9/17/22, 1:27 PM Robert C. Merton - Continuous Finance.

html

Robert C. Merton - Continuous Finance

Intro

These are some very high level (and nontechical) notes on the book, which is basically a collection
of the main papers of the author.

Notes

Take bKp as the measure of risk for some portfolio p relative to the efficient portfolio K : the measure
​

is ordering-consistent regardless of the reference K portfolio (more or less, this should mean that
the portfolio space of all p's wrt a given K is homeomorphic to the same p's with some other Q as
reference).

The security ordering wrt this measure is total. The efficient portfolio to use as the reference K
may be found with Rothschild-Stiglitz (or Markowitz etc).

There exist only three statistically indepdendent portfolio factors (the rest is data mining).

In stochastic calculus, all moments above 2 are asymptotically unrelevant as we zoom in, as in in
the infinitesimal. Intuitively, this is why most models just care about mean and variance.

"Well-behaved" returns, example: arithmetic Brownian motion (ABM): in the very short interval,
stock returns are dominated by the unknown/stochastic component. Empirically, a stock with an
annual return of 15% and a volatility of 15% as well, will show a return of like 0.05% already, which
will become even smaller at higher resolutions, of course.

Looking at a couple of formulas decomposing the ABM, to see why is it that mean and variance are
dominating on the stocastic turn, just observe that α, σ and the Brownian score are all O(1)
(constants) and h → ∞ for ∞ → 0.
​

Derivatives on returns described by an ABM as above have a closed form dynamic too: from
Xt − Xt−1 to F (Xt ) − F (Xt−1 ) the form is the same actually, except for an o(h) (little oh) which is
​ ​ ​ ​

stocastic on the interval. But such little oh makes a negligible contribution to the moments of dFt , ​

then, the contemporaneous, instantaenous changes of X and F are correlated regardless of F

But how does F behave over a non necessarily small (ie noninfinitesimal) interval? So more like
real life? This is basically taking the difference rather than the differential. The cumulative error of
the discrete approximation goes to zero certainly. The famous asymptotic is therefore neglegible in
effect as the resolutions increase. This means that stuff is well behaved.

Following the introduction of the Ito integral we can say that σ 2 (x, t) and α(x, t) are sufficient to
study the difference in security levels (prices), not returns between any two dates.
file:///home/ricky/Documents/Robert C. Merton - Continuous Finance.html 1/6
9/17/22, 1:27 PM Robert C. Merton - Continuous Finance.html

Ito's lemma is the stocastic counterpart of the fundamental theorem of calculus.

Security price dynamics with no rare events (jumps) may always be described in Ito's terms.

The fact that the behavior is normal in the instant follows from the IID hypothesis, and the
distribution of the differences themselves does not have to be normal: this is the central limit
theorem.

No rare events means: no jumps / regime changes. Jumps or not, still just the first two moments
matter. There is an Ito for Poisson-governed jump processes though!

Merton states that calibration in continuous time is empirically simple.

If prices are exactly lognorm distributed for all securities of a universe, then there exists a pair of
"perfect" mutual funds which can be proven to be better, in some linear combination, than any
other linear combination on the universe (two-fund theorems).

Prices in reality are not really lognormally distributed but rather fat-tailed. Wlog one of these assets
in the pair above is riskfree, the other a generic risky one.

Optimal consumption and portfolio rules (ie strategy) are possible in closed form if the preference
of the individual may be modeled in the HARA class (preference = risk aversion).

That is still true when adding a constant income, riskless stream for the investor, like a wage or
salary.

Such theory can be generalized for assets following not only lognorm (GBM, geometric brownian
motion returns/increments) returns but poissonian too (jumpy).

Some of the results above are shown to hold for other more advanced return models, such as
nonstationary brownian ones.

Closed-form portfolio strategies (what to hold at each time t) are variable (and built in a dynamic
programming-like fashion).

All investors in optimal portfolios converge to the same model parameters, elastically, for their
strategy: and same weights means perfect correlation (at least in the case of state-
indepdendent/totally arbitrary preferences).

With advanced preferences, a two-asset theorem is not enough, it must generalize to an N-assets
one.

Perpetual warrants may be proven to be priced exactly as the underlying stock, which may sound
paradoxical. But their yields, if any, must be higher than that of the stock.

For American puts, it is not true that their price is fully defined starting from the call. A put is
technically more difficult to price and less used than a call.

file:///home/ricky/Documents/Robert C. Merton - Continuous Finance.html 2/6

The value of an American option is always a non-decreasing function of its expiration date, unlike a
European one.

Black-Scholes is strictly speaking for Europeans only, but Americans have valuations that are often
comparable.

B-S: it is not true that if the market prices differently then an arbitrage is guaranteed.

B-S is a theorem, don't forget that! See pp 220-226 for a clarification on the assumptions and some
limiting arguments. Taking a step forward, he also relaxes some assumptions.

B-S may be used to "price" a firm based on its assets and claims.

Natively, B-S does not work with jumpy price processes.

Following a "long B-S hedge" and a "short" one are two options strategies.

The theory explains to us why option writers make money in relatively calm (no jump) periods (long
hedge).

If the underlying process is "random enough", one cannot be systematically long on or short, of
course.

By definition of a jump process, the variance of the steady part of the process price, which is the
geometric Brownian, is irrelevant.

In the jumpy case, the riskless portfolio cannot be built intuitively because you cannot hedge the
jump enough.

A proper jump is uncorrelated risk, non-systemtic (exogenous black swan). A closed form of B-S in
this case cannot be obtained, because one misses the distribution for Y , the jump value - but
everything is relatively amenable numerically.

Eventually, the associated formula looks like the real one, yet with a jump-inflated riskfree rate.

An option on a jumpy stock is more valuable than one that's not, all things being equal: intuitively,
the jump can scramble things up (or down) right before the expiration, regardless of moneyness
but especially for OOMs.

Basically, deep ITM and OOM (and shorter term) options tend to be priced above of their "B-S
price".

An alternative to B-S: binomial model (Cox-Rulenstein). It does not require continuous trading, it is
easier to define. In practice it is limited by the binomial ("one price or the other"/"up or down")
assumption.
The higher the resolution, the higher the "pathiness" (convergence in limit), of course:
in t → 0 seconds we may assume the price will be +1 tick or −1 tick / unit e.g. cent.

file:///home/ricky/Documents/Robert C. Merton - Continuous Finance.html 3/6

The binomial converges in limit to B-S, but for a proper choice of the binomial process, and
specifically the Ito solution of B-S itself.

Long "default free" bonds as a way to hedge against interest rates changes.

Term structure (yield curve) is seen as a consequence of the capital term structure results.

B-S can be used to price (corp) bonds.

σ 2 (std) is not sufficient to compare with each other debt from two different securities.

We cannot say that a higher rate means higher std, for bonds.

General use for B-S: pricing "contingent claims" (contingent claims analysis, CCA), e.g. call
options, multi-listed stocks etc, or corporate liabilities as a function of the stock.

CCA analysis can even be used to price MBSs and corporate employee compensation plans (e.g.
stock executive options). And, drive project and investment decisions at, say, power plants: like
when to start a project, or if at all. Or, in a single, long project (things like building an airplan), CCA
can be used to price intermediate goals.
CCA is a master tool for project evaluation.

CCA makes quantitative assets possible, whereas before and more traditionally, stuff is evaluated
quickly or ad-hoc (classic accountancy).

All things being equal (eg transaction costs, units etc) the number of shares held short to hedge a
long position are fewer than those held long to hedge a short. And, the minimum price at which a
perfect trader would sell a call exceeds the maximum at which he would buy the call.

American puts are an example of derivatives with a path-dependent payoff: they can be exercises
at any time before expiry so the price of the underlying at this or that point in time before the expiry
does matter. Europeans are not p-d, and this is the argument for them to be easier to price.

The more complex and structured the derivative, the more expensive and the more risk to
decimate returns. Think of very complex contract where you may get kicked out (ie worthless
expiration) at multiple conditions along the path.

No arbitrage safeguards postulate that the total price of the components may not exceed the price
of the product. Think carefully when buying complex products and evaluate how a partial
combination of multiple one-legged (simple) products compares to that, both in time (expiry ie
bermudiness) and space (path conditions).

Creating a derivative with an arbitrary payoff profile is computationally no more difficult than
creating a call option (eqq 14.9-10). The "assembly line" is CCA theory. And the risk exposure of
the investment bank is to be reconduced to a simple portfolio containing the underlying and a
riskless security traded on a secondary market (possibly a liquid one).

Those complex derivatives don't need to be traded on a market for the bank to be able to quantify
its risk. This is why, as long as that secondary and possibly liquid market access of above is
file:///home/ricky/Documents/Robert C. Merton - Continuous Finance.html 4/6
9/17/22, 1:27 PM Robert C. Merton - Continuous Finance.html

granted, the bank may go into quite complicated things. The "magic" of CCA is that this can be
applied to price virtually (virtually as in, not everything, like goodwill) any asset of the listed
company.

A hedging system for the bank relies on the net exposures if they sell both bearish and bullish stuff
(eg puts and calls). They can save on hedging costs by properly offsetting the way complex
products react to the underlying, so basically how convex or concave they are with respect to it.
The gamma and delta greeks measure this. And of course in an attempt to minimize costs, perfect
hedging is replaced by covariance (eg diversification) considerations: perfect hedging would kill the
profit for the bank (but when correlations melt and everything goes donw, it is the bank that gets
killed).

The most risk- and regulatory-friendly option is a put on the whole portfolio (protective put portfolio
insurance). This is a big difference between the majority of the investors and a handful of them: the
latter group pays for insurance like we do for things like a car or personal liability in the real world,
the former does not, judging protection as too expensive. And a modern investment bank is an
investment provider for other investors, not really an active investor itself, especially after the end
of the prop trading era with 2008.

For a number of reasons, the bank can better self-insure and create the portfolio in-house.

The put above needs to be synthetic (in the original case of nontrivial products).

Albeit criticised, CAPM seems to be able to explain a good deal of variability in assets.

Portfolios designed to have zero covariance with the market have returns well above the riskfree
rate (which is actually the minimum we expect, due to the inherently higher costs - think about
classic managed futures-based trend following and its rebalancing). We call the best of this class
the "beta stocks".

CAPM, empirically, has a better performance than its consumption-based version, CCAPM.

A consumption-based investment protects from both inflation and changes in the standard of living.
"Wages-linked" bonds are also fine. The adjustment with consumption means: the higher the cost
of living, the higher the pension.
This kind of theoretical product is known as SELFies bonds.
Collecting contributions is the challenge though: how do you measure and monitor somebody's
consumption accurately enough, anyway? That's all that jazz about privacy etc.

As we have seen above, B-S is the unlocker to price corporate liabilities accordingly: it relies on
non-historical and observable features, which is also the reason behind its popularity.

Puts and debts: if an entity takes the role of guarantor for a debt somebody has, they are in a
position equivalent to that of a put holder on their own assets. Before the end of the loan, in case
the guarantee defaults on its debt, they can sell the put to cover the grant. By pricing that "debt
put", the price for the premium the guarantee should pay may be found.

file:///home/ricky/Documents/Robert C. Merton - Continuous Finance.html 5/6

In the construction above, if the collateral put up by the guarantee for the grantor grows in
variance, or the loan duration rises, the protection costs should go with it. To keep premia stable
and sustainable, low risk collateral is usually put up (for listed swap-based retail products this is
basically mandated by law).

Endowment diversification paradox: if an endowment is really dependent on an asset class (eg MIT
<==> venture capital), that should be reflected in the allocations themselves, perhaps even
expressing this with short positions. Example: hedge-specific endowment allocations invested in oil
if the university has to produce energy for some facilities.

All its endowment theory is saying is revolving around a very basic principle: because of the nature
and operations of a university, endowments should also be used for hedging purposes. On the one
hand, their mandate is long-term and generational value creation in the present and future interest
of the university. On the other hand, because it is associated to a university seen as a modern
enterprised, hedges may not be ignored. In this sense, the endowment management may take
roles looking like those of a classic CFO in a production company.

Implicitly, the judgement on American endowments is not positive, due to the lack of such hedging
awareness and traditional market-heavy beta - something which is a characteristic trait of
American investors anyway, especially in the past ten-fifteen years thanks to an exceptional S&P
500 performance.

file:///home/ricky/Documents/Robert C. Merton - Continuous Finance.html 6/6