = 0

t+1 (t+1+M )
(M + 1) t 1 : xt+1 = (1 1
2 )
2
+ (1 1
1 )1 M
1 2 1 2
(t+1)
!
1 2 (t+1+M )
0 t : xt+1 = (1 1
1 )
1
+ (1 1
2 ) + 1 M
1 2 1 1
1
1 2

The interest rate is then

rt = it Et t+1 .

For the impulse-response function, the expected and actual values are the same, except
142 COCHRANE

at M , where though M 6= 0, EM 1 M = 0. Hence,

t (M + 1) : rt = 0 (77)
t
 
(t+1+M )
2
M t < 0 : rt = 1 M (78)
1 2 1 2
 t 
2 (t+1+M )
t = 0 : rt = i 1 M (79)
1 2 1 2
1 t 

1 1 (1 1 ) (t+1+M )
0 < t : rt = i + 1 1 M (80)
1 2 1 2 1 1

19.6 Impulse-response with long-term debt and price stickiness

I develop the exact nonlinear formulas for the value of surpluses and a linear approxi-
mation. The linear approximation turns out to be quite accurate in this application.

An interest rate rise from time t = 0 onwards is announced at time t = M . I

calculate for each value of the inflation shock M the percent change in a constant
surplus corresponding to that shock. Writing the surplus as ses , the value of nominal
debt before the shock satisfies

j1
!
X Y 1 B (j) X 1
= tM s. = s.
j=0 k=0
1 + f (k) PM
t=M
1

where B (j) is the observed maturity structure of the debt, and the observed forward


(j) (0)
rates are f (j) , (ft is the forward rate at time t for loans from t + j to t + j + 1; ft = it is
the one-period interest rate). After the shock, nominal interest rates increase by i, the
, with
price level jumps from PM to PM

PM

M M +M
e =e = .
PM

Here, M denotes the solution with = 0, so actual inflation after the shock is an-

nounced is M = M + M . The basic solution for inflation (70) includes a jump in
inflation when the shock is announced, and I have defined as additional unexpected
 INFLATION AND INTEREST RATES 143

changes in inflation. Surpluses rise to ses , giving

M 1 j1
! !
1 1 B (j) 0
(tM ) u (Ct )
X Y Y X
= ses (81)
1 + f (k) 1 + f (k) + i PM u0 (CM )
j=0 k=0 k=M t=M

To easily calculate the multiperiod discount factor on the right hand side, I use

u0 (Ct ) e(c+xt ) 1

0
= (c+x )
= e (xt x )
u (C ) e

Dividing pre and post shock values of (81), s cancels and

P QM 1 1
 Q
j1 1

j=0 k=0 1+f (k) k=M 1+f (k) +i B (j) P
(tM )
t=M
e M +M
= P Qj1  P 1 es .
j=0
1
k=0 1+f (k) B (j) t=M (tM ) e (xt xM )

Conversely, then, we can find the surplus required to support a given time -M shock
M whether that surplus comes from active or from passive fiscal policy by solving
for s,

P QM 1 1
 Q
j1 1

j=0 k=0 1+f (k) k=M 1+f (k) +i B (j) P
(tM )
t=M
e s
= Q  P 1 e(M +M ) .
(tM ) e (xt xM )
P j1 1
j=0 k=0 1+f (k) B (j) t=M
(82)

For each choice of M , then, I find the solution for inflation and interest rates by
(78)-(80); I compute the product of real rates in the bottom right term of (82), and I
compute the required percentage change in surplus s. To find the fiscal-theory / long-
term debt solution, I search for the M that produces s = 0. It is important to treat
the numerator and denominator of the last term of (82) equally. If one truncates the
denominator, truncate the numerator at the same point.
144 COCHRANE

19.7 Linearized valuation equation

To linearly approximate (82), write

P
t=M (tM )
e s
V P 1 e(M +M ) . (83)
(tM ) e (xt xM )
t=M

P (tM )
t=M
1 + s (1 + v) P (tM ) 1 1 (x

(1 (M + M )). (84)
t=M t xM )

1
1 + s (1 + v) P (tM ) 1 (x x
(1 (M + M )). (85)
t=M t M )
1 P


(tM )
t=M

P
t=M (tM ) 1 (xt xM )
s v + P (tM )
(M + M ). (86)
t=M

X 1
s v + (1 ) (tM ) (xt xM ) (M + M ). (87)

t=M

In numerical experimentation, it turns out that the exact and linearized approach
produce almost exactly the same answer to the first few decimals. So, the nonlinearity
of long-term present values is not an issue for this magnitude a few percent at most
of interest rate variation.

19.8 The Model with Money

This section derives the model with money (33). The utility function is
Z
max E et u(ct , Mt /Pt )dt.
t=0

The present-value budget constraint is

Z  
B 0 + M0 Rt Mt
= e s=0 rs ds ct yt + st + (it im
t ) dt
P0 t=0 Pt
 INFLATION AND INTEREST RATES 145

where
dPt
rt = it
Pt
and s denotes real net taxes paid, and thus the real government primary surplus. This
budget constraint is the present value form of

d(Bt + Mt ) = it Bt + im
t Mt + Pt (yt ct st ).

Introducing a multiplier on the present value budget constraint, we have

Rt
: et uc (t) = e s=0 rs ds .
ct

where (t) means (ct , Mt /Pt ). Differentiating with respect to time,

dct dmt Rt
et uc (t) + et ucc (t) + et ucm (t) = rt e s=0 rs ds
dt dt

where mt Mt /Pt . Dividing by et uc (t), we obtain the intertemporal first order con-
dition:
ct ucc (t) dct mt ucm (t) dmt
= (rt ) dt. (88)
uc (t) ct uc (t) mt

The first-order condition with respect to M is

1 Rt 1
: et um (t) = e s=0 rs ds (it im
t )
Mt Pt Pt
et um (t) = et uc (t) (it im
t )
um (t)
= it im
t . (89)
uc (t)

The last equation is the usual money demand curve.

Thus, an equilibrium ct = yt satisfies

 
ct ucc (t) dct mt ucm (t) dmt dPt
= dt + it dt (90)
uc (t) ct uc (t) mt Pt
um (t)
= it im
t (91)
uc (t)
Z R  
B0 + M 0 t
s=0 rs ds m Mt
= e st + (it it ) dt (92)
P0 t=0 Pt
146 COCHRANE

The last equation combines the consumers budget constraint and equilibrium c = y. I
call it the government debt valuation formula.

19.8.1 CES functional form

I use a standard money in the utility function specification with a CES functional form,

1 h 1 i 1
1
u(ct , mt ) = ct + mt1 .
1

I use the notation m = M/P , with capital letters for nominal and lowercase letters for
real quantities.

This CES functional form nests three important special cases. Perfect substitutes is
the case = 0 :
1
u(ct , mt ) = [ct + mt ]1 .
1
The Cobb-Douglas case is 1:
 1
1
1 1+ 1+
u(ct , mt ) c mt . (93)
1 t

The monetarist limit is :

1
u(ct , mt ) [min (ct , mt )]1 .
1

I call it the monetarist limit because money demand is then Mt /Pt = ct /, i.e. = 1/V
is constant, and the interest elasticity is zero. The separable case is = :

1 h 1 i
u(ct , mt ) = ct + m1
t .
1

In the separable case, uc is independent of m, so money has no effect on the intertem-

poral substitution relation, and hence on inflation and output dynamics in a new-
Keynesian model under an interest rate target. Terms in ( ) or ( ) with =
1/ and = 1/ will characterize deviations from the separable case, how much the
marginal utility of consumption is affected by money.
 INFLATION AND INTEREST RATES 147

With this functional form, the derivatives are

h i
c
1
uc = ct1 + mt1 t

h i
m
1
um = ct1 + mt1 t .

Equilibrium condition (91) becomes

 
um (t) mt
= = it im
t . (94)
uc (t) ct

The second derivative with respect to consumption is

ucc 1
= ( ) h i c c1
uc 1 1 t t
c + m t t

h i
cucc ( ) c1
t c1
t + m1
t
= h i
uc c1 + m1
t t

cucc ct1 + mt1

= 1
uc ct + mt1
  1 
1 + m

ct
t
cucc
=   1  .
uc
1+ m ct
t

The cross derivative is

mucm m1
= ( ) 1 t 1
uc ct + mt
 1
m
ct
t

= ( )  1 .
1+ m ct
t

or, using (94)  

mt
mucm (it im
t ) ct
= ( )   .
uc 1 + mt (i im ) ct t t
148 COCHRANE

19.8.2 Money demand

Money demand (94) can be written

 
mt 1
= (it im
t ) . (95)
ct

where = 1/ becomes the interest elasticity of money demand, in log form, and
governs the overall level of money demand.

The steady state obeys

 
m 1
= (i im ) . (96)
c
so we can write money demand (95) in terms of steady state real money as

it im

mt  m  t
= , (97)
ct c i im

avoiding the parameter . (Throughout, numbers without time subscripts denote steady
state values.)
m
The product c (i im ) , the interest cost of holding money, appears in many subse-
quent expressions. It is
 
m 1
(i im ) = (i im )1 .
c

With < 1, as interest rates go to zero this interest cost goes to zero as well.

19.8.3 Intertemporal Substitution

The first order condition for the intertemporal allocation of consumption (90) is

ct ucc (t) dct mt ucm (t) dmt

= dt + (it t ) dt
uc (t) ct uc (t) mt

where t = dPt /Pt is inflation. This equation shows us how, with nonseparable util-
ity, monetary policy can distort the allocation of consumption over time, in a way not
captured by the usual interest rate effect. That is the central goal here. In the case of
 INFLATION AND INTEREST RATES 149

complements, ucm > 0 (more money raises the marginal utility of consumption), larger
money growth makes it easier to consume in the future relative to the present, and acts
like a higher interest rate, inducing higher consumption growth.

Substituting in the CES derivatives,

 1  1
1 + m ct
t
dct mt
ct dmt
 1 ( )  1 = dt + (it t ) dt
ct mt
1+ m ct
t
1 + mt
ct

and using (94) to eliminate

   
mt mt
1+ ct (it im
t ) dc (it im
t ) dmt ct
t
  ( )   = dt + (it t ) dt (98)
1+ m t
(i im ) ct 1 + mt
(i i m ) mt
ct t t ct t t

We can make this expression prettier as

 
mt
dct (it im
t )

ctdct dmt

+ ( )   = dt + (it t ) dt
ct 1 + mt (i im ) ct mt
ct t t

Rexpressing in terms of the intertemporal substitution elasticity = 1/ and interest

elasticity of money demand = 1/, and multiplying by ,
 
mt
dct



ct(it im
t )

dct dmt

+   = dt + (it t ) dt. (99)
ct 1+ m t
(i im) ct mt
ct t t

We want to substitute interest rates for money. To that end, differentiate the money
demand curve

mt  m  it im

t
=
ct c i im

mt dmt dct
  m   i im  d (i im )
t t t t
=
ct mt ct c i im it im
t
  m  m
 m
dct dmt it it d (it it )
= mct
ct mt ct i im it im
t
150 COCHRANE

Substituting,
 
mt m
ct (it it )
 !
m
it im d (it im
  
dct c t t )
+   mt m = dt+ (it t ) dt.
ct 1 + mt (i im ) ct i im it it
ct t t


it im

dct m 1 t
+ ( )   d (it im
t ) = dt + (it t ) dt.
ct c 1 + mt (i im ) i im
ct t t

With xt = log ct , dxt = dct /ct m, approximating around a steady state, and approximat-
ing that the interest cost of holding money is small, m m


c (i i ) << 1, we obtain the
intertemporal substitution condition modified by interest costs,

dxt m d (it im
t )
+ ( ) = (it t ) . (100)
dt c dt

In discrete time,

m 
Et it+1 im m


Et xt+1 xt + ( ) t+1 (it it ) = (it Et t+1 ) .
c

For models with monetary control, one wants an IS curve expressed in terms of the
monetary aggregate. From (99), with the same approximations and m
= log(m),
   
dxt m dxt dm t
+ (i im ) = (it t ) dt. (101)
dt c dt dt

In discrete time,
 
m
(Et xt+1 xt ) + (i im ) [(Et xt+1 xt ) Et (m
t+1 m
t )] = (it t ) .
c
(102)

19.9 Three-equation model solution

I solve the three-equation model of Figure 25 by standard methods, incorporating the

Taylor rule in to monetary policy rather than conditioning on the equilibrium interest
rate and then constructing the underlying Taylor rule. Both methods give the same
answer, but a conventional calculation is more transparent in this case, and it verifies
 INFLATION AND INTEREST RATES 151

that both approaches give the same answer. This is a much condensed version of the
treatment in Cochrane (2016a).

While one can solve the model quickly via matrix techniques, here I use lag operator
techniques to write the solution for inflation analytically.

The model is

xt = Et xt+1 (it Et t+1 )

t = Et t+1 + xt

it = t + vti

vti = vt1
i
+ it

Substituting the Taylor rule,

xt = Et xt+1 (t + vti Et t+1 )

t = f Et t+1 + xt

Expressing the model in lag operator notation,

Et (1 L1 )xt = Et L1 t vti

Et (1 f L1 )t = xt

Forward-differencing the second equation,

Et (1 M L1 )(1 f L1 )t = Et (1 M L1 )xt

Then substituting into the first equation,

 
Et (1 M L1 ) 1 f L1 t = Et L1 t vti

M + f + 1 M f
 
2
Et 1 L + L t = vi.
1 + 1 + 1 + t
152 COCHRANE

Factor the lag polynomial

Et (1 1 L1 )(1 2 L1 )t = vi
1 + t

where q
2
M + f + (M + f + ) 4M f (1 + )
=
2 (1 + )
These lag operator roots are the inverse of the eigenvalues of the usual transition ma-
trix. The system is stable and solved backward for > 1; it is unstable and solved
forward for < 1.

The standard three-equation model uses > 1 so both roots are unstable, 1 < 1
and 2 < 1. Then, we can write

Et (1 1 L1 )(1 2 L1 )t = vi
1 + t

1
t = Et vi
(1 1 L1 )(1 2 L1 ) 1 + t
 
1 1 2
t = Et + vi
1 2 1 1 L1 1 2 L1 1 + t


1
j1 vt+j j2 vt+j
X X
i i
t = Et 1 + 2
1 + 1 2
j=0 j=0

Using the AR(1) form of the disturbance v i ,

1
j1 j + 2 j2 j t
X X
t = 1
1 + 1 2
j=0 j=0

 
1 1 2
t = + vi
1 + 1 2 1 1 1 2 t
 
1 2 (1 1 ) 1 (1 2 )
t = vti
1 + 1 2 (1 1 ) (1 2 )
 
1
t = vi
1 + (1 1 ) (1 2 ) t

Thus, to produce Figure 25, I simply simulate the AR(1) impulse-response, for {vti },
 INFLATION AND INTEREST RATES 153

calculate t by the last equation, and calculate it = t + vti .