Journal of Financial Economics 113 (2014) 270–299

Contents lists available at ScienceDirect

Journal of Financial Economics

journal homepage: www.elsevier.com/locate/jfec

Mortgage convexity$
Samuel G. Hanson
Harvard Business School, United States

a r t i c l e in f o abstract

Article history: Most home mortgages in the United States are fixed-rate loans with an embedded
Received 27 February 2013 prepayment option. When long-term rates decline, the effective duration of mortgage-
Received in revised form backed securities (MBS) falls due to heightened refinancing expectations. I show that
4 November 2013
these changes in MBS duration function as large-scale shocks to the quantity of interest
Accepted 18 November 2013
Available online 9 May 2014
rate risk that must be borne by professional bond investors. I develop a simple model in
which the risk tolerance of bond investors is limited in the short run, so these fluctuations
JEL classification: in MBS duration generate significant variation in bond risk premia. Specifically, bond risk
G10 premia are high when aggregate MBS duration is high. The model offers an explanation
G12
for why long-term rates could appear to be excessively sensitive to movements in short
rates and explains how changes in MBS duration act as a positive-feedback mechanism
Keywords:
that amplifies interest rate volatility. I find strong support for these predictions in the time
Bond return predictability
Duration series of US government bond returns.
Mortgage-backed securities & 2014 Elsevier B.V. All rights reserved.
Mortgage refinancing

1. Introduction yields declines. Conversely, as long-term interest rates rise,

the prepayment option moves out-of-the-money and the
A distinguishing feature of bond markets in the United effective duration of MBS rises. To quote a Wall Street
States is the prominent role of fixed-rated home mort- adage, a mortgage-backed security “goes up like a two-
gages. Most fixed-rate mortgages issued in the US grant year bond” when rates fall and “goes down like a six-year
the borrower the right to prepay the loan at any time bond” when rates rise. More formally, the negative rela-
without penalty. When long-term interest rates decline, tionship between price and yield is convex for a noncall-
the option to prepay the mortgage and refinance at the able bond. By contrast, the price-yield relation for callable
current interest rate moves toward the money. Because bonds such as MBS is typically concave, a feature that bond
lenders expect to be repaid sooner, the effective duration investors call “negative convexity.”
of outstanding mortgage-backed securities (MBS) falls; Aggregate mortgage refinancing activity varies signifi-
i.e., the sensitivity of MBS prices to changes in long-term cantly over time. Because refinancing entails large finan-
cial and nonfinancial costs, households typically follow a
trigger rule, refinancing only when their option is suffi-
☆
I am grateful to the referee, Jonathan H. Wright, for numerous ciently far in-the-money. Thus, depending on the past path
suggestions that significantly improved the paper. I thank seminar
participants at Boston University, the Federal Reserve Board, and the
of mortgage rates, there are times when many households
Copenhagen Business School FRIC'13 Conference on Financial Frictions, as move from being far from refinancing to being close to
well as John Campbell, Anna Cieslak, Robin Greenwood, Arvind Krishna- refinancing and vice versa. I argue that the resulting
murthy, Lasse Pedersen, Erik Stafford, Jeremy Stein, Larry Summers, changes in aggregate MBS duration act as massive supply
Adi Sunderam, and Dimitri Vayanos for helpful comments. I gratefully
shocks to the total quantity of interest rate risk that must
acknowledge funding from the Harvard Business School Division of
Research. be borne by investors in the broader bond market. Because
E-mail address: shanson@hbs.edu these shifts in bond market duration are large relative to

http://dx.doi.org/10.1016/j.jfineco.2014.05.002
0304-405X/& 2014 Elsevier B.V. All rights reserved.
 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299 271

bond investors' risk tolerance, they have a significant movements in short rates (Gürkaynak, Sack, and Swanson,
impact on equilibrium term premium; i.e., on the expected 2005) and, more generally, to the excess volatility of long-
return on long-term default-free bonds over short-term term yields (Shiller, 1979). Excess sensitivity reflects what
default-free bonds. In summary, time variation in aggre- might be called an “MBS duration spiral.” An initial shock
gate mortgage refinancing has a significant effect on the to the short rate directly raises long yields due to the
pricing of interest rate risk throughout the US fixed expectations hypothesis. The rise in long yields raises the
income market and, in particular, it has a large impact on duration of MBS. The term premium must rise to induce
US Treasury yields. risk-averse bond investors to bear the larger aggregate
Why do shifts in expected household mortgage refi- quantity of interest rate risk. The resulting rise in yields
nancing affect the aggregate amount of interest rate risk further raises MBS duration, which further raises the term
that bond market investors must bear? First, household premium, and so on. Thus, the fact that MBS duration is
borrowers only gradually exercise their prepayment increasing in long yields gives rise to a positive-feedback
options following a decline in prevailing mortgage rates. channel that generates excess sensitivity and excess vola-
Second, household borrowers do not alter their bond tility. The model also suggests that this positive-feedback
holdings to hedge their time-varying interest rate expo- mechanism is strongest when the mortgage market is on a
sure. That is, households do not adjust their asset portfo- refinancing cliff; i.e., when a small movement in long rates
lios to offset the time-varying interest rate risk they are significantly impacts refinancing behavior and, hence, MBS
bearing on the liability side. The gradual response of duration. The model therefore predicts that excess sensi-
mortgage prepayments to changes in mortgage rates and tivity and volatility should be most pronounced when the
the lack of household hedging means that households at MBS market is more negatively convex.
times are bearing more or less interest rate risk. Conver- I also show that an MBS duration shock should have a
sely, bond investors at times are bearing less or more risk. larger effect on the expected excess returns of long-term
Thus, shifting refinancing expectations generate a form of bonds than on those of intermediate-term bonds. This is a
aggregate market congestion: sometimes most households natural consequence of the fact that an MBS duration
are effectively borrowing long term, and other times they shock raises the current duration risk premium in bond
are borrowing short term. If the risk-bearing capacity of markets. However, because these shocks are transient,
professional bond investors is limited in the short run, shocks to MBS duration have a humped-shaped effect on
then the term premium must adjust to induce investors to the yield curve and the forward rate curve. The effects of a
bear these risks. supply shock on yields, which equals the effect on the
The sheer size of the MBS market within the US bond bond's average returns over its lifetime, is greater for
market plays a crucial role in this story. Bond markets have intermediate-term bonds than for long-term bonds when
witnessed several MBS duration supply shocks that are the supply shock is expected to be short-lived. That is, a
larger than the shift in duration induced by the Federal shock to MBS duration increases the curvature of the yield
Reserve's Quantitative Easing (QE) policies from 2008 curve. This suggests that transient shocks to MBS duration
to 2012. For instance, on several occasions the quarterly could account for some of the predictive power of the
change in MBS duration was equivalent to a $1 trillion Cochrane and Piazzesi (2005) factor, which picks up time
increase (in 2012 dollars) in the supply of ten-year variation in the curvature of the yield curve and is useful
Treasury notes, with a corresponding reduction in the for forecasting transitory variation in bond returns.
supply of short-term T-bills. By way of comparison, the I find strong support for these predictions in US interest
quarterly new-issue supply of ten-year Treasury notes in rate data between 1989 and 2012. Measures of MBS duration
2012 was roughly $65 billion. Thus, past shifts in MBS are strong predictors of excess government bond returns.
duration have arguably been very large relative to the risk And they appear to contain information that is not reflected
tolerance of bond market arbitrageurs. in traditional forecasting variables based on the current
I develop a simple model of this mortgage convexity shape of the yield curve. My analysis of the time signature
mechanism. The model has the following key ingredients. of these effects indicates that shocks to MBS duration have a
First, expectations of future household mortgage refinancing transitory impact on term premia, which largely dissipates
vary over time, which induces shifts in the effective duration over the next 6 to 12 months. As predicted, shocks to MBS
of household mortgage borrowing. Second, aggregate bond duration have a hump-shaped effect on the yield and
risks are priced by risk-averse, specialized bond investors, as forward rate curves. I also find that the excess sensitivity of
in Vayanos and Vila (2009). This ensures that demand curves long rates to short rates is more pronounced when the MBS
for aggregate bond risks slope downward, so the term market is on a refinancing cliff, i.e., when a move in long
premium must adjust to induce investors to absorb the rates has a larger impact on aggregate MBS duration. Lastly, I
aggregate supply of bond duration. Specifically, the model find that option-implied volatility of long yields is higher
predicts that measures of aggregate bond market duration, when the MBS market is more negatively convex.
which derive most of their power from variation in MBS I lack a good instrument for MBS duration that would
duration, should positively predict bond excess returns. allow me to cleanly identify the demand curve for interest
Furthermore, because shocks to MBS duration are fairly rate risk. As a result, I rely on predictive regressions that
transient, one would expect fluctuations in MBS duration provide indirect evidence consistent with the view that
to lead to high frequency variation in bond risk premia. shifts in MBS duration trace out a downward-sloping
The model shows how negative MBS convexity could demand for interest rate risk. These return forecasting
contribute to the excess sensitivity of long-term yields to regressions are analogous to a regression of prices (term
272 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299

premia) on quantities (duration supply). Thus, I take the Federal Reserve's Quantitative Easing policies, in which it
considerable care to address several natural concerns is typically argued that Fed purchases of long-term assets
raised by this indirect approach. reduce bond risk premia. Second, a vast literature is devoted
To begin, I control for an exhaustive set of factors that are to forecasting the excess returns on long-term bonds. Impor-
thought to impact term premia to address concerns that my tant contributions here include Fama and Bliss (1987),
forecasting results are driven by an omitted variable. I then Campbell and Shiller (1991), and Cochrane and Piazzesi
provide a host of further indirect evidence that is consistent (2005). Third, a variety of papers find excess sensitivity,
with the MBS story, but that would not be predicted by including Cochrane and Piazzesi (2002), Gürkaynak, Sack,
alternative explanations for my findings. First, I show that the and Swanson (2005), and Hanson and Stein (2012). Fourth,
return forecasting power of MBS duration has grown signifi- a number of papers argue that long-term yields are exces-
cantly over time as MBS markets have grown relative to the sively volatile, including Shiller (1979), Perli and Sack (2003),
rest of the US bond market. Second, I show that shifts in MBS and Duarte (2008). Finally, a growing literature, including
duration have far more forecasting power for US bond returns Gabaix, Krishnamurthy, and Vigneron (2007), Greenwood and
than for foreign bond returns. Third, I show that shifts in MBS Vayanos (2014), and Vayanos and Vila (2009), explores the
duration impact the spreads between corporate bonds, inter- implications of limited arbitrage in fixed income markets.
est rate swaps, and Treasuries in precisely the way that one The plan for the paper is as follows. Section 2 develops
would expect if MBS investors were using Treasuries and a simple model that generates several novel predictions.
swaps (but not corporate bonds) to hedge variation in MBS In Section 3, I find strong support for these predictions in
duration. Collectively, these additional findings provide further the time series of US government bond returns. Section 4
support for my story emphasizing shifts in MBS duration. concludes.
Bond investors often invoke shifts in MBS duration and the
portfolio-hedging flows they trigger when explaining large 2. Theoretical framework
movements in long-term interest rates. Many MBS investors
delta-hedge the time-varying duration of MBS. If there is a This section clarifies why shifts in MBS duration matter
drop in long rates that raises mortgage refinancing expecta- for bond pricing and develops a stylized model of the
tions and lowers MBS duration, these investors buy more mortgage convexity mechanism.
long-term Treasuries, financed by selling some short-term
T-bills, in order to maintain their prior asset duration. Given 2.1. Shocks to MBS duration
the prominence that these dynamics receive in practitioner
commentary, their relative absence from the literature on the Suppose a large number of households announce that
term structure of interest rates is somewhat surprising. The in six months time they will repay their existing mortgages
handful of exceptions includes Fernald, Keane, and Mosser at par and will take out new mortgages at the prevailing
(1994), Kambhu and Mosser (2001), Perli and Sack (2003), mortgage rate. And suppose for simplicity that this commit-
Duarte (2008), and Malkhozov, Mueller, Vedolin, and Venter ment to refinance in six months is binding. Suddenly, from the
(2013). Kambhu and Mosser (2001), Perli and Sack (2003), and perspective of investors who own the existing mortgages, the
Duarte (2008) each argue that these hedging flows impact mortgages behave just like short-term bonds, so mortgage
interest rate volatility, whereas the primary focus of my paper holders are bearing less interest rate risk. And over the
is to investigate the impact of MBS duration on the equili- following six months, households are bearing significantly
brium term structure of yields and term premia.1 In con- more interest rate risk. That is, they will be worse off if rates
temporaneous work, Malkhozov, Mueller, Vedolin, and Venter rise and better off if rates fall. To hedge this interest rate
(2013) also explore the impact of MBS duration on excess exposure, households could initially sell some long-term
returns. Their study places more emphasis on the implications Treasuries with the intention of buying them back in six
of MBS hedging for interest rate volatility, whereas I am months.2 However, due to a variety of frictions, costs, or a lack
primarily interested in understanding the pricing implications of financial sophistication, households do not hedge their
of these hedging flows. time-varying exposure to interest rate risk. So, on net, a
The ideas in this paper connect to several broader strands temporary reduction occurs in the amount of interest rate
of prior research. First, the idea that supply and demand risk that fixed income investors must bear in equilibrium. And
effects can have important consequences in bond markets is if the risk-bearing capacity of bond investors is limited, the
central to a number of recent papers, including Vayanos expected excess return on long-term bonds must fall to
and Vila (2009), Greenwood and Vayanos (2010, 2014), equate the supply and demand for bonds.
Krishnamurthy and Vissing-Jorgensen (2011, 2012), and In practice, households do not enter into binding agree-
Gagnon, Raskin, Remache, and Sack (2011). Important pre- ments to refinance. Instead, household refinancing behavior is
cursors to this recent work include Tobin (1958) and gradual, so shocks to long-term interest rates alter expected
Modigliani and Sutch (1966). This literature has featured mortgage refinancing. Critically, I assume households do not
prominently in the portfolio balance effect interpretation of hedge their interest rate exposure; i.e., households do not alter
their bond holdings to hedge the time-varying interest rate
1
Perli and Sack (2003) show that MBS hedging impacts swaption-
2
implied yield volatility. Duarte (2008) also finds that MBS hedging The household would invest the proceeds in short-term bills for six
forecasts high future realized volatility, which suggests that hedging months. If households were to hedge in this way, their desire to short
flows not only impact interest rate option prices, but also impact the long-term bonds would perfectly offset the temporary decline in mort-
underlying bond yields. gage duration.
 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299 273

risk they are assuming on the liability side of their balance emphasized here should be distinguished from MBS model
sheets. This plausible assumption is the key friction that risk as analyzed by Gabaix, Krishnamurthy, and Vigneron
generates time-varying risk sharing between household bor- (2007).
rowers and bond investors and, thus, shifts in aggregate bond
market duration. 2.2. Stylized model with short- and long-term bonds
For instance, when future mortgage refinancings are
expected to be high, households are effectively borrowing The main predictions can be illustrated using a simple
shorter term and, hence, are bearing more interest rate risk: dynamic model that features only short- and long-term
households are exposed to the risk that they will refinance at bonds at each date t. The expected excess return on long-
a less advantageous rate. Conversely, when expected refinan- term bonds over short-term bonds is the only endogenous
cings are low, households are effectively borrowing longer variable at each date. The Internet Appendix extends the
term, so investors are bearing more interest rate risk. The total model to allow for multiple bond maturities.
quantity of interest rate risk borne in the economy remains A set of risk-averse arbitrageurs (e.g., fixed income hedge
unchanged. A single long-term asset, namely, housing, is funds, fixed income dealers, etc.) price interest rate risk in
going to be financed long term at some interest rate. What bond markets. Assume these investors extend fixed-rate,
varies is the amount of interest rate risk that is borne by bond prepayable mortgages to households and also own some
investors as opposed to household borrowers. Shifts in the noncallable bonds (e.g., Treasuries): the same arbitrageurs
extent of aggregate risk sharing between households and are marginal buyers in both MBS and Treasury markets.
investors move the price of interest rate risk—the duration The key idea is that these arbitrageurs are specialists who are
risk premium—in bond markets. heavily exposed to risks specific to the bond market. The
If the total quantity of interest rate risk in the economy is limited risk-bearing capacity of these investors ensures that,
fixed, why do shifts in MBS duration move bond prices? One at least in the short term, demand curves for aggregate bond
natural answer, as in Gabaix, Krishnamurthy, and Vigneron risk factors slope downward. In particular, the required
(2007), is that, while this could be a wash from the perspec- return on long-term bonds must rise relative to short-term
tive of some representative household (e.g., when about to bonds to induce investors to bear more interest rate risk.
refinance, the representative household takes more interest I assume that, at each date t, arbitrageurs have mean–
rate risk on the liability side but is exposed to less interest rate variance preferences over wealth at t þ 1, with possibly time-
risk on bond holdings), the change in the quantity of bond risk varying risk aversion, γt. This is the discrete-time analog of
still looms large from the standpoint of a delegated invest- Vayanos and Vila (2009), who work in continuous time. The
ment manager who specializes in bond markets. If this key idea is that arbitrageurs are concerned with their interim
manager is forced to invest a large fraction of her own wealth wealth and, hence, with interest rate risk.They risk having to
in the bond fund, e.g., to mitigate agency problems, then the sell long-term bonds at a capital loss if interest rates rise
manager perceives an increase in the total quantity of risk. tomorrow.
Thus, risks that are idiosyncratic from an aggregate perspec- Let r t þ 1 denote the log return on one-period riskless
tive could be priced because of the large undiversified bonds between t and t þ 1. By definition r t þ 1 ¼ ytð1Þ , i.e., the
exposures of specialized intermediaries who are the marginal yield on one-period bonds at time t. I assume an exogen-
buyers of bonds.3 ous random process for the short rate. One can think of
Distinguishing between the interest rate risk of MBS and this as being pinned down by monetary policy or by a
MBS model risk is crucial. Interest rate risk derives from the stochastic short-term storage technology that is in per-
fact that the value of MBS depends on the level of rates, fectly elastic supply. Thus, from the standpoint of arbitra-
assuming that household prepayment behavior is a known geurs at time t, the only relevant uncertainty is about
deterministic function of rates. By contrast, model risk derives r t þ 2 ¼ ytð1Þ
þ 1 . The log return on two-period bonds from t to
from the fact that household prepayment behavior is not a t þ 1 is r ð2Þ ð2Þ ð1Þ ð2Þ
t þ 1 ¼ 2yt  yt þ 1 ¼ 2yt  r t þ 2 , so the excess
known function of interest rates. Instead, it is sometimes return on two-period bonds relative to one-period bonds
difficult to predict how prepayments will respond to changes is rxð2Þ ð2Þ ð2Þ
t þ 1 ¼ r t þ 1  r t þ 1 ¼ 2yt  r t þ 1  r t þ 2 . Arbitrageurs
in rates. This paper focuses on the time-varying interest rate choose their long-term bond holdings bt to solve
risk of the aggregate MBS market. The key insights are that
maxfbt  Et ½rxð2Þ ð2Þ
2
household prepayment behavior is gradual and that house- t þ 1   bt  ðγ t =2ÞVar t ½rxt þ 1 g
bt
holds do not hedge their time-varying interest rate exposure.
¼ maxfbt  ð2yð2Þ
2
Thus, the total quantity of interest rate risk borne by profes- t  r t þ 1  Et ½r t þ 2 Þ  bt  ðγ t =2ÞVar t ½r t þ 2 g:
bt
sional bond investors can vary over time even if prepayment
ð1Þ
behavior is a deterministic function of the path of rates.
As a result, the time-varying quantity of interest-rate risk I assume that r t þ 1 and γ t Var t ½r t þ 2  follow exogenous
stochastic processes. Thus,

3
bt ðyð2Þ
t Þ ¼ ðγ t Var t ½r t þ 2 Þ
1
 ð2ytð2Þ r t þ 1  Et ½r t þ 2 Þ: ð2Þ
A second answer is to think about an overlapping generations
setting where households borrow when young and lend when old. I model the MBS positive-feedback channel in a simple
Because the young are bearing less interest rate risk when refinancing way. Let Qt denote the total dollar quantity of duration risk
expectations are low, this means that the old are bearing more risk,
which impacts equilibrium term premia. In this telling, time-varying risk
that needs to be held by fixed income arbitrageurs. In practical
sharing between the old and young induces time variation in bond risk terms, I associate Qt with the total number of ten-year
premia. Treasury duration equivalents summed across all US fixed
274 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299

income markets. I allow Qt to depend on yð2Þ t . I assume that of expected returns because it contains an expectations
Q t ¼ Q t ðytð2Þ Þ with Q 0t ðytð2Þ Þ 40 to capture the fact that MBS hypothesis component, ðEt ½r t þ 2   r t þ 1 Þ=2. Thus, adding
duration is increasing in long-term yields. As discussed above, Q t ðyð2Þ
t Þ to a regression that contains the term spread can
Q 0t ðyð2Þ
t Þ 4 0 is a property that holds locally due to time raise the forecasting power, particularly if the time series
variation in risk sharing between household mortgage bor- variation in ðEt ½r t þ 2   r t þ 1 Þ=2 is high.
rowers and bond investors. And the vast size of the MBS More generally, the aggregate supply of duration contains
market relative to the broader US bond market ensures that valuable information about term premia that cannot be easily
shifts in MBS duration have a significant effect on aggregate recovered from the yield curve. For example, duration could
bond market duration. help to forecast excess bond returns, even controlling for the
The equilibrium long-term yield must clear the market term spread, the Cochrane and Piazzesi (2005) factor, or other
at each date, bt ðyð2Þ ð2Þ
t Þ ¼ Q t ðyt Þ, which delivers the fixed- simple yield-based proxies for term premia. Nonetheless, it is
point condition not my intention to argue that duration is an unspanned state
variable in the sense of Duffee (2011) or Joslin, Priebsch, and
yð2Þ ð2Þ
t ¼ ðr t þ 1 þEt ½r t þ 2 Þ=2 þ ðγ t =2ÞVar t ½r t þ 2 Q t ðyt Þ: ð3Þ
Singleton (2013). In classic term structure approaches, if the
To ensure that this equilibrium is locally stable, I assume true model is known, yields can be inverted to obtain the full
that 1 4 ðγ t =2ÞVar t ½r t þ 2 Q 0t ðytð2Þ Þ: This ensures that a small set of state variables. An unspanned state variable is a variable
exogenous shock to yð2Þ t leads to a larger increase in that is useful for forecasting returns but has (almost) no
investor demand for bond duration than in duration impact on current yields and cannot be recovered in this way.4
supply [i.e., ∂bt ðyð2Þ ð2Þ
t Þ=∂yt 4Q t ðyÞ].
0
Specifically, the Internet Appendix shows that I can nest all of
The expected excess return on long-term bonds relative my predictions in an affine term structure model similar to
to short-term bonds is Vayanos and Vila (2009) in which MBS duration is a spanned
state variable that impacts current yields. Thus, assuming a
E½rxð2Þ ð2Þ
t þ 1  ¼ γ t Var t ½r t þ 2 Q t ðyt Þ; ð4Þ stationary data generating process, all the information about
and the yield spread is future bond returns would be contained in current yields, and
duration would not add any further information.
yð2Þ ð1Þ ð2Þ
t  yt ¼ ðEt ½r t þ 2   r t þ 1 Þ=2 þðγ t =2ÞVar t ½r t þ 2 Q t ðyt Þ: ð5Þ In practice, however, the true model generating bond
Naturally, in the limiting cases in which arbitrageurs are yields is not known. Furthermore, it seems likely that
risk neutral ðγ t ¼ 0Þ or have to bear no interest rate risk the true model evolves over time due to changes in
ðVar t ½r t þ 2 Q t ðyð2Þ the macroeconomy, the conduct of monetary policy, the
t Þ ¼ 0Þ, the expectations hypothesis holds.
market structure, and the behavior of market participants.
For instance, the impact of MBS duration on term premia
2.3. Predictions
and bond market spreads has grown over time, arguably
because fluctuations in duration have grown larger relative
The model generates several novel predictions that I
to investor risk tolerance. Thus, in practice, it is natural
now develop in turn.
that duration is a useful summary statistic containing
additional information about future excess bond returns.
2.3.1. Forecasting excess bond returns
The model predicts that measures of aggregate
bond market duration, which derive most of their power 2.3.2. Excess sensitivity of long-term rates to short rates
from variation in MBS duration, should positively predict The negative convexity of the US fixed income market
bond excess returns. naturally generates excess sensitivity of long-term yields to
movements in short-term rates. To see this, consider the static
Proposition 1. Both the yield spread and aggregate bond case in which yð2Þ ¼ ðr 1 þ E½r 2 Þ=2 þ ðγ=2ÞVar½r 2 Q ðyð2Þ Þ and
market duration, Q t ðytð2Þ Þ, will positively forecast excess bond consider a change in r1, holding fixed all other parameters
returns. This remains true in a multivariate forecasting including E½r 2 . Because 1 4 ðγ=2Þ Var½r 2 Q 0 ðyð2Þ Þ 4 0,
regression so long as γ t Var t ½r t þ 2  varies over time.
∂yð2Þ 1 1 1=2 1
¼ ∑ ððγ=2ÞVar½r 2 Q 0 ðyð2Þ ÞÞi ¼ 4 :
Why isn't the slope of the yield curve a sufficient statistic ∂r 1 2i¼0 1  ðγ=2ÞVar½r 2 Q 0 ðyð2Þ Þ 2
for forecasting excess bond returns? In this simple model, the ð6Þ
term spread contains information about the expected path of
future short rates and term premia, both of which vary over This excess sensitivity stems from an MBS duration spiral.
time. Duration measures pertain solely to the latter and, thus, A small initial shock to the short rate of dr1 directly raises
can improve the forecasting power of regressions that include long-term yields by ð1=2Þ dr 1 due to the expectations hypoth-
only the term spread. Formally, so long as γ t Var t ½r t þ 2  varies esis. The rise in yields extends the duration of MBS, which
over time, both the yield spread and duration positively
predict excess returns in a multivariate forecasting regression. 4
As Duffee (2011) explains, this would arise if the evolution of short
According to Eq. (4), expected returns equal the product of the rates under the risk-neutral (pricing) measure is independent of some
time-varying quantity of risk, Q t ðytð2Þ Þ, and the time-varying variable, implying that it has no impact on current yields, even though
price per unit of risk, γ t Var t ½r t þ 2 . Aggregate duration pro- that variable is relevant for forecasting future short rates under the
objective measure. Such a situation could arise if a variable had offsetting
vides an accurate signal of the quantity of risk. Eq. (5) shows effects on the evolution of future short rates and future term premia, e.g.,
that the yield spread contains information about both the some scary, bad news that raises future expected term premia but lowers
quantity of risk and the price of risk, but it is a noisy indicator future expected short rates.
 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299 275

raises the term premium by ð1=2Þ½ðγ=2Þ Var½r 2 Q 0 ðyð2Þ Þ dr 1 ; uncertainty is about r t þ 2 þEt þ 1 ½r t þ 3 , so that
the resulting rise in yields further extends the duration of h i ∂yð2Þ 

MBS, which further raises the term premium by ð1=2Þ yð2Þ ð2Þ
t þ 1  Et yt þ 1   ð2Þ ð2Þ  ðEt þ 1 Et Þðr t þ 2 þr t þ 3 Þ:
∂r 1 y ¼ Et ½yt þ 1 
½ðγ=2ÞVar½r 2 Q 0 ðyð2Þ Þ2 dr 1 , and so on.5
ð8Þ
The same excess sensitivity point can be made in terms
of forward rates. Working with forward rates allows for Thus,
cleaner empirical tests because there is no direct expecta- h i
tions hypothesis term. The forward rate is f ¼
ð2Þ Var t ytð2Þ
þ1
ð2Þ
2yð2Þ r 1 ¼ E½r 2  þ γVar½r 2 Q ððr 1 þf Þ=2Þ. Consider a change !2
1=2   
in r1, holding fixed all other parameters including E½r 2 .6   Var t r t þ 2 þ Et þ 1 r t þ 3 :
Thus, 1  ðγ=2ÞVar½r 2 Q 0
ðEt ½ytð2Þ
þ 1 Þ

ð2Þ
ð9Þ
∂f ðγ=2ÞVar½r 2 Q 0 ðyð2Þ Þ
¼ 40: ð7Þ
∂r 1 1  ðγ=2ÞVar½r 2 Q 0 ðyð2Þ Þ
Proposition 4. All else equal, long-term interest rate volati-
lity, Var t ½yð2Þ
t þ 1 , is increasing in the negative convexity of the
Proposition 2. If aggregate bond market duration is increas- fixed income market, Q 0 ðEt ½yð2Þ t þ 1 Þ.
ing in yields ðQ 0 ðyð2Þ Þ 4 0Þ, long-term yields and forward rates Excess volatility of long-term rates is a natural corollary
will be excessively sensitive to movements in short-term of excess sensitivity. However, the negative convexity of
rates. Furthermore, both long-term real and nominal rates the MBS market acts as a positive-feedback loop that
will exhibit excess sensitivity. amplifies the effects of any primitive shock (e.g., short-
Thus, the model departs from Vayanos and Vila (2009), term rate, investor risk tolerance, or bond supply) that
who assume that Q 0 ðyð2Þ Þ o 0, meaning that arbitrageurs moves bond yields. Thus, the growth of the MBS market
must hold a lower net supply of long-term bonds when could have led to a secular increase in excess volatility.
long-term yields are high. The assumption that Q 0 ðyð2Þ Þ o 0 However, there is an additional time series prediction:
implies that distant forwards underreact to movements in Long rates should be particularly volatile at times when
short rates. However, the bulk of the empirical evidence the MBS market is most negatively convex. Perli and Sack
suggests that distant forwards overreact, not underreact, (2003) and Duarte (2008) develop a similar hypothesis and
to movements in short rates.7 find support for it in recent US data.
The model also generates a novel prediction about how
the excess sensitivity of long-term forwards varies over 2.3.4. Allowing for multiple bond maturities
time. In the Internet Appendix, I extend the model with only
short- and long-term bonds to allow for multiple bond
Proposition 3. The excess sensitivity of long-term rates to maturities. The extension is a discrete-time version of the
short rates is more pronounced when the fixed income no-arbitrage term structure model developed by Vayanos
market is more negatively convex ð∂2 yð2Þ =∂r 1 ∂Q 0 ðyð2Þ Þ 4 0Þ. and Vila (2009) and Greenwood and Vayanos (2014). I add
the MBS convexity effect to this model in a simple fashion. I
2.3.3. Excess volatility of long-term interest rates assume that the aggregate supply of duration that must be
The model also has implications for the volatility held by arbitrageurs rises when interest rates rise. However, I
of long-term yields. Suppose it is time t and consider assume that these duration shocks dissipate quickly: dura-
the long-term yield at time t þ 1: yð2Þ t þ 1 ¼ ðr t þ 2 þEt þ 1 tion shocks are transitory and are expected to mean-revert
½r t þ 3 Þ=2 þ ðγ=2ÞVar½r t þ 3 Q ðyð2Þ
tþ1 Þ. Assume the only over time. These assumptions are a simple way to capture
the dynamics of MBS duration explained in the paper.
5 Solving the model, I obtain a discrete-time affine model
Changes in expected future short rates also lower risk premia in the
model: ∂yð2Þ =∂E½r 2  ¼ ∂yð2Þ =∂r 1 4 1=2. The MBS convexity effect suggests of the term structure with two state variables, the current
that Federal Open Market Committee forward guidance could impact short rate and the current level of MBS duration, which
term premia. Intuitively, the amplification mechanism is the same for an depends on the past path of interest rates. The model
independent change in either r 1 or E½r 2 . generates three predictions.
6
By considering sufficiently distant forwards, it is reasonable to
assume that expected short rates in the distant future are insensitive to
current short rates, i.e., that ∂E½r 2 =∂r1  0. Even if ∂E½r2 =∂r 1 a 0, then 1. An MBS duration shock raises the current expected
ð2Þ
∂f =∂r1 4∂E½r 2 =∂r 1 , so the excess sensitivity result still carries through. excess returns on long-term bonds over short-term
7
Vayanos and Vila (2009) assume that Q 0 ðyð2Þ Þo 0 because it enables bonds. Furthermore, because MBS duration is expected
their baseline model—featuring an stochastic short rate, no independent
to quickly revert to its long-run mean, this effect is
supply shocks, and constant arbitrageur risk tolerance—to explain why
the term spread forecasts excess bond returns. The intuition is that a rise short-lived in expectation.
in short rates, which directly raises long yields and lowers the term 2. A stronger MBS convexity effect (i.e., a more negatively
spread due to the expectations hypothesis, reduces the amount of long convex MBS universe) increases the sensitivity of long-
bonds that arbitrageurs must hold and, thus, the term premium. To term yields and forward rates to movements in
jointly explain the forecasting power of the term spread and excess
sensitivity, one could entertain a more complicated model in which
short rates.
Q 0t ðyð2Þ
t Þ 4 0 and, in addition to short rate shocks, there are independent
3. A stronger MBS convexity effect increases the volatility
shocks to the supply of duration or arbitrageur risk tolerance. of long-term yields and forwards.
276 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299

These three results echo those from the simple model. short-term bonds. As a result, banks significantly reduce
However, the multiple maturity extension delivers two their delta hedges. That is, the banks buy more long-term
additional predictions that I can take to the data. Treasuries, financed by selling short-term T-bills, to main-
tain their previous asset duration. When borrowers refi-
4. An MBS duration shock has a larger effect on the nance, banks unwind these transactions. Thus, banks have
expected excess returns on longer-term bonds than a temporarily elevated hedging demand for longer-term
those on intermediate-term bonds. This is a natural bonds. As a result, there is a temporary reduction in the
consequence of the fact that an MBS duration shock amount of interest rate risk that bond arbitrageurs must
raises the current duration risk premium. The expected bear, so the expected excess return on long-term bonds
returns on long duration bonds move more than those must fall in equilibrium.9
on intermediate-duration bonds. Clearly, the delta-hedging banks do not play an impor-
5. However, because they are expected to quickly dissi- tant role in this story. Hedging flows in the model with
pate, shocks to MBS duration have a humped-shaped banks correspond one-for-one to changes in the aggregate
effect on the yield curve and the forward rate curve, i.e., quantity of duration risk born by investors in the model
a shock to MBS duration increases the curvature of the without banks. As a result, equilibrium bond prices in the
yield curve. As in Greenwood and Vayanos (2014), the modified model are the same as in the simpler model that
effects of a supply shock on yields, which equals the omitted banks.10
effect on the bond's average returns over its lifetime,
are greater for intermediate-term bonds than for long-
3. Empirical analysis
term bonds when the supply shock is transient.
3.1. Measures of MBS duration and convexity
These two predictions mean that it is impossible to
infer whether there was, say, decline in the duration risk The aggregate duration of the MBS market is not a
premium simply by asking whether the yield curve flat- simple function of the current mortgage rate. MBS dura-
tened. And, notably, the last prediction suggests that tion also depends critically on the distribution of out-
transient shocks to MBS duration could account for some standing mortgage coupons and, in this way, reflects the
of the predictive power of the tent-shaped combination of past path of mortgage rates. Furthermore, times arise
forward rates identified by Cochrane and Piazzesi (2005). when aggregate refinancing and thus MBS duration are
more or less sensitive to changes in long yields.
2.3.5. The role of delta-hedging: stock versus flow effects Fortunately, the major brokerage firms publish widely
One important simplification in the current model is followed estimates of the effective duration of the US MBS
that only one class of investors owns MBS and prices market. Duration is the semi-elasticity of price with
interest rate risk. In practice, two sets of intermediaries respect to yield [DUR ¼ P 0 ðyÞ=P is the percentage change
own MBS. One set of intermediaries delta-hedge the in price for a small change in yield] and is the most widely
embedded prepayment option and, thus, bear a constant used measure of interest rate risk. Bond convexity mea-
amount of interest rate risk over time. Other investors do sures the curvature of the price-yield relation:
not delta-hedge and, thus, bear a time-varying amount of CONV ¼ ð1=2ÞP ″ ðyÞ=P. Thus, the percentage change in price
risk.8 following a Δy parallel shift in the yield curve is
Does it matter whether some MBS holders delta-hedge %ðΔP=PÞ   DUR  Δy þCONV  ðΔyÞ2 : ð10Þ
the prepayment option? In principle, the answer is “no”
because the relevant hedging flows correspond one-for- The duration of a negatively convex bond rises with yields.
one with changes in the aggregate quantity of duration By contrast, the duration of a positively convex bond falls
risk. To see this, consider a modification of the model with yields.
featuring a set of banks that extend all mortgages to
households. Assume banks do not bear interest rate risk 9
Instead of dynamically delta-hedging the interest rate exposure
and instead delta-hedge their interest rate exposure. of MBS, investors can statically hedge their exposures by purchasing
Specifically, as the prepayment option moves into-the- interest rate options. Regardless of whether MBS investors pursue a
money, the banks buy noncallable bonds (e.g., Treasuries) dynamic or a static hedging strategy, other investors must take the other
to keep the total duration of their assets fixed (and equal side of these trades. Thus, in the aggregate, fixed income investors must
bear a time-varying amount of interest rate risk that impacts equilibrium
to that of their liabilities). Finally, assume that the remain- term premia.
ing supply of Treasuries is held by risk-averse bond 10
There are reasons to think that the extent of hedging could play
arbitrageurs, and continue to assume that households do a nontrivial role in practice. The MBS market might not be perfectly
not hedge their time-varying interest rate exposure. integrated with the broader bond market, and the extent of hedging
could reflect the degree of integration. Suppose there are two types of
Suppose long-term rates suddenly drop, which
MBS investors: (1) delta-hedging intermediaries (e.g., the government-
raises mortgage refinancing expectations. From the per- sponsored enterprises and banks) who maintain a constant asset dura-
spective of banks, the existing mortgages now behave like tion, and (2) sleepers who inelastically buy a fixed quantity of MBS
irrespective of the duration risk they are taking (e.g., foreign official
holders). The duration that must be absorbed by arbitrageurs is
8
Conversations with market participants suggest that the θt Q MBS
t ðyð2Þ
t Þ. Thus, the fraction of MBS held by hedgers, an indicator of
government-sponsored enterprises and commercial banks have histori- the degree of market integration, determines the extent to which shifts in
cally been the most prominent delta-hedgers of MBS. MBS duration impact term premia.
 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299 277

MBS
For callable bonds such as MBS, PðyÞ is calculated using 2. DURt is the effective duration of the Barclays MBS
a stochastic term structure model and a prepayment Index and measures the percentage change in US
model that forecasts MBS cash flows as a function of Agency MBS market value following a shift in the
future rates. Using the riskless yield curve and parameters yield curve.
governing interest rate volatility as inputs, one simulates a 3. DUR_CNTRBMBS t  ðMV MBS
t =MV AGG
t Þ  DURMBS
t is the con-
large number of scenarios for the future path of rates. One tribution of MBS to Barclays Aggregate Index duration:
computes the expected cash flows in each scenario using
the mortgage prepayment model and then discounts these
expected cash flows using the implied zero-coupon curve Duration contribution of MBS:
DUR_CNTRBMBS
Duration contribution of other sectors:
DUR_CNTRBOTH
zfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflffl{ zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl
ffl{
t t
in that scenario. The model-implied price is computed by !
taking the probability-weighted average of the discounted AGG MV MBSt MBS MV MBS t OTH
DURt ¼  DURt þ 1   DURt :
cash flows in each scenario.11 MV AGG t MV AGG t
In the baseline results, I use estimates of MBS duration ð11Þ
from Barclays Capital, which are available from Datastream
beginning in 1989. Aside from data availability, starting
MBS
the analysis in the late 1980s is sensible because the Scaling DURt this way captures the fact that shifts in
MBS market rose to prominence only in the mid-1980s. MBS duration have had a growing impact on aggregate
The Barclays (formerly Lehman Brothers) Fixed Income bond market duration due to the growth of the MBS
Indices are the most widely followed set of bond indices in market. DUR_CNTRBMBS t proxies for the transient com-
the US. However, as shown in the Internet Appendix, ponent of aggregate bond market duration due to MBS
I obtain nearly identical results using the duration of the and constitutes my preferred forecasting variable.
Bank of America (formerly Merrill Lynch) US Mortgage
Master index, which is available starting in 1991. The
correlation between the Barclays and Bank of America
MBS duration measures is 0.79. 3.2. Understanding MBS duration
The Barclays US MBS index covers mortgage-backed
pass-through securities guaranteed by Government Before proceeding with the analysis, I explain several
National Mortgage Association (Ginnie Mae), Federal facts about MBS duration that play a key role in the story.
National Mortgage Association (Fannie Mae), and Federal First, due to the embedded refinancing option, the average
Home Loan Mortgage Corporation (Freddie Mac), collec- duration of MBS is low. Second, MBS duration rises as
tively known as US Agency MBS. The index is composed of interest rates rise. This negative convexity property under-
pass-through securities backed by conventional fixed-rate lies the positive-feedback dynamic emphasized in the
mortgages. The MBS index does not include nonagency or paper. Third, shocks to the MBS duration are transient,
private-label MBS (e.g., MBS backed by Jumbo, Alt-A, or having a half-life of roughly 5.5 months. Finally, shifts in
subprime mortgages).12 MBS duration are large relative to investors' risk-bearing
I examine three related measures of duration. capacity. In combination, the last two properties mean that
shocks to MBS duration generate significant, short-lived
AGG
1. DURt is the effective duration of the Barclays Aggre- shifts in bond risk premium.
gate Index and measures the percentage change in the
US bond market value following a parallel shift in the
Treasury yield curve. Barclays Aggregate Index is a
proxy for the broad US fixed income market and 3.2.1. Average MBS duration is low
includes Treasuries, Agency debentures, US Agency Panel A of Fig. 1 plots my three duration measures over
MBS, investment-grade corporates, and some asset- time. Table 1 presents summary statistics for the main
backed securities. Many bond portfolios are bench- variables in the paper. My monthly sample runs from
MBS
marked relative to the Barclays Aggregate Index. January 1989 to April 2011. DURt averages 3.35 years
with a minimum of 0.58 years in May 2003 and a
maximum of 4.83 years in May 1994. That is, the typical
price-yield sensitivity of MBS is roughly equivalent to that
11
To ensure the model-implied price equals the market price,
of a 3.5 year, zero-coupon bond.
analysts plug their discount function using an option-adjusted spread The low average duration of MBS reflects two factors.
(OAS), the constant spread one must add to the riskless curve in all First, the self-amortizing nature of fixed-rate mortgages
scenarios so the model-implied price equals the market price. Effective lowers their duration relative to nonamortizing bonds.
duration and convexity are computed at the current OAS.
12 Second, and more important, the prepayment option
Conventional mortgages satisfy several size, FICO, and loan-to-
value (LTV) requirements. For instance, mortgages in government- reduces the duration of MBS. A position in a callable bond
sponsored enterprise MBS are subject to the conforming loan size limit is equivalent to a position in a similar noncallable bond
set by Congress, typically have FICO scores over 620, and have a plus a short position in an interest rate call option to
maximum LTV of 80%. Barclays MBS Index is formed by grouping MBS repurchase the bond at par. Over the past 25 years, the
pools into generic pools based on agency, program (e.g., 30-year or 15-
year), mortgage coupon, and origination year. A generic pool is included
delta of the typical prepayment option has been substan-
in the index if it has a contractual maturity greater than one-year and tial. The option is typically struck slightly out-of-the-
more than $250 million outstanding. money at origination, so the steady decline in rates since
278 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299

2
5
Effective duration (years)

4 1

3
0

-1
1

0 -2

12

12
12

12
12

12

12
12

12
12

12

12

12
12

12

12

12

12

12

12

12

12

12

12

12

12

m
m

m
m

m
m

m
m

04

10

12
88

94

00

02

06

08
90

92

96

98
88

90

92

94

96

98

00

02

04

06

08

10

12

20
19

19

20

20

20

20

20

20
19

19

19

19
19

19

19

19

19

19

20

20

20

20

20

20

20
Aggregate duration MBS duration Detrended mortgage refinancing
Average MBS coupon - Current mortgage rate
MBS contribution to aggregate duration

1000

Ten-year equivalents, billions of 2012 dollars

1.5

500
Effective duration (years)

0.5 0

0
-500

-0.5
-1000
-1
19 12

19 12

19 12

19 12

19 12

20 12

20 12

20 12

20 12

20 12

20 12

20 12

12
12

12

12

12

12

12

12

12

12

12

12

12

12

m
m

88

90

92

94

96

98

00

02

04

06

08

10

12
88

90

92

94

96

98

00

02

04

06

08

10

12

19
19

19

19

19

19

19

20

20

20

20

20

20

20

Aggregate duration MBS contribution

Non-MBS contribution

Fig. 1. Effective duration of the US fixed income market. Panel A plots effective duration measures based on Barclays bond indices from 1989m1–2012m4:
the effective duration of the Barclays Aggregate Index ðDURAGG Þ, the effective duration of the Barclays Mortgage-backed Securities (MBS) Index ðDURMBS Þ,
and the duration contribution of the MBS Index to Aggregate duration ðDUR_CNTRBMBS Þ. Panel B shows the decomposition from Eq. (11) in 12-month
changes. Specifically, the change in Aggregate duration is the sum of the MBS contribution and the non-MBS contribution. Panel C plots detrended
refinancing activity based on the Mortgage Bankers' Association Refinancing Index versus aggregate refinancing incentives (the average coupon on
outstanding MBS minus the primary mortgage rate). Panels D shows detrended MBS duration in ten-year Treasury equivalents (in billions of 2012 dollars).
Panel A: Effective duration in levels. Panel B: 12-month changes in effective duration. Panel C: Detrended refinancing activity and refinancing incentives.
Panel D: Detrended duration in ten-year US Treasury equivalents (in billions of 2012 dollars).

the mid-1980s means most prepayment options have prepayments. Empirically, a 100 basis points (bps) increase
remained near the money.13 in the ten-year Treasury yield has been associated with a
MBS AGG
1.37-year increase in DURt and a 0.34 increase in DURt
14
since 1989.
3.2.2. Negative convexity of MBS
MBS convexity has played a larger role in US bond
MBS are typically negatively convex. Their duration
markets over time for two reasons. First, the MBS market
rises as interest rates rise due to a decline in expected
has grown significantly. The MBS index grew from 25% of
the Barclays Aggregate Index in 1989 to 40% in 2008 but
13
The price of the callable bond is P C ðyÞ ¼ PðyÞ  CðPðyÞÞ, where PðyÞ is had fallen to 30% as of early 2012 due to the post-crisis
the price of the underlying noncallable bond and CðPÞ is the price of the surge in Treasury borrowing. Second, advances in informa-
interest rate call. Because 0o C0 ðPÞ o 1,
tion technology and heightened competition in mortgage
Duration of callable Noncallable duration equivalents
banking have reduced the costs of refinancing. As a result,
 P 0C ðyÞ ¼ 0
ð1 C ðPðyÞÞÞ
Duration of noncallable Duration of noncallable
refinancing now responds more aggressively to changes in
0 0
  P ðyÞ o  P ðyÞ ; primary mortgage rates, so the MBS market has become
more negatively convex. For instance, prior to 2000, a 100
Convexity of callable Noncallable duration equivalents
P ″C ðyÞ ¼ ð1  C0 ðPðyÞÞÞ
bps increase in the ten-year yield was associated with a
ðDuration of noncallableÞ2
1.09-year increase in the MBS duration. Since 2000, a 100
Convexity of noncallable Call optionΓ 4 0
″ ″ 0 2
 P ðyÞ  C ðPðyÞÞ  ðP ðyÞÞ :

Because P ″ ðyÞ 40 and C″ ðPÞ 4 0, the sign of P ″C ðyÞ is ambiguous. However,

because callable bonds become morenegatively convex as rates decline, 14
Estimating ΔDURMBS
t ¼ a þ b  Δyð10Þ
t þ εt using monthly differences,
P ″C ðyÞo 0 unless the prepayment option is deeply out-of-the-money. I find b ¼ 1.37 ðR2 ¼ 20:29Þ.
 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299 279

Table 1
Summary statistics.
This table presents means, medians, standard deviations, extreme values, and monthly time series autocorrelations (denoted ρ) of variables between
1989m1 and 2011m4. Panel A presents summary statistics for US Treasury excess returns, yield spreads, and forwards from Gürkaynak, Sack, and Wright
(2007). These variables are all measured in percentage points. Panel B presents summary statistics for various measures of effective duration in years based
on Barclays Capital bond indices. Panel C reports the corresponding summary statistics for effective convexity. The effective convexity measures are
available beginning only in 1997m4.

Variable N Mean Median St Dev Min Max ρ

Panel A: US Treasury excess returns, yield spreads, and forwards (%)

rxð5Þ 268 3.00 3.33 4.39  8.49 12.46 0.922

tþ1
rxð10Þ 268 4.98 5.74 7.73  16.41 20.96 0.884
tþ1
rxð20Þ 268 6.32 6.35 12.98  29.35 39.17 0.861
tþ1
yð5Þ ð1Þ 268 0.90 0.76 0.83  0.54 2.50 0.973
t  yt
ytð10Þ  yð1Þ 268 1.57 1.20 1.28  0.58 3.79 0.977
t
ytð20Þ  yð1Þ 268 2.01 1.57 1.54  0.82 4.68 0.978
t
ð1Þ 268 4.45 4.59 2.24 0.16 9.70 0.976
ft
ð2Þ 268 4.93 4.98 2.01 0.62 9.32 0.971
ft
ð3Þ 268 5.33 5.35 1.80 1.42 9.19 0.970
ft
ð4Þ 268 5.67 5.60 1.64 2.31 9.17 0.970
ft
ð5Þ 268 5.96 5.76 1.52 3.02 9.16 0.969
ft

Panel B: Effective duration (years)

DURAGG 268 4.53 4.56 0.26 3.71 5.12 0.863

t
DURMBS 268 3.35 3.49 0.91 0.58 4.83 0.875
t
DUR_CNTRBMBS
t
268 1.08 1.12 0.29 0.20 1.78 0.851
DUROTH 268 5.13 5.16 0.25 4.47 5.61 0.954
t
DUR_CNTRBOTH 268 3.46 3.45 0.18 2.93 3.88 0.960
t
MV MBS =MV AGG 268 0.33 0.33 0.04 0.26 0.41 0.991
t t

Panel C: Effective convexity (years)

CONV AGG 169  0.18  0.19 0.17  0.52 0.20 0.844

t
CONV MBS 169  1.50  1.54 0.45  2.43  0.44 0.825
t
CONV MBS  ðMV MBS =MV AGG Þ 169  0.52  0.54 0.16  0.85  0.15 0.818
t t t

bps increase in the ten-year yield was associated with a mortgage rate from Freddie Mac's survey, ct yM;t .
1.57-year increase in the MBS duration.15 As shown in Panel C, a strong positive relation exists
Another way to illustrate the negative convexity of the between the two series. The corresponding time series
MBS market is to examine aggregate refinancing behavior. regression has a R2 of 0.65 and suggests that a 100 bps
MBS duration is low when expected future refinancing is increase in refinancing incentives boosts aggregate refi-
high. And current refinancing activity is a strong signal of nancing by 85% relative to trend.
refinancing over the near-term. Panel C of Fig. 1 plots the
Mortgage Bankers' Association Refinancing Index, which 3.2.3. Shocks to MBS duration are transient
reflects the raw number of mortgage applications classi- Shocks to the bond market duration are transient.
fied as refinancings, versus a measure of aggregate refi- Specifically, the one-month autocorrelation of the duration
nancing incentives. Specifically, I show the Hodrick and of the Barclays Aggregate Index, DURAGG t , is 0.88. The
Prescott (1997) filtered version of the log-refinancing transient nature of aggregate duration dynamics is driven
index, essentially the percentage deviation of refinancings by the MBS component of the index. Specifically, the one-
MBS
from an estimated trend. A borrower has strong incentives month autocorrelation of DURt is also 0.88, which
to refinance when the primary rate is below her mortgage implies that MBS duration shocks have a half-life of 5.5
coupon, so my measure of aggregate refinancing incen- [ ¼ln(0.5)/ln(0.88)] months. By contrast, the duration of
tives is the difference between the average coupon of MBS the non-MBS portion of the index is far more persistent
in the Barclays MBS Index and the 30-year primary with a one-month autocorrelation of 0.96, implying a half-
life for non-MBS duration shocks of 17.5 months.
Why are shocks to MBS duration so transient? In the
15
Beginning in 1997, Barclays estimates the effective negative case of a downward shock to rates, the explanation is fairly
convexity of the MBS index. If I estimate ΔDURMBS
t ¼
a þ b  NCONV MBS  Δyð10Þ þ εt , I obtain a ¼0.02 and b ¼0.97 ðR2 ¼ 0:75Þ
mechanical. A decline in rates raises expected prepay-
t t
precisely as one would expect. Furthermore, the estimated negative ments, thus lowering the amount of duration risk that
convexity has trended up over time. holders of outstanding MBS face over the near-term.
280 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299

As homeowners respond to this interest rate shock, pre- 3.3.1. Basic forecasting results
payments rise. When a borrower prepays, she replaces a I use data from Gürkaynak, Sack, and Wright (2007) on
high coupon bond with a deep in-the-money prepayment the nominal Treasury yield curve as updated regularly by
option with a lower coupon bond with a slightly out-of- the Federal Reserve Board. The log excess return on an
the-money prepayment option. Borrowers effectively n-year zero-coupon bonds is defined as rxðnÞ ðnÞ
t þ 1 ¼ n  yt 
restrike the interest rate call options embedded in their ðn  1Þ  yðn
tþ1
 1Þ
 yð1Þ
t . Panel A of Table 1 provides summary
mortgages. The net effect of this refinancing transaction statistics on 12-month excess returns ðrxðnÞ t þ 1 Þ and yield
raises the total quantity of duration that investors need to spreads ðyðnÞ ð1Þ
t  yt Þ for n ¼5,10, and 20. I also summarize
ðmÞ
bear. However, because prepayments are gradual, there are information on the instantaneous forward rates f t for m
temporary declines in aggregate MBS duration, which then ¼ 1, 2, 3, 4, 5.
predictably revert over time. By contrast, in the case of an Table 2 uses the three bond duration measures to forecast
upward shock to rates, there is no expectation of such excess returns on ten-year nominal US Treasuries. Specifi-
rapid mean reversion in MBS duration. This is because cally, Table 2 presents forecasting regressions of the form
upward shocks to rates are expected to slow refinancing
activity, so there is no mechanical restriking effect. rxtð10Þ 0 ð10Þ
þ 1 ¼ a þ b  DURt þ c xt þ εt þ 1 : ð12Þ

3.2.4. Shifts in MBS duration are large relative to risk- For the sake of comparability with the recent literature, the
bearing capacity regressions are estimated with monthly data, so each month
Due to the vast size of the MBS market, shifts in MBS I am forecasting excess returns over the following 12 months.
duration have a large effect on the aggregate amount of To deal with the overlapping nature of returns, t-statistics are
interest rate risk born by investors. And shifts in MBS duration based on Newey and West (1987) standard errors allowing
drive almost all high-frequency variation in aggregate bond for serial correlation at up to 18 lags. I estimate these
market duration. This is shown in Panel B of Fig. 1 which uses regressions with and without other forecasting variables
Eq. (11) to decompose 12-month changes in aggregate bond identified in the literature on bond risk premia. Specifically,
market duration into an MBS component and a non-MBS I control for the term spread following Campbell and Shiller
component. The MBS component accounts for the vast bulk of (1991) and the first five forward rates following Cochrane
the variation in aggregate duration. The R2 from a regression and Piazzesi (2005). I obtain similar results controlling for
AGG
of DURt on DUR_CNTRBMBS t is 0.61 in levels, 0.81 in the ten-year forward rate spread following Fama and Bliss
12-month changes, and 0.91 in one-month changes. (1987) or if I control for longer-dated (e.g., seven-, ten-, and
To get a sense of the dollar magnitudes, investors convert 20-year) forwards as in Cieslak and Povala (2013).
AGG
aggregate duration statistics into ten-year US Treasury equiva- Table 2 shows that aggregate duration (DURt ),
MBS
lents. If the market value of some asset class X is QX with MBS duration ðDURt Þ, and MBS duration contribution
effective duration DX and the duration of a ten-year US ðDUR_CNTRBMBS t Þ are each very strong predictors of bond
Treasury is DT10 , then X represents Q X  ðDX =DT10 Þ ten-year excess returns. MBS duration contribution is the strongest
Treasury equivalents. Panel D of Fig. 1 plots the detrended predictor, both in univariate specifications and in the multi-
component of MBS duration in ten-year Treasury equivalents variate regressions. Duration adds incremental forecasting
[the trend is estimated using the Hodrick and Prescott (1997) power over and above the term spread, the forward rate
filter]. I convert historical dollars to 2012 dollars using the spread, and arbitrary combinations of forward rates. Speci-
consumer price index. The figure shows that the resulting fically, as shown in Table 2, duration significantly raises the
shifts in duration supply are massive. In 26 sample months, R2 in these specifications, and the R2 's are already high. For
detrended MBS ten-year equivalents exceed þ$500 billion or instance, Column 2 shows that using the first five forward
 $500 billion. rates as predictors delivers a forecasting R2 of 0.31. If MBS
These are arguably very large shifts in the quantity of duration contribution is added to this multivariate specifi-
duration risk relative to investors’ risk-bearing capacity. I can cation as in Column 11, the forecasting R2 jumps to 0.46.
compare the shifts in MBS duration with the Fed's Quantita- Comparing Columns 2 and 11, the inclusion of duration
tive Easing operations. From 2008Q4 to 2010Q1, the Fed measures roughly halves the absolute magnitude of the
purchased $1,250 billion MBS with an effective duration of coefficients on forward rates in the augmented Cochrane
roughly three years. Because the duration of the ten-year and Piazzesi (2005) regression. I return to this observation
Treasury note was roughly eight years at this time, this below when I explain why MBS duration could help to
amounted to a reduction of $469 billion [$469¼$1,250  (3/ account for some of the forecasting power of Cochrane and
8)] ten-year equivalents of duration that needed to be held by Piazzesi's tent factor.
investors (assuming no further change in interest rates). Thus, Fig. 2 illustrates the basic forecasting result. Specifically,
Panel D shows that numerous shifts in MBS duration since the Panel A shows a scatter plot of 12-month future excess
late 1980s have been of a comparable scale to the Fed's QE returns versus MBS duration contribution. These plots
operations. Furthermore, these shifts can occur rapidly, which correspond to the multivariate estimates in Column 10.
is likely to be relevant if capital is slow-moving. Specifically, I plot the component of excess returns that is
orthogonal to the term spread versus the component of
3.3. Forecasting excess bond returns duration that is orthogonal to the term spread. The strong
positive relation is evident in Panel A. Panel B plots 12-month
I now show that measures of MBS duration are strong future excess returns versus initial MBS duration contribution
predictors of excess government bond returns. over time. The figure shows that MBS duration adds
 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299 281

Table 2
Forecasting excess bond returns using measures of mortgage-backed securities (MBS) duration.
Regressions of 12-month excess returns on ten-year Treasuries on the effective duration of the Barclays Aggregate Index, the effective duration of the
Barclays MBS Index, and the effective duration contribution of the Barclays MBS Index:

rxtð10Þ 0 ð10Þ
þ 1 ¼ a þ b  DURt þ c xt þ εt þ 1 :

The regressions are estimated with monthly data from 1989m1–2011m4, so each month I forecast the excess return over the following 12 months. To deal
with the overlapping nature of returns, t-statistics are based on Newey and West (1987) standard errors allowing for serial correlation at up to 18 lags.
I estimate these regressions with and without other forecasting variables identified in the literature on bond risk premia. Specifically, I control for the term
spread following Campbell and Shiller (1991) and the first five forward rates following Cochrane and Piazzesi (2005).

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)

DURAGG
t 6.725 11.482 7.749
[2.84] [5.44] [2.73]

DURMBS
t 2.905 4.437 3.929
[2.76] [6.48] [4.59]

DUR_CNTRBMBS
t 11.253 14.629 12.143
[3.70] [5.78] [4.44]

ytð10Þ  yð1Þ
t 2.178 2.928 3.231 2.999
[3.02] [3.88] [4.67] [4.57]

ð1Þ
ft  37.854  30.755  27.953  22.653
[  6.77] [  4.91] [  4.59] [  3.89]

ð2Þ
ft 217.606 171.892 146.201 114.797
[7.00] [4.77] [4.68] [3.78]

ð3Þ
ft  511.359  403.066  340.160  268.889
[  6.92] [  4.80] [  4.74] [  3.78]

ð4Þ
ft 533.181 418.391 357.217 286.175
[6.66] [4.72] [4.60] [3.65]

ð5Þ
ft  201.349  156.236  136.066  109.305
[  6.26] [  4.47] [  4.29] [  3.36]

Constant 1.555 2.065  25.513  51.686  34.283  4.740  14.948  6.634  7.158  15.515  11.768
[1.04] [0.48] [  2.34] [  5.08] [  2.56] [  1.18] [  4.58] [  1.27] [  1.85] [  4.23] [  1.95]

Number of observations 268 268 268 268 268 268 268 268 268 268 268
R2 0.13 0.31 0.05 0.26 0.35 0.12 0.37 0.42 0.17 0.40 0.46

significant forecasting power because duration extensions in duration shocks relative to the scale of arbitrage capital, the
1991, 1994, 1999, 2004, 2007, 2008, and 2011 were each effects do not seem implausibly large. I return to this point
followed by high bond excess returns. Conversely, duration below when I compare the implied price impact from these
contractions in 1993, 1995, 1998, 2002, 2003, and 2009 were regressions with estimates from the literature evaluating the
followed by low excess returns. Thus, Fig. 2 shows that the Federal Reserve's Quantitative Easing policies.
basic forecasting result is not driven by one subsample or by a These forecasting regressions suggest a simple market-
handful of outlying events. In untabulated regressions, I obtain timing strategy for diversified investors (e.g., endowments and
strong results when I separately examine the 1989–1999 pensions) who allocate capital between the bond market and
subsample or the 2000–2011 subsample. other financial markets. Specifically, diversified investors
What is the economic magnitude of the estimated should overweight long-term bonds when MBS duration is
effects? The coefficient in Column 10 indicates that a high and underweight long-term bonds when MBS duration
one-year increase in DUR_CNTRBMBS t raises expected excess is low. Unlike the specialist bond arbitrageurs emphasized in
returns on ten-year bonds by 14.629% over the following the model, the changes in the aggregate quantity of interest
12 months. A 1 standard deviation move in MBS duration rate risk perceived by these investors is miniscule because
contribution is 0.29 years, so this implies that a 1 standard bonds are only a small portion of their overall financial
deviation increase in duration raises expected excess portfolios. Thus, such diversified investors perceive the result-
returns by 4:20% ¼ 14:629  0:29. Assuming this shift ing time-variation in expected returns as time–varying alpha
has no effect on expected returns beyond 12 months as opposed to a time-varying bond risk premium.
(analysis below suggests this is a reasonable assumption), this Naturally, the t-statistics from the forecasting regres-
corresponds to a rise in ten-year yields of 42 [¼420/10] bps sions can be translated directly into a statement about
today. Thus, the estimated effects are highly economically the Sharpe ratio of strategy that buys (sells) long-term
significant. However, given the size and speed of the MBS bonds to the extent that DUR_CNTRBMBS t is above (below) its
282 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299

3.3.2. MBS duration accounts for all the forecasting power of

20 aggregate duration
Table 3 shows that all of the forecasting power of
DURAGG derives from the MBS component. Specifically, I
10
decompose DURAGG into the sum of DUR_CNTRBMBS
Future excess return

t and
DUR_CNTRBOTH t as in Eq. (11). The transient MBS compo-
0 nent accounts for the vast majority of all high-frequency
variation in aggregate bond market duration. The 12-
-10
month autocorrelations of DURAGG, DUR_CNTRBMBS t , and
DUR_CNTRBOTH t are 0.16, 0.04, and 0.64, respectively. Thus,
perhaps not surprisingly, the horse races in Columns 4, 8,
-20 and 12 of Table 3 show that all of the near-term forecasting
power in aggregate bond market duration is attributable to
-1 -0.5 0 0.5 1
MBS duration. A similar conclusion holds if, instead of
MBS duration contribution
DUR_CNTRBOTH t , I focus more narrowly on the duration of
Future excess return Fitted values the Barclays Treasury Index.
In summary, both the duration of the Barclays Aggre-
gate Index and the Barclays MBS Index positively forecasts
20 1
excess bond returns over the following 12 months. And the
MBS duration contribution duration contribution of MBS, which scales MBS duration
10 relative to the broader market, is the strongest forecaster.
Future excess return

0.5
However, the measured duration of the Treasury and other
0 non-MBS fixed income markets do not reliably forecast
0
bond returns at the quarterly or annual frequencies con-
sidered here, at least not in my 1989–present sample. This
-10
-0.5 is perhaps not surprising if one considers a preferred-habit
model with slow-moving capital. In such a model, capital
-20
-1
flows in response to persistent changes in bond risk
premia, which neutralizes some, but not all, of the effects
19 12

19 12

19 12

19 12

19 12

20 12

20 12

2 0 12

20 12

20 12

20 12

12

of persistent shocks to duration supply on expected

m

m
m

m
m

m
88

04
94

10
90

96

98

00

06

08
92

02

returns. Thus, from a medium-frequency return forecast-

19

Future excess return MBS duration contribution ing perspective, it could be particularly useful to isolate the
transient component of shocks to bond duration supply.
Fig. 2. The basic forecasting result using mortgage-backed securities
(MBS) duration contribution. This figure depicts the regression of future
And the preceding analysis suggests that MBS duration is
12-month excess bond returns on the term spread and DUR_CNTRBMBS t , an excellent proxy for this transient component of dura-
corresponding to Column 10 of Table 2. Specifically, Panel A shows a tion supply.17
scatter plot of the component of future excess returns that is orthogonal
to the term spread versus the component of DUR_CNTRBMBS that is
t
3.3.3. Time signature of MBS duration effects
orthogonal to the term spread. Panel B shows the corresponding time
series plot. Panel A: Scatter plot. Panel B: Time-series plot. Variation in MBS duration is expected to be associated
with relatively high frequency variation in expected
returns for several reasons. First, if current MBS duration
unconditional average.16 Estimating a univariate regression positively predicts bond excess returns over the following
with quarterly data yields a t-statistic ofp2.91,
ffiffiffiffiffiffi implying a instant, the current duration would not be expected to
strategy Sharpe ratio of 0:303 ¼ 2:91C 92 per quarter. reliably predict returns on 12-month forward basis
Assuming that strategy returns are independent pover
ffiffiffi time, because MBS duration is not itself very persistent. Second,
this implies an annual Sharpe ratio of 0:606 ¼ 4  0:303. slow-moving capital effects as in Duffie (2010) could exist,
The fact that this timing strategy appears to be so profitable which implies that the short-run demand curve for dura-
suggests that diversified investors do not fully exploit it. tion risk is more inelastic than the long-run demand curve.
This could be because various frictions combine to limit the Specifically, suppose that, in the short run, interest rate
speed at which arbitrage capital flows across markets as in risk is priced by specialized bond market investors with a
Duffie (2010). Or this could simply be because many fixed risk tolerance. However, over longer horizons, diver-
diversified investors are unaware of the high Sharpe ratio sified investors could allocate financial capital between the
offered by the strategy. bond market and other financial markets as in Duffie and
Strulovici (2012). Thus, following a positive shock to
duration supply, bond risk premia could be expected to
jump. The anticipation of large future bond returns would

16
Suppose r t þ 1 ¼ α þ β  xt þ εt þ 1 and consider returns on the strat-
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 17
Admittedly, because my measures of MBS duration reflect
egy wt þ 1 ¼ r t þ 1 ðxt  E½xt Þ. Then, SR½wt þ 1  ¼ βs2 ½xt C s2 ½εt þ 1 s2 ½xt  ¼
pffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi forward-looking expectations of future MBS duration but my measures
t½bOLS C T , where t½bOLS  ¼ βC T  1 s2 ½εt þ 1 ðs2 ½xt Þ  1 is the population of Treasury duration are not forward-looking, this horse race between
t-statistic for the ordinary least squares estimator of β. MBS and non-MBS duration is not a completely fair fight.
 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299 283

Table 3
Forecasting excess bond returns, mortgage-backed securities (MBS) vs. non-MBS duration.
Regressions of 12-month excess returns on ten-year Treasuries on the effective duration of the Barclays Aggregate Index, the effective duration
contribution of the Barclays MBS Index, and the effective duration contribution of the non-MBS index:

rxtð10Þ 0 ð10Þ
þ 1 ¼ a þ b  DURt þ c xt þ εt þ 1 :

The regressions are estimated with monthly data from 1989m1–2011m4, so each month I forecast the excess return over the following 12 months. To deal
with the overlapping nature of returns, t-statistics are based on Newey and West (1987) standard errors allowing for serial correlation at up to 18 lags.
I estimate these regressions with and without other forecasting variables identified in the literature on bond risk premia. Specifically, I control for the term
spread following Campbell and Shiller (1991) and the first five forward rates following Cochrane and Piazzesi (2005).

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)

DURAGG
t 6.725 11.482 7.749
[2.84] [5.44] [2.73]

DUR_CNTRBMBS
t 11.253 8.872 14.629 13.586 12.143 10.178
[3.70] [3.25] [5.78] [6.63] [4.44] [3.48]

DUR_CNTRBOTH
t  14.620  8.022  13.676  3.179  13.200  5.577
[  2.10] [  1.11] [  1.92] [  0.44] [  1.73] [  0.68]

ytð10Þ  yð1Þ
t 2.928 2.999 2.053 2.912
[3.88] [4.57] [2.74] [4.14]

ð1Þ
ft  30.755  22.653  33.422  23.239
[  4.91] [  3.89] [  5.39] [  4.00]

ð2Þ
ft 171.892 114.797 183.721 117.115
[4.77] [3.78] [5.04] [3.86]

ð3Þ
ft  403.066  268.889  432.257  274.699
[  4.80] [  3.78] [  4.87] [  3.96]

ð4Þ
ft 418.391 286.175 460.215 295.311
[4.72] [3.65] [4.87] [3.99]

ð5Þ
ft  156.236  109.305  178.142  114.393
[  4.47] [  3.36] [  4.85] [  3.82]

Constant  25.513  7.158 55.500 23.133  51.686  15.515 49.011  3.268  34.283  11.768 48.946 10.278
[  2.34] [  1.85] [2.31] [0.88] [  5.08] [  4.23] [1.98] [  0.12] [  0.56] [  1.95] [1.92] [0.34]

Number of
observations 268 268 268 268 268 268 268 268 268 268 268 268
R2 0.05 0.17 0.12 0.20 0.26 0.40 0.23 0.41 0.35 0.46 0.39 0.47

draw in capital from other markets, raising the risk meaningfully from three to four quarters out, and generally
tolerance of bond investors and, thus, reducing the impact little predictive power is evident beyond five quarters.
of supply on bond risk premia over time in the case of a To investigate the timing of these effects more para-
permanent supply shock. metrically, I estimate first-order vector autoregression
To formally explore the time signature of my main (VAR) including excess returns, MBS duration, and the
forecasting result, I use MBS duration to forecast quarterly term spread using quarterly data:
returns from one to eight quarters ahead. Fig. 3 presents
quarterly forecasting regressions of the form rxtð10Þ ð10Þ
þ 1=4 ¼ a1 þb1  rxt þ c1  DURt þd1  ðyt yð1Þ
t Þ þ ε1;t þ 1=4 ;

rxtð10Þ MBS
þ ðj  1Þ=4-t þ j=4 ¼ aðjÞ þ bðjÞ  DUR_CNTRBt DURt þ 1=4 ¼ a2 þ b2  rxt þ c2  DURt þ d2  ðyð10Þ  ytð1Þ Þ þ ε2;t þ 1=4 ;
t

þ c0ðjÞ xt þεtð10Þ
þ ðj  1Þ=4-t þ þ j=4 ð13Þ yð10Þ ð1Þ ð10Þ
t þ 1=3  yt þ 1=4 ¼ a3 þ b3  rxt þ c3  DURt þd3  ðyt  yð1Þ
t Þ þ ε3;t þ 1=4 :

ð14Þ
for j ¼ 1; …; 8. Fig. 3 then plots the coefficients bðjÞ versus the
horizon j. This series of regressions provides a simple non- Using DUR_CNTRBMBS t ¼ ðMV MBS
t =MV AGG
t Þ  DURMBS
t as the
parametric way to trace out the impulse response of quarterly measure of MBS duration, Fig. 4 shows the resulting simple
excess bond returns following a movement in MBS duration. impulse response of quarterly excess bond returns to a
Specifically, Panel A of Fig. 3 plots the coefficients bðjÞ from shock to MBS duration. As expected, the effect on excess
univariate regressions, and Panel B plots the coefficients from bond returns from a shock to MBS duration is concentrated
multivariate specifications that control for the term spread. in the next four quarters. Furthermore, using a parametric
Fig. 3 shows that the forecasting power of bond duration is VAR, paints a very similar picture of the timing of the effects
largely located in the following two quarters. The effect decays to the nonparametric approach show in Fig. 3.
284 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299

10
10

Excess returns
5
Coefficient (%)

-5
0 1 2 3 4 5 6 7 8
-5 Quarters

1 2 3 4 5 6 7 8 95% Confidence Interval Impulse response function

Horizon (quarters)
Fig. 4. Impulse response of quarterly excess bond returns to shock to
mortgage-backed securities (MBS) duration. Using quarterly data, I estimate
a first order vector autoregression (VAR) of the form shown in Eq. (14).
10 For simplicity, I show results only for DUR ¼ DUR_CNTRBMBS . I plot the
(simple) impulse response function from a shock to DUR_CNTRBMBS in
quarter t ¼ 0 on excess bond returns from quarters t ¼ 1,…, 8. Confidence
intervals for the estimated impulse response are shown in gray. For
simplicity, I use a quarterly data set of nonoverlapping observations,
5 sampled in March, June, September, and December.
Coefficient (%)

DUR_CNTRBMBS t in Panel B. As shown in Fig. 5, the coefficients

0
bðnÞ are increasing in maturity n, consistent with the idea
that shifts in MBS duration impact the duration risk premia
in bond markets.
How do shifts in MBS duration impact bond yields and
-5
the current shape of the yield curve? Due to their transitory
1 2 3 4 5 6 7 8
nature, MBS duration supply shocks would be expected to
Horizon (quarters) have a hump-shaped effect on the yield curve and forward
Fig. 3. Coefficients by quarterly forecast horizons. This figure plots the rate curve. Although a transitory rise in the duration risk
coefficients bðjÞ on MBS duration contribution from estimating premium has the largest impact on the price and current
expected excess returns of long-term bonds, it has the largest
rxð10Þ MBS
t þ ðj  1Þ=4-t þ j=4 ¼ aðjÞ þ bðjÞ  DUR_CNTRBt þ
impact on the yield of intermediate-term bonds. As in
c0ðjÞ Xt þ εtð10Þ
þ ðj  1Þ=4-t þ j=4 Greenwood and Vayanos (2014), the intuition is that the
for quarters j ¼ 1,…, 8, effectively tracing out a nonparametric version of impact bond yields equals the effect on a bond's average
the impulse to quarterly excess bond returns following a movement in returns over its lifetime. As a result, a temporary rise in
mortgage-backed securities (MBS) duration. Panel A plots the coefficients
from univariate forecasting regressions. The multivariate specifications in
duration risk premium has a greater impact intermediate-
Panel B controls for the term spread. Confidence intervals, based on term yields than on long-term yields when the supply shock
Newey and West (1987) standard errors allowing for serial correlation at is expected to be short-lived.
up to six months, are shown as dashed lines. Panel A: Univariate These results are shown in the middle and right-most
coefficients by quarterly forecast horizon. Panel B: Multivariate coeffi-
plots in Fig. 5. Specifically, I plot the slope coefficients
cients by quarterly forecast horizon.
versus maturity from estimating

3.3.4. Results for multiple maturities yðnÞ ðnÞ ðnÞ

t  yt þ 1 ¼ aðnÞ þ bðnÞ  DURt þεt þ 1 ð16Þ
If changes in DURt shift the duration risk premium in US
and
bond markets, this would be expected to have a larger effect
ðnÞ ðnÞ
on the expected returns of long-term bonds than on inter- f t f t þ 1 ¼ aðnÞ þbðnÞ  DURt þ εtðnÞ
þ1 ð17Þ
mediate bonds. For instance, the coefficient on DURt when
forecasting 20-year excess bond returns should be larger for n ¼ 0; 1; 2; …; 20. In other words, I use DURt to forecast
than when forecasting ten-year excess bond returns. This is the decline in yields and forwards over the following 12
illustrated in the left-most plots in Fig. 5 where I estimate months for each maturity. Fig. 5 shows that high level of
current MBS duration has a humped shaped effect on the
rxtðnÞ ðnÞ
þ 1 ¼ aðnÞ þbðnÞ  DURt þεt þ 1 ð15Þ yield and forward rate curves. Specifically, the expected
decline in rates over the following 12 months is largest for
for n ¼ 2; …:20. I then plot the coefficients, bðnÞ , versus intermediate yields and forwards, having its maximal
MBS
maturity n. I show results for DURt in Panel A and effect on two-year yields.
 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299 285

Univariate return forecasting coefficients by maturity Univariate yield forecasting coefficients by maturity Univariate forward forecasting coefficients by maturity
10 0.8 0.8

8 0.6 0.6
Coefficient (%)

Coefficient (%)

Coefficient (%)
0.4 0.4
6
0.2 0.2
4
0 0
2 -0.2 -0.2

0 -0.4 -0.4
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Maturity (years) Maturity (years) Maturity (years)

Univariate return forecasting coefficients by maturity Univariate yield forecasting coefficients by maturity Univariate forward forecasting coefficients by maturity
10 0.8
0.8
8 0.6
0.6
Coefficient (%)

Coefficient (%)

Coefficient (%)
0.4
6 0.4
0.2 0.2
4
0 0
2
-0.2 -0.2

0 -0.4 -0.4
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Maturity (years) Maturity (years) Maturity (years)

Fig. 5. Excess return, yield, and forward rate forecasting coefficients by bond maturity. This figure plots the coefficients bðnÞ from estimating

rxðnÞ ðnÞ
t þ 1 ¼ aðnÞ þ bðnÞ  DURt þ εt þ 1

versus maturity n, for n ¼ 2,…,20. I then plot the coefficients bðnÞ versus maturity n from estimating

yðnÞ ðnÞ ðnÞ

t  yt þ 1 ¼ aðnÞ þ bðnÞ  DURt þ εt þ 1

ðnÞ ðnÞ
f t  f t þ 1 ¼ aðnÞ þ bðnÞ  DURt þ εðnÞ
t þ 1;

for n ¼ 0, 1, 2,…,20. For simplicity, I rescale the coefficients bðnÞ so that they reflect the impact of one standard deviation shift in DURt . Confidence intervals,
based on Newey and West (1987) standard errors allowing for serial correlation at up to 18 months, are shown as dashed lines. Panel A shows results for
mortgage-backed securities (MBS) duration, and Panel B shows results for MBS duration contribution. In each panel, the left-most graph shows the excess
return forecasting results by maturity, the middle graph shows yield-change forecasting results by maturity, and the right-most graph shows forward-
change forecasting results. Panel A: MBS duration. Panel B: MBS duration contribution.

This suggests that transient shocks to MBS duration in the maturity structure of borrowing (holding constant
could account for some of the predictive power of the tent- total dollar borrowing) can significantly impact term
shaped combination of forward rates identified by premia. And this indicates that the Fed's Quantitative
Cochrane and Piazzesi (2005). The Cochrane and Piazzesi Easing policies can impact the duration risk premium
(2005) factor picks up time variation in the curvature of through a broad portfolio balance channel, consistent with
the yield curve and is useful for forecasting transitory the findings of Gagnon, Raskin, Remache, and Sack (2011)
variation in bond returns, i.e., variation at frequencies and Greenwood and Vayanos (2014).
higher than a standard business cycle frequency. Thus,
the variation in MBS duration could help explain why a 3.3.5. Robustness checks
tent-shaped combination of forwards can explain transi- Running a regression of returns on quantities does not,
tory variation in bond risk premia. Of course, any fast- in general, allow one to cleanly identify a demand curve.
moving state variable that affects bond risk premia should Instead, to nail down the MBS duration supply channel, I
have a hump-shaped effect on yields and forwards as basic would like to show that some component of MBS duration
matter of no-arbitrage bond pricing logic. However, the that is exogenous with respect to interest rates positively
literature has generally struggled to produce economically forecasts excess bond returns. Otherwise, one might worry
plausible state variables that might induce high-frequency that MBS duration is simply correlated with some omitted
variation in bond risk premium. I argue that MBS duration variable that is driving demand for bonds. Unfortunately,
is one such variable and, thus, the fact that it fights with constructing a valid and powerful instrument for MBS
the Cochrane and Piazzesi (2005) factor is natural. duration is difficult. However, because the issues
In summary, the findings for multiple maturities are addressed here are of first-order importance for under-
highly consistent with the model I develop in the Internet standing of bond market dynamics and the determination
Appendix and with prior work by Vayanos and Vila (2009) of risk premia more generally, evidence that is admittedly
and Greenwood and Vayanos (2014), who argue that shifts somewhat indirect should be acceptable. Specifically, I am
286 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299

forced to rely on predictive regressions that provide equivalents in Fig. 1, I estimate that a 1 standard
MBS
indirect evidence consistent with the MBS hedging story. deviation decline in DURt corresponds to a $503 billion
In the Internet Appendix, I address several concerns reduction in ten-year equivalents in recent years.18
raised by this indirect approach. First, one could be con- The first LSAP was announced in late 2008 and early 2009
cerned that variation in DURt does itself not drive the term amidst unusually strained market conditions (Krishnamurthy
premium but is simply correlated with an omitted variable and Vissing-Jorgensen, 2013). Because LSAP1 represented a
that does. The Internet Appendix shows that the coefficient cumulative reduction in $750 billion ten-year equivalents,
on DURt is highly robust to the inclusion of additional my estimates imply a 54 bps [¼36 bps  (750/503)] decline
controls that are thought to be associated with term premia. in ten-year Treasury yields. This implied effect is smaller than
This should help to alleviate most natural concerns about event study-based estimates of the impact of LSAP1 on ten-
omitted variable bias. Second, I address the standard econo- year yields, which average nearly 100 bps. However, this
metric concerns that arise when estimating time series difference makes sense because intermediary risk-bearing
forecasting regressions. These econometric issues are not a capacity was limited at the time of LSAP1. Table 4 shows that
significant concern in this context. Finally, one could worry the implied effect is slightly larger for univariate regressions
that the results are somehow specific to the Barclays indices I using DUR_CNTRBMBS t or for multivariate regressions that
use throughout. The Internet Appendix shows that similar control for the term spread. For LSAP2 and LSAP3,
forecasting results obtain when using the Bank of America which were announced in late 2010 and late 2011, respec-
MBS indices instead of the Barclays indices. tively, the effects implied by my estimates are in line
with the estimated LSAP announcement effects in the prior
3.3.6. Comparison with prior estimates of the price impact of literature.
duration supply shocks In summary, the price impact of supply implied by my
I now compare the price-impact effects implied by my regressions is broadly in line with those found in the
regressions with estimates from the recent literature evaluat- recent event-study literature on LSAPs. To be clear, my
ing the Federal Reserve's Large-Scale Asset Purchase (LSAP) results do not show that LSAPs have lowered the duration
programs. While there are competing interpretations risk premium in bond markets or that they work at all.19
(Krishnamurthy and Vissing-Jorgensen, 2013), to draw this However, my evidence does provide an out-of-sample
comparison, I assume that LSAP's primarily work by impacting proof of concept for LSAP policies that seek to impact
term premia. To be clear, I am not arguing that mortgage market-wide term premia.
refinancing explains why LSAP policies do or do not work.
However, both mortgage refinancing waves and LSAPs are
sources of duration supply shocks that have the potential to 3.3.7. A decomposition of aggregate MBS duration
impact term premia. Thus, it is interesting to compare the I now decompose aggregate MBS duration. The
MBS
price impact of supply shocks from LSAPs and mortgage decomposition shows that DURt reflects aggregate refi-
refinancing waves. nancing incentives, prepayment burnout, and several
Different price-impact magnitudes for LSAP- and mort- other factors. I then show that both aggregate refinancing
gage refinancing-induced duration shocks are expected for incentives and these other terms appear to contribute to
MBS
several reasons. First, investor risk-bearing capacity likely the forecasting power of DURt .
varies over time, so larger effects for LSAPs should be Let yM;t denote the primary mortgage rate, and take
expected to the extent they occurred when risk-bearing first-order expansion of the duration of MBS class i (e.g.,
capacity was low. Further, the LSAP-induced shocks would MBS backed by Fannie Mae guaranteed 30-year mortgages
have been expected to last longer than mortgage duration with a 6% coupon that were originated three years ago)
shocks, suggesting that LSAPs should have a larger impact on about its coupon cit:
yields due to their more persistent effect on term premia.
However, LSAPs were arguably far more advertised than DURit ¼ Dit  CONV it  ðyM;t cit Þ þϵit ; ð18Þ
mortgage duration shocks, suggesting that mortgage shocks
could have a larger impact than LSAPs if, as in Duffie (2010), where ϵit is the approximation error. Letting X t ¼ ∑i wit X it
investors were more inattentive in the former case. Thus, it is and st ½X it ; Y it  ¼ ∑i wit ðX it  X t ÞðY it  Y t Þ denote the market-
not obvious whether larger effects should be expected for weighted average and covariance, respectively, aggregate
MBS duration shocks or for LSAPs. MBS duration is given by
Table 4 compares the estimated effect of supply shocks
measured in ten-year Treasury duration equivalents. DURt ¼ Dt þCONV t  ðct  yM;t Þ þst ½CONV it ; cit  þ ϵt : ð19Þ
Table 4 provides more detail on the LSAPs, estimates of
their impact on ten-year yields from the prior literature, 18
I obtain this estimate by regressing the deviation of
and the estimated effects implied by the MBS duration MBS
Q MBS
t ðDURMBS
t =DURT10
t Þ from trend (in constant 2012 dollar) on DURt
effect. I use my estimates to predict the effect of LSAP from 2008 to the present.
announcements on ten-year Treasury yields. Specifically, 19
Besides moving the duration risk premium, LSAP policies could impact
using specifications of the form ytð10Þ  ytð10Þ þ1 ¼ aþb 
yields throughout the bond market through a variety of distinct channels.
DURt þ εð10Þ
t þ 1 as in Fig. 5, I estimate that a 1 standard
This is a lively area of scholarly debate. See Krishnamurthy and Vissing-
MBS Jorgensen (2011, 2013). Furthermore, LSAPs could have local effects. Compar-
deviation decline in DURt has lowered ten-year Treasury ing securities with similar durations, D'Amico, English, Lopez-Salido, and
yields by 36 bps over my 1989–present sample. Based on Nelson (2012) and Cahill, D'Amico, Li, and Sears (2013) find larger intraday
the relation between DURMBS t and MBS ten-year declines in yields on those securities that the Federal Reserve is purchasing.
 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299 287

Table 4
Comparison with literature on the Federal Reserve's Large-Scale Asset Purchase (LSAP) programs.
This table compares the implied price impact from my regressions using mortgage-backed securities (MBS) duration with that estimated by papers
examining the Fed's LSAP programs. I obtain the implied effect for each LSAP as
 
Implied Effect in basis points ðbpsÞ ¼ Effect of 1s move in DURt on yð10Þ in bps
LSAP size in ten  year Treasury duration equivalents

ten  year Treasury duration equivalents corresponding to 1s move in DURt
Li and Wei (2013) estimate that the $300 billion of Treasury purchases associated with LSAP1 reduced duration supply by $169 billion ¼ (4.5  300)/8 billion ten-
year Treasury equivalents. Thus, accounting for the purchases of $200 billion of agency debt and $1,250 billion MBS, I estimate that LSAP1 reduced the aggregate
supply of duration by $750 billion¼ (4.5  300þ4.5  200þ 3  1250)/8 billion ten-year Treasury equivalents. Gagnon, Raskin, Remache, and Sack (2011) examine
movements in yields on eight LSAP1 announcement dates and estimate that LSAP1 reduced ten-year yields by 91 bps. Applying the coefficients from time-series
regressions, Gagnon, Raskin, Remache, and Sack (2011) estimate an impact of LSAP1 of 61 bps. Krishnamurthy and Vissing-Jorgensen (2011) examine movement in
yields on five LSAP1 announcement dates and estimate that LSAP1 reduced ten-year yields by 107 bps. Using a no arbitrage model, Li and Wei (2013) estimate that
LSAP1 lowered yields by 99 bps. Jarrow and Li (2012) also estimate a no arbitrage model and estimate that LSAP1 and LSAP2 lowered ten-year yields by 70 bps.
Assuming that roughly 80% of this corresponds to LSAP1, this implies that LSAP1 lowered yields by 56 bps.
Li and Wei (2013) estimate that LSAP2 reduced the supply of duration by $400 billion ten-year Treasury equivalents. Krishnamurthy and Vissing-Jorgensen (2011)
examine movement in yields on three LSAP2 announcement dates and estimate that LSAP2 reduced ten-year yields by 30 bps. Using a no-arbitrage model, Li and
Wei (2013) estimate that LSAP2 lowered yields by 26 bps. Jarrow and Li (2012) also estimate a no-arbitrage model and estimate that LSAP1 and LSAP2 lowered ten-
year yields by 70 bps. Assuming roughly 20% of Jarrow and Li's (2012) estimate corresponds to LSAP2, this implies that LSAP2 lowered yields by 14 bps.
LSAP3 is often referred to as the Maturity Extension Program. Li and Wei (2013) estimate that LSAP3 reduced the supply of duration by $400 billion ten-year
Treasury equivalents. Li and Wei (2013) estimate that LSAP3 lowered ten-year yields by 25 bps. Using regressions similar to those in Greenwood and Vayanos (2014),
Hamilton and Wu (2012) estimate that LSAP3 lowered ten-year yields by 17 bps.

Estimated effect on ten-year yield (bps)

Supply LSAP evaluation studies Effect implied by MBS supply shocks

shock
Event description Ten-year Average Minimum Maximum DURMBS
t DUR_CNTRBMBS
t
Treasury study study study
duration
equivalents
Univariate Multivariate Univariate Multivariate

LSAP1 Purchase of $300billion $750 83 55 107 54 68 69 79

Treasuries, $200 billion billion
Agencies, $1,250 billion
MBS.Announced 11/2008-
3/2009

LSAP2 Purchase of $600 billion of $400 24 15 30 29 36 37 42

long-term Treasuries. billion
Announced 11/2010

LSAP3 Buy $400 billion long-term $400 21 17 25 29 36 37 42

Treasuries and sell $400 billion
billion short-term
Treasuries. Announced
9/2011

Effect of 1s move 36 46 40 46
in DURt on yð10Þ
in bps

Ten-year Treasury 503 503 436 436

duration
equivalents
corresponding
to 1 s move in
DURt

There are two main terms20: term can be further decomposed using deviations from
time series average of Dt and CONV t :
1. Aggregate refinancing incentives: The first term depends Dt þ CONV t  ðct  yM;t Þ ¼ ½D þCONV  ðct yM;t Þ
on the aggregate refinancing incentive, ðct  yM;t Þ. This
þ ½ðDt  DÞ þðCONV t  CONVÞ  ðct  yM;t Þ: ð20Þ

20 Because Dt and CONVt likely vary only slowly over time,

Further insight into the residual ϵt can be gained using a second-order
Taylor expansion of DURit. This suggests that aggregate MBS duration is this suggests that ðct  yM;t Þ should be a strong linear
decreasing in the cross-sectional variation in mortgage outstanding coupons. predictor of excess returns.
288 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299

2. Aggregate prepayment burnout: st ½CONV it ; cit  measures such as aggregate prepayment burnout, that are not
the extent to which low callability is concentrated in reflected in the simpler measures.
high coupon MBS. Aggregate MBS duration is high
when aggregate burnout is high. For instance,
some analysts argue that burnout has been elevated 3.3.8. The growing impact of MBS duration on term premia
in recent years because borrowers who are still in Have shifts in MBS duration become a more or less
higher coupon mortgages cannot refinance due to important driver of term premia over time? My story
negative home equity or tighter mortgage underwriting suggests that these effects should have grown far more
standards. important. The rapid growth of the MBS market means
that these duration shocks have become far more
This works largely as one might expect in the data. significant relative to the risk tolerance of bond investors.
Specifically, a strong negative relation exists between The significant growth of the MBS market relative
MBS
DURt and ðct  yM;t Þ: to the rest of the bond market is illustrated in Panel A of
Fig. 6.
DURMBS
t ¼ 3:095 0:952  ðct  yM;t Þ; R2 ¼ 0:48: ð21Þ
½18:22 ½  5:56 Fortunately, I can investigate the evolution of these
effects by examining data prior to 1989. Although I have
Adding a time trend (t) and interactions terms, I obtain
data on the model-based effective duration of MBS begin-
DURMBS
t ¼ 3:610 0:835  ðct  yM;t Þ 0:276  ðct  yM;t Þ  t ning only in 1989, Barclays publishes a cruder duration
½16:76 ½  5:18 ½  0:58
measure called Macaulay duration-to-worst dating back to
1:041  t; R2 ¼ 0:58: ð22Þ 1976.21 I use dur to denote the Macaulay analogs of the
½  2:01
more commonly used modified duration measures (DUR).
Thus, MBS duration has trended down by a little Thus, going back to 1976, I can ask whether dur t
MBS
and
more than one year since 1989, and weak evidence MBS MBS
dur_cntrbt ¼ ðMV MBS
t =MV AGG
t Þdur t have predictive
exists that CONVt has become slightly more negative
power for excess bond returns. While the model-based
over time.
effective duration measures provide a more accurate
Consistent with the previous decomposition, Columns
measure of the sensitivity of MBS prices to changes in
1–3 of Table 5 show that the aggregate refinancing
yields, Macaulay duration-to-worst contains similar infor-
incentive, ct  yM;t , is a strong negative predictor of excess
mation in the time series. These two measures are shown
bond returns. My measure of aggregate refinancing incen-
in Panel B of Fig. 6. For instance, from 1989 to the present
tives is the weighted-average coupon on MBS in the
when both measures are available, the correlation
Barclays index minus the current 30-year primary mort-
between effective duration and Macaulay duration-to-
gage rate from Freddie Mac. The forecasting power of this
worst of MBS is 0.91 in levels and 0.91 in 12-month
variable derives from comparing the primary mortgage
changes. I can also construct my measure of aggregate
rate (a variable that is almost a perfect linear combination
refinancing incentives, i.e., the difference between the
of forwards) with the average coupon on outstanding MBS
average MBS coupon and current primary mortgage rates
(a variable not spanned by forwards). Naturally, ct is itself
ðct  yM;t Þ going back to 1976.
a very slow moving variable that reflects historical path
Because Panel A of Fig. 6 shows that the MBS market
dependencies. Columns 7–9 of Table 5 show that I also
grew significantly relative to the rest of the bond market
obtain similar results if I forecast returns using a normal-
from 1976 to the early 1990s, examining the forecasting
ized version of the Mortgage Banking Association's refi-
power of MBS duration over this early period is useful.
nancing index, REFIt. Specifically, the Hodrick and Prescott
Table 6 shows subsample forecasting results. I find that the
(1997) filtered version of the log-refinancing index is a
three measures of MBS duration have no predictive power
strong negative predictor of returns. Thus, Table 5 shows
for excess bond returns over the 1976–1988 subsample,
that my main findings are not an artifact of using MBS
but they have significant predictive power from 1989 to
duration, an admittedly complex model-based construct,
the present (including both the 1989–1999 and 2000þ
and supports the broader hypothesis that aggregate mort-
subsamples). Fig. 7 makes a similar point and shows the
gage refinancing activity plays a key role in driving bond
results from 60-month (20-quarter) rolling regressions of
risk premia.
three-month excess returns on MBS duration measures,
Finally, Columns 4–6 and 10–12 of Table 5 show
controlling for the term spread. The results show an
forecasting horse races between my preferred measure of
unmistakable upward trend in the predictive power of
MBS duration ðDUR_CNTRBMBS t Þ and refinancing incentives
MBS duration. Over time, the coefficients have grown and
ðct  yM;t Þ and refinancing activity (REFIt), respectively.
are being estimated more precisely, so the t-statistics have
Consistent with the previous decomposition, these horse
grown. The fact that these effects have grown so much
races suggest that much of the information that
DUR_CNTRBMBS t contains about future excess returns is also
contained in ct yM;t (and REFIt). However, the horse races 21
Macaulay duration-to-worst is the Macaulay duration correspond-
suggest that model-based MBS duration estimates do ing to the yield-to-worst. Specifically, one assumes a deterministic set of
contain additional information above and beyond these future MBS cash flows based on an assumed prepayment profile (typi-
cally the expected prepayment profile, which depends on current interest
simpler measures. This is not surprising as MBS duration rates). The yield-to-worst and Macaulay duration-to-worst are then
reflects the forward-looking supply expectations of market simply the yield and Macaulay duration assuming this set of future bond
participants. Furthermore, DUR_CNTRBMBS t captures effects, cash flows.
 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299 289

Table 5
Horseraces between mortgage-backed securities (MBS) duration and mortgage refinancing measures.
Regressions of 12-month excess returns on ten-year Treasuries on the effective duration contribution of the Barclays MBS Index and either aggregate
refinancing incentives ðct  yM;t Þ computed as the difference between the average MBS coupon and the Freddie Mac primary market 30-year mortgage rate
or the log-deviation of the Mortgage Bankers' Association refinancing index from its Hodrick and Prescott (1997) trend ðREFI t Þ:

0
rxtð10Þ MBS
þ 1 ¼ a þ b  DUR_CNTRBt þ c  REFI t þ d xt þ εð10Þ
t þ 1:

The regressions are estimated with monthly data from 1989m1–2011m4, so each month I forecast the excess return over the following 12 months. To deal
with the overlapping nature of returns, t-statistics are based on Newey and West (1987) standard errors allowing for serial correlation at up to 18 lags.
I estimate these regressions with and without other forecasting variables identified in the literature on bond risk premia. Specifically, I control for the term
spread following Campbell and Shiller (1991) and the first five forward rates following Cochrane and Piazzesi (2005).

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)

ct  yM;t  2.571  7.766  6.867 2.444  4.702  5.158

[  1.26] [  4.46] [  3.29] [0.72] [  1.44] [  1.51]

REFI t  4.670  6.772  6.027  2.264  4.448  4.432

[  3.33] [  6.31] [  4.23] [  1.12] [  2.87] [  2.59]

DUR_CNTRBMBS
t 15.470 7.686 4.624 7.682 7.378 5.551
[2.72] [1.90] [1.17] [2.04] [2.83] [2.37]

ytð10Þ  yð1Þ
t 4.537 4.038 3.317 3.298
[4.70] [3.92] [4.55] [4.65]

ð1Þ
ft  26.123  23.254  17.055  14.383
[  3.53] [  3.91] [2.27] [  1.95]

ð2Þ
ft 131.748 113.968 94.061 74.597
[3.09] [3.50] [2.43] [2.03]

ð3Þ
ft  318.861  274.439  246.166  196.714
[  3.34] [  3.78] [2.85] [  2.42]

ð4Þ
ft 347.721 299.820 281.555 228.361
[3.59] [4.00] [3.13] [2.66]

ð5Þ
ft  134.407  116.017  112.104  91.553
[  3.61] [  3.89] [3.15] [  2.64]

Constant 4.303  4.202  1.734  11.061  10.900  6.056 5.273  0.210 0.632  2.978  8.102  5.626
[2.73] [  1.87] [  0.32] [  1.94] [  3.12] [  1.13] [4.17] [0.13] [0.13] [0.73] [2.27] [  0.95]

Number of observations 268 268 268 268 268 268 256 256 256 256 256 256
R2 0.05 0.42 0.50 0.19 0.45 0.51 0.17 0.42 0.51 0.21 0.46 0.52

stronger over time, precisely as I would predict, provides on coupon-bearing long-term foreign government bonds
further indirect evidence for my MBS hedging story. over short-term foreign bond using the Shiller, Campbell,
and Schoenholtz (1983) approximation.22 I compute for-
3.3.9. Predicting foreign bond returns eign government bond returns for ten developed coun-
Do shifts in the duration of US mortgage-backed secu- tries: Belgium, Canada, Denmark, France, Germany, Italy,
rities impact term premia in foreign bond markets? This Japan, Sweden, Switzerland, and the UK. I then estimate
question is addressed in Table 7, which investigates whether univariate forecasting regressions of the form
the duration of US MBS forecasts the excess returns on long-
rxFORðnÞ FORðnÞ
c;t þ 1 ¼ a þ b  DURt þ εc;t þ 1 ð23Þ
term foreign bonds. What one should expect to find here
depends on the degree of integration between the bond
markets in different countries. At one extreme, if foreign and multivariate regressions controlling for the US term
bond markets were completely segmented from the US spread and foreign country term spread
bond market, US MBS duration would be expected to have rxFORðnÞ USðnÞ
 yUSð1Þ
c;t þ 1 ¼ a þ b  DURt þ c  ðyc;t t Þ
no forecasting power for foreign excess bond returns. At the
other extreme, if there were a single integrated global bond þ d  ðyFORðnÞ
c;t  ytFORð1Þ Þ þ εc;t
FORðnÞ
þ1 : ð24Þ
market, term premia would be expected to move in lock-
step across national bond markets. In between these two
extremes, one would expect US MBS duration to have some
22
predictive power for excess foreign bond returns, albeit less This approximation says that log excess return on an n-year coupon
bond over the short rate ðyð1Þ ðnÞ ðnÞ ðnÞ ðnÞ ðn  1Þ ð1Þ
t Þ is rxc;t þ 1  Dt yc;t  ðDt  1Þyc;t þ 1  yt ;
than for US excess bond returns. where Y ðnÞ ðnÞ ðnÞ
c;t is the yield on an n-year coupon bond, yc;t ¼ lnð1þ Y c;t Þ, and
Following Campbell (1999, 2003), I work with Interna- DðnÞ ðnÞ  n
t ¼ ½1  ð1 þ Y c;t Þ =½1  ð1þ Y ðnÞ
c;t Þ
1
 is the approximate duration on an
tional Financial Statistics data and compute excess returns n-year coupon bond.
290 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299

results provides additional comfort that the results do not

60
reflect an omitted global risk factor.

3.4. Excess sensitivity of long-term yields and forwards to

40 short rates
%

The model suggests that distant forward rates should be

20 excessively sensitive to movements in short-term interest
rates (see Proposition 2). As shown in Hanson and Stein
(2012), this general excess sensitivity result emerges very
0 strongly in the data. This is consistent with Gürkaynak,
Sack, and Swanson (2005), who find that distant forwards
80 2

84 2

88 2

92 2
94 2

06 2
78 2

82 2

86 2

90 2

98 2

02 2

08 2

12 2
96 2

00 2

04 2

10 2

12
19 m1
19 m1

19 m1
19 m1
19 m1

19 m1
19 m1

19 m1

19 m1

20 m1
19 m1
19 m1
20 m1
20 m1

20 m1
20 m1

20 m1
20 m1
are excessively sensitive to macroeconomic announce-

m
76
19

ments. And, as emphasized by Hanson and Stein (2012),

MBS % of Aggregate Index MBS % of MBS+Treasuries one observes significant excess sensitivity for distant real
forward rates, i.e., forwards extracted from the Treasury
Inflation Protected Securities (TIPS) yield curve. While this
8 is puzzling from the standpoint of textbook macro-finance
theories, excess sensitivity follows naturally from models
6
that emphasize supply and demand effects in bonds mar-
kets. For instance, an extension in MBS duration increases
the exposure of investors to movements in short-term real
Years

4 rates, necessitating a rise in the real term premium.

Going further, the model predicts that distant forwards
2 should be particularly sensitive to short rates when the MBS
market is more negatively convex (see Proposition 3). When
the market is more negatively convex, a given change in
0
short rates has a larger effect on the total quantity of interest
rate risk and, hence, a larger impact on distant forward term
82 2
84 2

90 2
92 2
94 2

98 2
00 2
02 2

08 2
10 2
78 2
80 2

86 2
88 2

96 2

04 2
06 2

12 2
12
19 m1
19 m1

19 m1
19 m1
19 m1

20 m1
19 m1
19 m1

19 m1
19 m1

19 m1
19 m1
20 m1
20 m1

20 m1
20 m1
20 m1

20 m1
m
76

premia. To test this prediction, I examine whether the high-

19

MBS effective duration MBS Macaulay duration-to-worst frequency sensitivity of distant forwards varies with MBS
convexity. In Table 8, I estimate daily regressions of the form
Fig. 6. Growth of mortgage-backed securities (MBS) market and MBS
¼ a þ b  Δyt$ð2Þ þ c  NCONV MBS
Xð10Þ
duration measures. Panel A shows the market value of the Barclays MBS Δf t t
Index as a fraction of the market value of the Barclays Aggregate Index (a
proxy for broad the investment-grade bond market) from 1976–2012 and as þ d  ðΔyt$ð2Þ  NCONV MBS
t Þ þΔεXð10Þ
t ; ð25Þ
a fraction of the sum of the value of the Barclays MBS and Treasury indices.
where NCONV MBS t ¼ CONV MBS
t (negative one times the
Panel B compares the effective duration of the Barclays MBS Index from
1989–2012 to the Macaulay duration-to-worst of the index from 1976–2012. convexity of the Barclays MBS Index) for X ¼ $ and TIPS;
Panel A: The growth of the MBS market. Panel B: MBS duration measures. i.e., I examine changes in both ten-year nominal and ten-year
real forwards. I estimate these regressions from 1999 to the
To assess the differential forecasting power for foreign present to include data on real forwards from Gürkaynak,
bond excess returns as compared with US domestic bond Sack, and Wright (2010).23 The theory predicts that d 40.
returns, for each country I re-run the regressions with Excess sensitivity should be more pronounced when the
FORðnÞ USðnÞ MBS market is more negatively convex.
rxc;t þ 1  rxc;t þ 1 on the left-hand side. I then report the
coefficient on DURt from this specification as well as the Consistent with this prediction, Table 8 shows that both
associated t-statistic, labeled as “Difference from USA” in distant nominal and real forwards are more sensitive to
MBS
the table. Panel A shows results for DURt , and Panel B movements in short-term nominal rates when the MBS
MBS
shows results for DUR_CNTRBt . MBS market is more negatively convex. Because NCONVt is
Table 7 shows that evidence exists of moderate cross- persistent at daily frequencies, t-statistics are based on
country spillovers. US MBS duration has some limited Newey and West (1987) standard errors allowing for serial
ability to forecast excess returns on long-term government correlation up to 20 business days. From 1999 to the present,
MBS
bonds in Canada, Denmark, Japan, and the UK. However, in NCONVt has averaged 1.528 with a standard deviation of
each case, the magnitude of the forecasting relations for 0.531. Thus, on average, a 100 bps increase in the two-year
foreign bonds is much weaker, both economically and nominal yield has been associated with a 54 bps [¼0.322þ
statistically, than the corresponding magnitude for US 0.140  1.528] increase in ten-year nominal forwards and a
bonds. In short, the evidence is consistent with the view 30 bps increase in ten-year real forwards. However, when
MBS
that the bond markets of, say, the US and the UK are NCONVt is 2 standard deviations above average, the same
partially integrated, so a duration supply shock in the US 100 bps increase in the nominal short rates has been
has some impact on bond risk premia in the UK. At the
same time, the fact that the US return forecasting results 23
The convexity of Barclays MBS Index is available for 1997–present.
are so much stronger than the foreign return forecasting I obtain similar results for nominal forwards for this period.
 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299 291

Table 6
Forecasting excess bond returns, subsample results.
Regressions of 12-month excess returns on ten-year Treasuries on the effective duration of the MBS Index, the effective duration contribution of the MBS
index, and aggregate refinancing incentives

rxtð10Þ 0 ð10Þ
þ 1 ¼ a þ b  DURt þ c xt þ εt þ 1 :
MBS
For mortgage-backed securities (MBS) duration, DURMBSt (effective) refers to option-adjusted effective duration which is available beginning in 1989. durt
(Macaulay) refers to Macaulay duration-to-worst which is available beginning in 1976. A similar notation applies in the case of MBS duration contribution.
Aggregate refinancing incentives ðct  yM;t Þ, computed as the difference between the average MBS coupon in the Barclays Index and the Freddie Mac
primary market 30-year mortgage rate, is also available beginning in 1976. The regressions are estimated with monthly data from 1976m1–2011m4, so
each month I forecast the excess return over the following 12 months. To deal with the overlapping nature of returns, t-statistics are based on Newey and
West (1987) standard errors allowing for serial correlation at up to 18 lags. I estimate these regressions with and without controlling for the term-spreads.

(1) (2) (3) (4) (5) (6) (7) (8)

1989 þ 1989–1999 2000þ 1989 þ 1976þ 1976–1988 1989–1999 2000þ

Panel A: Subsample results for MBS duration

DURMBS
t (effective) 4.437 5.324 4.355
[6.48] [3.33] [9.44]

MBS
dur t (Macaulay) 5.968 1.505  0.962 10.198 4.383
[4.57] [0.76] [  0.27] [3.05] [5.14]

ytð10Þ  yð1Þ
t 3.231 3.062 2.938 2.935 3.481 4.516 3.278 2.333
[4.67] [1.93] [4.30] [4.23] [3.34] [1.76] [2.44] [2.92]

Constant  14.948  19.279  12.830  27.100  8.601 2.063  48.689  17.803

[  4.58] [  2.65] [  4.73] [  4.08] [  0.93] [0.11] [  2.99] [  4.01]

Number of observations 268 132 136 268 424 156 132 136
2
R 0.37 0.34 0.49 0.34 0.15 0.13 0.41 0.37

Panel B: Subsample results for MBS duration contribution

DUR_CNTRBMBS (effective) 14.629 19.993 11.886

t
[5.78] [3.86] [8.64]

MBS
dur_cntrbt (Macaulay) 14.551 6.733 9.220 33.513 11.891
[3.80] [2.18] [1.71] [2.96] [5.76]

ytð10Þ  yð1Þ
t 2.999 3.004 2.837 2.319 2.579 3.756 3.140 2.218
[4.57] [1.98] [4.07] [3.52] [2.64] [1.58] [2.84] [2.89]

Constant  15.515  21.018  12.307  20.371  8.244  9.578  46.573  16.979

[  4.23] [  3.04] [  4.42] [  3.39] [  1.63] [  1.84] [  3.07] [  4.22]

Number of observations 268 132 136 268 424 156 132 136

R2 0.40 0.38 0.47 0.32 0.20 0.15 0.42 0.36

Panel C: Subsample results for mortgage refinancing incentive ðct  yM;t Þ

ct  yM;t  7.766  1.684  3.455  9.119  6.717

[  4.46] [  1.50] [  2.91] [  4.24] [  3.78]

ytð10Þ  yð1Þ
t 4.537 4.531 8.205 6.066 3.828
[4.70] [3.94] [3.87] [3.54] [4.14]

Constant  4.202  4.479  15.428  5.710  3.209

[  1.87] [  1.79] [  5.32] [  1.70] [  1.21]

Number of observations 268 424 156 132 136

R2 0.42 0.19 0.30 0.48 0.33

associated with a 68 bps increase in ten-year nominal movements in short-term nominal yields or short-term
$ð2Þ
forwards and a 42 bps increase in ten-year real forwards. forwards [if I replace Δyt$ð2Þ with Δf t in Eq. (25)].
Thus, variation in MBS convexity has been associated with
meaningful variation in the sensitivity of distant forwards to 3.5. Excess volatility of long-term interest rates
movements in short rates. Furthermore, this is true whether I
MBS
look at NCONVt or the scaled version of this variable, The model suggests that, all else being equal, long-term
MBS
NCONV t  ðMV MBSt =MV AGG
t Þ; or I examine sensitivity to interest rate volatility is elevated when the MBS market is
292 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299

Return forecasting coefficients Return forecasting t-statistic

20 10

10 5
Coefficient (%)

t-statistic
0
0

-10
-5
82

88

94

10

88

94

98

04
98

04

84
80

84
86

90
92

96

00
02

06
08

12

80
82

86

90
92

96

00
02

06

10
12
08
19

19
19

19
19

19

20
20

20
20

20

19
19

19

19
19

19

20
20

20

20
20
20
19

19

19

20

19

19

19

20
19

20

19
Return forecasting coefficients Return forecasting t-statistic
50 10

0 5
Coefficient (%)

t-statistic

0
-50

-5
-100
88
90
92

06
08

88

94

98

04
84
80
82
84
86

94
96
98
00
02
04

12

80
82

86

90
92

96

00
02

06

10
12
10

08
19
19
19
19

19
19
19
20
20
20

20

19
19

19

19
19

19

20
20

20

20
20
20

20
19
19
19

20
20

19

19

19

20
19

Return forecasting coefficients Return forecasting t-statistic

5 2

0
Coefficient (%)

-2
t-statistic

-4
-5

-6

-10
-8
82

88

94

10

88

94

98

04
98

04

84
80

84
86

90
92

96

00
02

06
08

12

80
82

86

90
92

96

00
02

06

10
12
08
19

19
19

19
19

19

20
20

20
20

20

19
19

19

19
19

19

20
20

20

20
20
20
19

19

19

20

19

19

19

20
19

20

19

Fig. 7. Coefficients from 60-month rolling excess return forecasting regressions, 1980m12–2012m1. This figure depicts results from 60-month rolling
regressions of three-month excess bond returns on mortgage-backed securities (MBS) duration measures, controlling for the term spread

rxð10Þ ð10Þ
t þ 1=4 ¼ a þ b  dur t þ c  ðyt  yð1Þ ð10Þ
t Þ þ εt þ 1=4 :
MBS MB
Panel A shows results for MBS Macaulay duration ðdur ¼ dur Þ, Panel B shows results for MBS Macaulay duration contribution ðdur ¼ dur_cntrb Þ, and
Panel C shows results for aggregate refinancing incentives ðdur ¼ c  yM Þ. The left-hand graph in each Panel shows the estimated regression coefficients on
dur along with confidence intervals as dashed lines. The right-hand graph shows the associated t-statistics. I compute standard errors using Newey and
MBS
West (1987) standard errors allowing for serial correlation at up to six months. Panel A: MBS Macaulay duration ðdur ¼ dur Þ. Panel B: MBS Macaulay
MB
duration contribution ðdur ¼ dur_cntrb Þ. Panel C: Aggregate refinancing incentives ðdur ¼ c  yM Þ.

more negatively convex. Intuitively, the positive-feedback To test this prediction, I follow the approach of Perli
dynamic underlying MBS duration spirals is stronger when and Sack (2003), who relate weekly observations of
the mortgage market is on a refinancing cliff; i.e., when a option-implied interest rate variance to measures of MBS
small movement in rates significantly impacts aggregate convexity. Specifically, Perli and Sack (2003) assume that
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
refinancing behavior. Δyt ¼ 1 þ βxt  εt , where s2ε;t follows an ARð1Þ process
Table 7
Using US mortgage-backed securities (MBS) duration to forecast foreign bond excess returns.
This tables presents univariate forecasting regressions of the form

rxFORðnÞ FORðnÞ
c;t þ 1 ¼ a þ b  DURt þ εc;t þ 1 ;

and multivariate regressions of the form

rxFORðnÞ USðnÞ
c;t þ 1 ¼ a þ b  DURt þ c  ðyc;t  yUSð1Þ
t Þþ d  ðyFORðnÞ
c;t  yFORð1Þ
t Þ þ εFORðnÞ
c;t þ 1 :

Using International Financial Statistics data on long-term government bond yields and short-term interest rates, I compute the excess returns on long-term government bond returns for the ten developed countries:
Belgium, Canada, Denmark, France, Germany, Italy, Japan, Sweden, Switzerland, and the UK. Because the long-term yields are for coupon-bearing bonds, I follow Campbell (1999, 2003) and use the Shiller, Campbell, and
Schoenholtz (1983) approximation to compute excess returns over short-term government bills (when available) or money market rates (when bills are not available). The regressions are estimated with monthly data from
1989m1–2011m4. To deal with the overlapping nature of the regressions, t-statistics are based on Newey and West (1987) standard errors allowing for serial correlation at up to 18 lags. To assess the differential forecasting
power between foreign bond and domestic bond excess returns, for each country I rerun the regressions with rxFORðnÞ USðnÞ
c;t þ 1  rxc;t þ 1 on the left-hand side. I then report the coefficient on DURt from this specification as well as the

associated t-statistic, labeled as “Δ from USA.” Panel A shows results for DURMBS
t , and Panel B shows results for DUR_CNTRBMBS
t .

S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299

USA Belgium Canada Denmark France Germany Italy Japan Sweden Switzerland UK

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18) (19) (20) (21) (22)

Panel A: Forecasting foreign bond excess returns using DURMBS

t

DURMBS 2.42 2.91 0.44 0.77 0.69 1.90 0.83 1.06 0.21 0.74 0.20 0.58 0.44 0.18 1.51 2.13 1.03 1.41 0.14 1.37 0.92 1.50
t
[3.53] [5.81] [0.52] [1.44] [0.86] [4.36] [1.19] [1.78] [0.31] [1.59] [0.23] [0.88] [  0.56] [  0.20] [1.21] [2.49] [1.07] [1.73] [  0.15] [2.85] [0.99] [1.99]

yUSðnÞ
t  yUSð1Þ
t
1.94 1.37 0.56 1.63 1.52 1.93 1.04 2.10 1.38 2.12 1.49
[3.13] [1.80] [0.88] [2.28] [2.00] [2.75] [0.86] [2.74] [1.56] [3.19] [2.12]

ytFORðLÞ  yFORðSÞ 1.74 1.79 1.35 1.05 1.84 1.03 2.15 1.39 1.80 0.79
t
[2.21] [2.58] [2.29] [1.53] [2.51] [0.79] [0.98] [1.95] [3.31] [1.18]

Constant  4.64 9.62 1.11  4.82 1.19  6.63 0.00  4.59 1.81 3.60 1.83  4.82 4.05 2.52  2.75  10.59 0.09  5.18 2.73  7.72  0.53  5.45
[ 1.74] [ 3.80] [0.40] [  2.16] [0.48] [ 4.17] [0.00] [ 1.82] [0.80] [  1.62] [0.69] [  2.00] [1.58] [0.71] [ 0.78] [  3.95] [  0.03] [ 1.84] [0.98] [ 3.88] [ 0.19] [ 1.77]

Observations 268 268 268 268 268 268 268 268 268 268 268 268 268 268 268 268 268 268 268 268 268 268

R2 0.14 0.29 0.00 0.26 0.02 0.39 0.01 0.22 0.00 0.17 0.00 0.30 0.00 0.03 0.05 0.25 0.01 0.11 0.00 0.37 0.02 0.22

Δ from USA  1.98 2.10 1.72  1.07  1.59 1.79 2.21 2.22  2.22  2.30 2.86  3.12 0.91  0.79  1.39  1.51  2.56 1.74  1.50  1.40
[ 3.73] [  4.88] [ 3.73] [  2.66] [  3.35] [ 3.40] [  4.65] [ 4.98] [ 3.70] [ 4.28] [  3.09] [  3.46] [ 0.87] [  0.88] [  1.93] [ 2.13] [  3.30] [ 2.78] [ 1.85] [  1.89]

Panel B: Forecasting foreign bond excess returns using DUR_CNTRBMBS

t

DURCNTRBMBS 8.76 10.05 2.14 2.27 2.82 5.31 3.53 3.61 1.88 2.56 2.69 3.22 2.14 1.10 5.57 6.32 4.31 4.49 2.15 4.46 4.07 5.39
t
[4.31] [5.70] [0.98] [1.44] [1.45] [4.00] [1.45] [2.03] [0.96] [1.50] [1.34] [1.78] [  0.94] [ 0.44] [2.77] [2.84] [1.69] [1.86] [1.10] [ 3.05] [1.76] [2.37]

ytUSðLÞ  yUSðSÞ
t
1.90 1.36 0.69 1.61 1.52 1.97 1.01 2.01 1.36 2.13 1.55
[3.43] [1.74] [0.89] [2.29] [1.98] [2.76] [0.83] [2.53] [1.52] [3.17] [1.96]

ytFORðLÞ  yFORðSÞ 1.70 1.57 1.31 0.96 1.78 0.98 1.58 1.27 1.61 0.71
t
[2.07] [2.12] [2.12] [1.35] [2.36] [  0.75] [0.69] [1.66] [3.01] [1.06]

Constant  6.01 10.66 0.26  4.58 0.47  5.88  1.05  4.86 0.47 3.81 0.41  6.34 4.89 3.12 3.70  9.53  1.30  5.15 0.06  7.77  1.84  6.30
[  2.28] [ 4.05] [0.10] [  1.82] [0.22] [ 3.27] [  0.35] [ 1.83] [0.19] [  1.35] [ 0.18] [ 2.35] [1.58] [0.75] [ 1.79] [  3.03] [ 0.41] [ 1.45] [0.03] [ 3.84] [ 0.75] [  1.83]

Observations 268 268 268 268 268 268 268 268 268 268 268 268 268 268 268 268 268 268 268 268 268 268

R2 0.19 0.33 0.01 0.26 0.03 0.37 0.02 0.22 0.01 0.17 0.02 0.32 0.01 0.03 0.07 0.23 0.02 0.11 0.01 0.38 0.04 0.23

Δ from USA  6.63  7.48  5.94  4.66 5.23  6.19  6.88 7.42  6.07  6.69  10.91  11.57  3.20  3.96 4.45 5.70  6.61 5.56  4.70  4.72

293
[ 5.29] [ 5.17] [ 5.57] [  4.81] [ 4.06] [ 3.92] [  5.38] [ 4.80] [ 4.70] [ 4.09] [  3.88] [  3.65] [ 1.33] [ 1.46] [ 3.04] [ 2.99] [4.06] [3.12] [2.30] [  2.26]
294 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299

Table 8
Response of US long-term forwards to changes in short-term rates.
Daily regressions of changes in nominal and real forward rates on short-term nominal rates beginning in 1999, allowing for a differential response
depending on mortgage-backed securities (MBS) negative convexity. MBS negative convexity is simply negative one times MBS convexity. Specifically, I
estimate

¼ a þ b  Δy$ð2Þ þ d  ðΔy$ð2Þ
Xð10Þ
Δf t t þ c  NCONV MBS
t t  NCONV MBS
t Þ þ ΔεXð10Þ
t

and

Xð10Þ $ð2Þ $ð2Þ

Δf t ¼ a þ b  Δf t þ c  NCONV MBS
t þ d  ðΔf t  NCONV MBS
t Þ þ ΔεXð10Þ
t ;
for X ¼ $ and TIPS. t-statistics, based on Newey and West (1987) standard errors allowing for serial correlation at a lag of up to 20 business days, are shown
in brackets.

$ TIPS $ TIPS $ TIPS $ TIPS

(1) (2) (3) (4) (5) (6) (7) (8)

Δy$ð2Þ 0.322 0.124 0.475 0.142

[2.67] [1.86] [3.99] [1.98]

$ð2Þ
Δf 0.264 0.074 0.364 0.084
[3.33] [1.41] [4.97] [1.66]

NCONV MBS 0.000  0.001 0.000  0.001

[0.06] [  0.58] [0.13] [  0.65]

NCONV MBS  ðMV MBS =MV AGG Þ 0.002 0.000 0.001  0.001
[0.23] [0.06] [0.09] [  0.16]

Δy$ð2Þ  NCONV MBS 0.140 0.114

[1.74] [2.51]

Δy$ð2Þ  NCONV MBS  ðMV MBS =MV AGG Þ 0.107 0.278

[0.48] [1.96]

$ð2Þ
Δf  NCONV MBS 0.155 0.135
[2.94] [4.00]

$ð2Þ
Δf  NCONV MBS  ðMV MBS =MV AGG Þ 0.247 0.354
[1.77] [3.74]

Constant 0.000 0.000  0.001  0.001 0.000 0.000  0.001 0.000

[0.01] [0.19] [  0.27] [  0.28] [0.02] [0.19] [  0.16] [0.26]

R2 0.21 0.12 0.21 0.12 0.31 0.19 0.31 0.19

s2ε;t ¼ α0 þ α1 s2ε;t  1 þ ut , so that ten-year swaptions and examine the residuals from this
1 þ βxt 2 regression. As Perli and Sack (2003) argue, it is desirable to
s2Δy;t ¼ ð1 þβxt Þs2ε;t ¼ α0 ð1 þβxt Þ þ α1 s strip out structural or cyclical fluctuations in interest rate
1 þβxt  1 Δy;t  1
volatility in this way because MBS convexity is expected to
þð1 þ βxt Þut : ð26Þ
have fairly transient effects on implied volatility. Using the
There are three effects: a level effect, a persistence effect, resulting residualized measure, s~ 2Δy;t , I then estimate
and a volatility of volatility effect. However, if xt is fairly
persistent, then ð1 þ βxt Þ=ð1 þβxt  1 Þ  1. Furthermore, Perli s~ 2Δy;t  1 ¼ α0 ð1 þβ  NCONV t  1 Þ þ α1 s~ 2Δy;t  1 þ ut : ð28Þ
and Sack (2003) find that the volatility of volatility effect is
ignorable in practice. Thus, using xt ¼ NCONV t , I focus Data on MBS convexity are available beginning in 1997, so I
solely on the level effect estimate specification (28) using weekly data from 1997 to
the present. t-statistics are based on Newey and West (1987)
s2Δy;t ¼ α0 ð1 þ β  NCONV t Þ þ α1 s2Δy;t  1 þ ut : ð27Þ standard errors allowing for serial correlation at a lag of up to
A simultaneity problem arises here as NCONVt is a function 12 weeks.
of s2Δy;t . To deal with this, I substitute NCONVt-1 for NCONV t The results are shown in Table 9 and indicate that
in the above. option-implied yield variance is higher when the MBS
Following Perli and Sack (2003), I measure s2Δy;t using the market is more negatively convex. This conclusion holds
MBS
implied yield variance (the square of implied volatility) from whether I look at NCONVt or the scaled version of this
three-month by ten-year swaptions (i.e., the option to enter variable, NCONV MBS
t  ðMV MBS
t =MV AGG
t Þ, and when I control
into a ten-year swap at prespecified rate in three months). I for lagged swap yields in Eq. (28). The estimates in Column
first regress the implied variance from three-month by ten- 2 imply that a 1 standard deviation increase in NCONV
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
year swaptions on the implied variance from two-year by raises implied yield volatility by 0:10%½ ¼ 0:020  0:519.
 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299 295

Table 9
Implied variance of US long-term yields and mortgage-backed securities (MBS) negative convexity.
Weekly regressions of swaption-implied yield variance on MBS negative convexity beginning in 1997

s~ 2t ðΔyð10Þ Þ ¼ a þ b  s~ 2t  1 ðΔyð10Þ Þ þ c  ytð10Þ MBS

 1 þ d  NCONV t  1 þ ut :
MBS negative convexity is negative one times MBS convexity. I follow the empirical approach of Perli and Sack (2003). Specifically, the left-hand side
variable, s~ 2t ðΔyð10Þ Þ, is the residual from a regression of ten-year by three-month forward swaption-implied variance (the square of implied volatility) on
ten-year by two-year forward implied yield variance. t-statistics, based on Newey and West (1987) standard errors allowing for serial correlation at a lag of
up to 12 weeks, are shown in brackets.

(1) (2) (3) (4)

s~ 2t  1 ðΔyð10Þ Þ 0.941 0.941 0.939 0.939

[49.81] [50.40] [49.61] [49.86]

ytð10Þ
1 0.004 0.006
[1.07] [1.64]

NCONV MBS
t1 0.013 0.020
[2.14] [2.29]

NCONV MBS MBS AGG

t  1  ðMV t  1 =MV t  1 Þ 0.048 0.078
[2.14] [2.70]

Constant  0.021  0.051  0.026  0.072

[  2.28] [  1.62] [  2.42] [  2.31]

Number of observations 785 785 785 785

R2 0.89 0.89 0.89 0.89

The standard deviation of s~ Δy;t  1 is 0.48%, suggesting that reflect idiosyncratic shifts in the demand as in a flight to
movements in MBS convexity are associated with econom- quality episode (e.g., Duffee, 1996; Longstaff, 2004, etc.) or
ically significant shifts in interest rate volatility. idiosyncratic shifts in supply (e.g., Greenwood and Vayanos,
2010; and Lou, Yan, and Zhang, 2013).
3.6. Impact of MBS duration on corporate and swap spreads In my stylized model, the expected excess returns on
any long-term fixed income asset over short-term bonds is
In practice, MBS investors hedge the interest rate equal to a term premium earned by all long-term fixed
exposure of MBS using either Treasuries or interest rate income assets plus a risk premium that is specific to that
swaps. Regardless of the specific hedging instrument, asset. In particular, because long-term government bonds
other investors must take the other side of these trades. and swaps are both exposed to movements in the general
Thus, in the aggregate, fixed income investors must bear a level of interest rates, the impact of MBS duration on the
time-varying amount of interest rate risk, which impacts overall term premium is independent of the fraction of
equilibrium term premia. In a simple model in which MBS hedged with government bonds and swaps—just as in
Treasuries and swaps are perfect substitutes, one would the simpler model developed above.
not expect these hedging flows to impact spreads between However, MBS hedging flows now have an impact on
swaps and Treasuries. But different long-term fixed the spread between long-term fixed income assets. For
income assets with the same duration are not perfect instance, the government-specific risk premium is high
substitutes. Once we relax the assumption of perfect when MBS duration is high. Because MBS investors hedge
substitutability, hedging flows triggered by shifts in MBS with government bonds, the government-specific risk
duration could impact yield spreads between duration- premium must rise to induce arbitrageurs to hold more
matched fixed income assets and give rise to predictable Treasuries. Because these hedging flows have no impact on
variation in the excess returns on various fixed income the corporate-specific risk premium, they result in tighter
assets over duration-matched government bonds. than normal spreads between long-term corporate and
In the Internet Appendix, I consider a stylized model with government bonds. Furthermore, if a sufficiently large
three long-term fixed income assets: government bonds, volume of MBS hedging takes place in swap markets,
interest rate swaps, and high-grade corporate bonds. I and the idiosyncratic movements in swaps are large
suppose that some fraction of MBS duration is hedged using relative to the idiosyncratic movements in Treasuries, an
interest rate swaps and the remainder is hedged using increase in MBS duration would be expected to raise the
government bonds. While all long-term fixed income assets spread between swaps and Treasuries.
are exposed to interest rate risk, some components of returns In summary, the analysis suggests that high levels of
are specific to government bonds, swaps, and corporate MBS duration are associated with narrow current spreads
bonds. For corporate bonds, this could reflect changes in between corporate bonds and duration-matched Treasu-
credit risk. And, for all three assets, these components could ries, should forecast a future widening of corporate bond
be reflecting shifts in supply and demand for specific assets. spreads, and should predict low future returns on corpo-
For instance, in the case of government bonds, this could rate bonds relative to those on duration-matched
296 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299

Table 10
The impact on mortgage-backed securities (MBS) duration on corporate bond and swap spreads.
Panel A shows regressions of three-month changes in spreads over Treasuries on contemporaneous three-month changes in MBS duration:

ΔSPREADt ¼ a1 þ b1  ΔDUR_CNTRBMBS
t þ c01 Δxt þ vt :
Panel B uses the current level of MBS duration to forecast changes in spreads over the next 12 months:

ΔSPREADt þ 1 ¼ a2 þ b2  DUR_CNTRBMBS
t þ c02 xt þ ut þ 1 :
Panel C uses the current level of MBS duration to forecast excess returns over duration-matched Treasuries over the following 12 months:.

 MATCH
rxDUR
t þ1 ¼ a3 þ b3  DUR_CNTRBMBS
t þ c03 xt þ εt þ 1
All regressions begin in 1999m1. To deal with the overlapping nature of the regressions, t-statistics are based on Newey and West (1987) standard errors
allowing for serial correlation at up to six lags in Panel A and up to 18 lags in Panels B and C. In Panel A, I control for changes in the term spread, the Chicago
Board Options Exchange Market Volatility Index (VIX), and the past returns on the stock market, all of which Collin-Dufresne, Goldstein, and Martin (2001)
show are useful for explaining changes in corporate credit spreads. In Panel B and C, I control for the initial level of the term spread, the VIX, and the initial
level of the spread for the each asset. Corporate bond spreads in Panel A and B for investment-grade corporates, Aaa-rated corporates, and Baa-rated
corporate are the (option-adjusted) spreads on Barclays indices over duration-matched Treasuries.The ten-year swap spread is from Bloomberg. The
returns in Panel C are the excess returns on these indices over duration-matched Treasuries.For swaps, I do not have an exact measure of excess returns
over duration-matched Treasuries. Thus, I use the Shiller, Campbell, and Schoenholtz (1983) approximation. I compute this excess return using the yield on
ten-year swaps and the ten-year par Treasury yield from Gurknayak, Sack, and Wright (2007).

IG corporate spread Aaa corporate spread Baa corporate spread 10-yr swap spread

(1) (2) (3) (4) (5) (6) (7) (8)

Panel A: Contemporaneous 3-month changes in spreads over duration-matched Treasuries

ΔDUR_CNTRBMBS
t  0.43  0.16  0.14 0.04  0.72  0.35 0.13 0.17
[  3.56] [  2.21] [  1.92] [0.50] [  3.52] [  2.75] [3.40] [5.22]

Δðyð10Þ
t  yð1Þ
t 0.05 0.04 0.06  0.08
[0.52] [0.46] [0.48] [  2.98]

ΔVIX t 0.03 0.03 0.03 0.00

[2.78] [2.40] [2.70] [1.98]

MKTRF t  0.01 0.00  0.02 0.00

[  2.40] [1.14] [  3.84] [1.21]

Constant 0.01 0.03 0.00 0.00 0.01 0.04  0.01  0.01

[0.19] [0.79] [0.06] [0.17] [0.13] [0.94] [  0.97] [  1.10]

Number of observations 265 265 265 265 265 265 265 265
R2 0.06 0.51 0.01 0.32 0.10 0.50 0.09 0.22
bps effect 1 s DUR  11.2  4.0  3.5 0.9  18.8  9.0 3.3 4.3

Panel B: Forecasting future 12-month changes in spreads over duration-matched Treasuries

DUR_CNTRBMBS
t 0.85 0.63 0.31 0.30 1.18 0.92  0.17  0.16
[2.68] [3.72] [1.46] [2.81] [2.93] [4.07] [  2.00] [  2.87]

yð10Þ
t  yð1Þ
t  0.12  0.08  0.11  0.09
[  1.44] [  1.65] [  1.23] [  4.32]

VIX t 0.00 0.01 0.01 0.00

[0.15] [0.81] [0.76] [0.44]

SPREADt  0.47  0.82  0.61  0.47

[  1.89] [  4.53] [  2.88] [  4.29]

Constant  0.87 0.24  0.33 0.34  1.24 0.04 0.15 0.56

[  2.52] [0.68] [  1.70] [1.64] [  2.61] [0.11] [1.55] [4.25]

Number of observations 256 256 256 256 256 256 256 256
R2 0.07 0.35 0.02 0.44 0.08 0.37 0.06 0.40
bps effect 1-s DUR  24.7  18.4  8.9  8.6  34.5  26.8 4.9 4.7

Panel C: Forecasting 12-month excess returns over duration-matched Treasuries

DUR_CNTRBMBS
t  4.65  3.59  2.28  1.58  6.53  5.10 1.20 0.49
[  2.75] [  3.87] [  1.99] [  2.31] [  3.17] [  4.66] [2.08] [1.02]

yð10Þ
t  yð1Þ
t 0.51 0.41 0.50 0.22
[1.11] [1.74] [1.05] [1.91]
 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299 297

Table 10 (continued )

IG corporate spread Aaa corporate spread Baa corporate spread 10-yr swap spread

(1) (2) (3) (4) (5) (6) (7) (8)

VIX t 0.03 0.05  0.07  0.01

[0.25] [0.87] [  0.71] [1.01]

SPREADt 2.49 0.15 3.44 3.86

[1.74] [0.16] [3.09] [4.48]

Constant 5.21  0.57 2.38  0.11 7.37 0.29  1.10  2.43

[2.82] [  0.30] [2.41] [  0.11] [2.97] [0.13] [  2.26] [  2.52]

Number of observations 256 256 256 256 256 256 256 256
R2 0.07 0.32 0.06 0.13 0.08 0.38 0.05 0.32

Treasuries. And because swaps are used to hedge MBS, matched Treasuries, while swaps are expected to outper-
each of these predictions should be reversed for swaps. form duration-matched Treasuries.
I find evidence consistent with each of these additional Finally, Perli and Sack (2003), Wooldridge (2001), and
predictions in Table 10. In Panel A, I report regressions of Reinhart and Sack (2000) argue that many MBS investors
the form switched from hedging with Treasuries to hedging with
interest rate swaps beginning in the late 1990s. Several
ΔSPREADt ¼ a1 þb1  ΔDUR_CNTRBMBS
t þ c01 Δxt þ vt : ð29Þ
episodes, including the flight-to-quality dislocations in the
Specifically, I regress three-month changes in spreads on fall of 1998 and the jump in the convenience premium on
the contemporaneous change in MBS duration contribution Treasuries following the Treasury buyback announcement
and controls. I find that increases in MBS duration are in early 2000, convinced many MBS investors that hedging
associated with tighter spreads between corporate bonds with Treasuries exposed them to far greater basis risk than
and Treasuries and with wider swap spreads. This holds in hedging with swaps. Interestingly, the time series evi-
univariate regressions as well as in multivariate regressions dence is consistent with such a shift in hedging activity.
that control for changes in the term spread, the Chicago In untabulated results, I find that the tendency for swap
Board Options Exchange Market Volatility Index (VIX), and spreads to widen when MBS duration rises became sig-
the past returns on the stock market, which Collin- nificantly more pronounced after 1997.
Dufresne, Goldstein, and Martin (2001) show are useful In summary, the evidence is consistent with previous
for explaining changes in corporate credit spreads. Spreads work, including Cortes (2003, 2006) and Feldhutter and
are in percentage points, so the coefficient of  0.43 in Lando (2008), which argues that MBS hedging plays a
Column 1 suggests that a one-year rise in DUR_CNTRBMBS t is significant role in explaining the level of US swap spreads.
associated with a 43 bps decline in investment-grade And this spread-based evidence supports the broader
spreads. Based on the univariate estimates, a 1 standard argument of this paper: Shifts in MBS duration and the
deviation change in DUR_CNTRBMBS t is associated with a 11 associated hedging flows function as large-scale supply
bps decline in all investment-grade corporate spreads, a 4 shocks that have significant effects on bond market pricing.
bps decline in Aaa spreads, a 19 bps decline in Baa spreads,
and a 3 bps rise in ten-year swap spreads. 4. Conclusion
In Panel B, I use the level of MBS duration to forecast
spread changes over the next 12 months: Changes in the effective duration of mortgage-backed
securities function as large-scale duration supply shocks in
ΔSPREADt þ 1 ¼ a2 þ b2  DUR_CNTRBMBS
t þ c02 xt þ ut þ 1 : ð30Þ
US bond markets. As a result, bond term premia are high
I find that a high level of MBS duration predicts that when aggregate MBS duration is high, and changes in MBS
corporate spreads will widen over the next 12 months and duration capture variation in bond risk premia that is not
that swap spread spreads will tighten. Again, this holds reflected in traditional forecasting variables. The fact that
true in univariate forecasting regressions as well as in these recurring and transient supply shocks have large
multivariate regressions that control for the initial level of effects on bond prices highlights the critical role of limited
spreads and other conditioning variables. and slow-moving arbitrage capital, even in markets as
Finally, I forecast the excess returns on corporate bonds deep and liquid as the US bond market.
and swaps over duration-matched Treasuries over the The negative convexity of MBS—the fact that MBS dura-
following 12 months: tion rises when interest rates rise—generates a positive-
 MATCH feedback loop that helps explain the excess sensitivity of
rxDUR
tþ1 ¼ a3 þb3  DUR_CNTRBMBS
t þc03 xt þεt þ 1 : ð31Þ
long rates to short rates and excess volatility of long rates.
By construction, these regressions provide almost identical MBS convexity has the potential to amplify a variety of
information to those in Panel B. Specifically, a high level shocks within US bond markets. For instance, one might
of MBS duration predicts that the returns on long-term expect the effective risk tolerance of fixed income arbitra-
corporate bonds will underperform those on duration- geurs to decline following losses. If arbitrageurs are long
298 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299

duration and lose money when rates rise, negative MBS Cortes, F., 2003. Understanding and modelling swap spreads. Bank of
convexity could add another positive-feedback loop to US England Quarterly Bulletin (Winter), 2003.
Cortes, F., 2006. Understanding the term structure of swap spreads. Bank
bond markets. Specifically, a decline in arbitrageur risk of England Quarterly Bulletin (Spring), 2006.
tolerance would lead to a rise in yields, which would cause D'Amico, S., English, W., Lopez-Salido, D., Nelson, E., 2012. The Federal
the duration of MBS to extend, raising required returns and Reserve's large-scale asset purchase programs: rationale and effects.
Finance and Economica Discussion Series working paper 2012-85.
causing yields to rise further. The resulting losses would Federal Reserve Board, Washington, DC.
further lower arbitrageur risk tolerance, and so on. Duarte, J., 2008. The causal effect of mortgage refinancing on interest rate
Thus, MBS convexity could sometimes act as a destabi- volatility: empirical evidence and theoretical implications. Review of
Financial Studies 21 (3), 1689–1731.
lizing force within US bond markets, as it arguably did at Duffee, G.R., 1996. Idiosyncratic variation of Treasury bill yields. Journal of
the beginning of the three most recent Fed tightening cycles Finance 51, 527–551.
in 1994, 1999, and 2003. While it could be desirable for Duffee, G.R., 2011. Information in (and not in) the term structure. Review
of Financial Studies 24, 2895–2934.
households to have access to mortgages that embed a no-
Duffie, D., 2010. Asset price dynamics with slow-moving capital. Journal
penalty prepayment option, this could come at some cost in of Finance 65, 1238–1268.
terms of financial stability. More speculatively, my analysis Duffie, D., Strulovici, B., 2012. Capital mobility and asset pricing. Econo-
offers a potential explanation for the finding that failures of metrica 80, 2469–2509.
Fama, E.F., Bliss, R.R., 1987. The information on long-maturity forward
the expectations hypothesis are more pronounced in the US rates. American Economic Review 77 (4), 680–692.
than in most other developed nations (see, e.g., Campbell, Feldhutter, P., Lando, D., 2008. Decomposing swap spreads. Journal of
2003). Because the US is one of the few countries with long- Financial Economics 88, 375–405.
Fernald, J.K., Keane, F., Mosser, P.J., 1994. Mortgage security hedging and
term fixed rate mortgages that embed a true interest rate the yield curve. FRBNY Quarterly Review (Fall/Summer), 92–100.
call, the existence of a significant MBS convexity effect Gabaix, X., Krishnamurthy, A., Vigneron, O., 2007. Limits of arbitrage:
suggests a novel solution to this puzzle. theory and evidence from the mortgage-backed securities market.
Journal of Finance 62 (2), 557–595.
Finally, my analysis indicates that shocks to MBS duration Gagnon, J., Raskin, M., Remache, J., Sack, B., 2011. Large-scale asset
impact term premia throughout the US bond market (for purchases by the Federal Reserve: did they work?. Economic Policy
government debt, MBS, and corporate debt alike) as predicted Review (May), 41–59.
Greenwood, R., Vayanos, D., 2010. Price pressure in the government bond
by a model such as Vayanos and Vila (2009) in the portfolio-
market. American Economic Review: Papers and Proceedings 100 (2),
balance tradition. This finding is relevant for the ongoing 585–590.
debate about the effectiveness of Quantitative Easing policies, Greenwood, R., Vayanos, D., 2014. Bond supply and excess bond returns.
Review of Financial Studies 27 (3), 663–713.
the purchases of long-term bonds by the Federal Reserve and
Gürkaynak, R.S., Sack, B., Swanson, E.T., 2005. The sensitivity of long-term
other global central banks. For instance, Krishnamurthy and interest rates to economic news: evidence and implications for macro-
Vissing-Jorgensen (2013) argue that the markets for MBS, economic models. American Economic Review 95 (1), 425–436.
corporate bonds, and Treasuries are highly segmented from Gürkaynak, R.S., Sack, B., Wright, J.H., 2007. The US Treasury yield curve:
1961 to the present. Journal of Monetary Economics 54 (8),
one another, so QE policies do not work through broad 2291–2304.
channels such a shifting the market-wide price of interest Gürkaynak, R.S., Sack, B., Wright, J.H., 2010. The TIPS yield curve and
rate risk. My results do not show that QE policies have inflation compensation. American Economic Journal: Macroeco-
nomics 2 (1), 70–92.
lowered the duration risk premium in bond markets. How- Hamilton, J.D., Wu, C., 2012. The effectiveness of alternative monetary
ever, they do provide something of a proof of concept for QE policy tools in a zero lower bound environment. Journal of Money,
policies that seek to impact market-wide term premia. Credit, and Banking 44 (s1), 3–46.
Hanson, S.G., Stein, J.C., 2012. Monetary policy and long-term real rates.
Comparably sized shocks to the quantity of duration appear Finance and Economics Discussion Series working paper 2012-46.
to have had significant effects on term premia in the past. Federal Reserve Board, Washington, DC.
Hodrick, R.J., Prescott, E.C., 1997. Postwar U.S. business cycles: An
empirical investigation. Journal of Money, Credit and Banking 29 (1),
References 1–16.
Jarrow, R.A., Li, H., 2012. The impact of quantitative easing on the U.S. term
Cahill, M.E., D'Amico, S., Li, C., Sears, J.S., 2013. Duration risk versus local structure of interest rates. Research paper 2-2012. Cornell University
supply channel in Treasury yields: evidence from the Federal Samuel Curtis Johnson Graduate School of Management, Ithaca, NY.
Reserve's asset purchase announcements. Finance and Economics Joslin, S., Priebsch, M., Singleton, K.J., 2013. Risk premiums in dynamic
Discussion Series working paper 2013-35. Federal Reserve Board, term structure models with unspanned macro risks. Journal of
Washington, DC. Finance 69 (3), 1197–1233.
Campbell, J., Shiller, R., 1991. Yield spreads and interest rate movements: Kambhu, J., Mosser, P.C., 2001. The effect of interest rate options hedging
a bird's eye view. Review of Economic Studies 58, 495–514. on term-structure dynamics. Federal Reserve Bank of New York
Campbell, J.Y., 1999. Asset prices, consumption, and the business cycle. In: Quarterly Review (December), 51–70.
Taylor, J.B., Woodford, M. (Eds.), Handbook of Macroeconomics, vol. 1. Krishnamurthy, A., Vissing-Jorgensen, A., 2011. The effects of quantitative
, North Holland, Amsterdam, pp. 1231–1303. easing on interest rates: channels and implications for policy. Brook-
Campbell, J.Y., 2003. Consumption-based asset pricing. In: Constanti- ings Papers on Economic Activity (Fall), 215–265.
nides, G., Harris, M., Stulz, R. (Eds.), Handbook of the Economics of Krishnamurthy, A., Vissing-Jorgensen, A., 2012. The aggregate demand for
Finance, vol. IB. , North Holland, Amsterdam, pp. 803–887. Treasury debt. Journal of Political Economy 120 (2), 233–267.
Cieslak, A., Povala, P., 2013. Expected returns in Treasury bonds. Unpub- Krishnamurthy, A., Vissing-Jorgensen, A., 2013. The in and outs of LSAPs.
lished working paper, Northwestern University working paper. Evan- In: Economic Policy Symposium, Jackson Hole, WY. Federal Reserve
ston and Chicago, IL. Bank, Kansas City, MO.
Cochrane, J.H., Piazzesi, M., 2005. Bond risk premia. American Economic Li, C., Wei, M., 2013. Term structure modeling with supply factors and the
Review 95 (1), 138–160. Federal Reserve's large-scale asset purchase programs. International
Cochrane, J.H., Piazzesi, M., 2002. The Fed and interest rates: a high- Journal of Central Banking 9 (1), 3–39.
frequency identification. American Economic Review: Papers and Longstaff, F.A., 2004. The flight-to-liquidity premium in U.S. Treasury
Proceedings 92 (2), 90–95. bond prices. The Journal of Business 77 (3), 511–526.
Collin-Dufresne, P., Goldstein, R.S., Martin, J.S., 2001. The determinants of Lou, D., Yan, H., Zhang, J., 2013. Anticipated and repeated shocks in liquid
credit spread changes. Journal of Finance 56, 2177–2207. markets. Review of Financial Studies 26 (8), 1891–1912.
 S.G. Hanson / Journal of Financial Economics 113 (2014) 270–299 299

Malkhozov, A., Mueller, P., Vedolin, A., Venter, G., 2013. Mortgage risk and Shiller, R.J., 1979. The volatility of long-term interest rates and expecta-
the yield curve. Unpublished working paper. London School of tions models of the term structure. Journal of Political Economy 87,
Economics, London, UK. 1190–1219.
Modigliani, F., Sutch, R., 1966. Debt management and the term structure Shiller, R.J., Campbell, J.Y., Schoenholtz, K.L., 1983. Forward rates and
of interest rates: an empirical analysis of recent experience. Journal of future policy: interpreting the term structure of interest rates.
Political Economy 75, 569–589. Brookings Papers on Economic Activity 1, 173–217.
Newey, W.K., West, K.D., 1987. A simple, positive semi-definite, hetero- Tobin, J., 1958. Liquidity preference as behavior towards risk. Review of
skedasticity and autocorrelation consistent covariance matrix. Econ- Economic Studies 25, 124–131.
ometrica 55, 703–708. Vayanos, D., Vila, J.-L., 2009. A preferred-habitat model of the term
Perli, R., Sack, B., 2003. Does mortgage hedging amplify movements in structure of interest rates. Unpublished working paper. London
long-term interest rates? Journal of Fixed Income 13 (3), 7–17. School of Economics, London, UK.
Reinhart, V., Sack, B., 2000. The economic consequences of disappearing Wooldridge, P., 2001. The emergence of new benchmark yield curves.
government debt. Brookings Papers on Economic Activity 31 (2), Fourth Quarterly Review: International Banking and Financial Market
163–220. Developments, 48–57.