Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
1 / 30
Suppose the real-world probability measure P quanties the markets belief about the frequencies of future states. The risk-neutral probability measure Q describes futures prices of Arrow Debreu securities. In 2011, Stephen A. Ross began circulating a working paper called The Recovery Theorem. The Ross Recovery Theorem gives sucient conditions under which P can be obtained from Q.
2 / 30
Theorem 1 in Ross (2011) states that: if markets are complete, and if the utility function of the representative investor is state independent and intertemporally additively separable and: if the state variable is a time homogeneous Markov process X with a nite discrete state space, then: one can recover the real-world transition probability matrix of X from the assumed known matrix of Arrow Debreu state prices.
3 / 30
Jiming Yu and I were intrigued by Rosss conclusion, but bothered by the restrictions on preferences that Ross made to generate it. In particular, it seems impossible to test whether a representative investors utility function is additively separable or not. Fortunately, there is a concept in the nancial literature which can be used to derive Rosss conclusion without restricting preferences. We exploit this concept to replace Rosss restrictions on preferences with restrictions on beliefs, which we believe to be more testable.
4 / 30
In 1990, John Long introduced the notion of a numeraire portfolio. A numeraire is any self-nancing portfolio whose price is alway positive. Long showed that if any set of assets is arbitrage-free, then there always exists a numeraire portfolio comprised of just these assets. The dening property of this numeraire portfolio is the following surprising result: If the value of the numeraire portfolio is used to deate each assets dollar price, then each deated price evolves as a martingale under the real-world probability measure.
5 / 30
Sit ,
Let M be the positive P martingale used to create Q from P: EP Sit MT SiT |Ft = , Mt S0T S0t t [0, T ], i = 0, 1, . . . , n.
1 Fact: If M creates Q from P, then M creates P from Q. To conserve 1 probability, M must be a positive Q local martingale. Let L denote the product of the money market account S0 and this reciprocal:
Lt
S0t , Mt
for t [0, T ].
L is clearly a positive stochastic process. Since L is just the product of the 1 MMA and the Q local martingale M , L grows in Q expectation at the risk-free rate. As result, L is the value of some self-nancing portfolio. Multiplying both sides of the top equation by Mt and substituting in L implies: Sit SiT EP |Ft = , t [0, T ], i = 0, 1, . . . , n. LT Lt In words, L is the value of the numeraire portfolio that Long introduced.
Carr/Yu (NYU Courant) Risk, Return, and Ross Recovery September 13, 2012 8 / 30
To gain intuition, suppose each spot price Si is positive and has continuous sample paths over time (i.e. no jumps). The value L of the numeraire portfolio is always positive and now it would be continuous as well. Let Rit implies:
Sit Lt
be the ratio at time t of each spot price Sit to Lt . Its formula o dRit dSit dLt dSit dLt = + Rit Sit Lt Sit Lt dLt Lt
2
The last two terms are of order dt and arise due to the unbounded variation of the sample paths of L.
9 / 30
dRit dSit dLt dSit dLt = + Rit Sit Lt Sit Lt Let rt be the spot interest rate at time t. If: EP then since L is a portfolio: EP dLt = rt + Lt dLt Lt
2
and so taking expected value in the top equation implies: EP dRit = rt rt = 0. Rit
Hence, the ratio Rit Sit is a P local martingale. By restricting asset Lt volatilities the stronger condition of martingality can be achieved.
Carr/Yu (NYU Courant) Risk, Return, and Ross Recovery September 13, 2012 10 / 30
In a continuous world, the quadratic variations and covariations of martingale components under P are the same as under Q . In a continuous setting, the bounded variation components ( = expected returns) under P arise by adding the short rate r to the covariations dSitit dLtt . S L So for continuous semi-martingales, the dynamics of each spot price under P would be determined.
11 / 30
One could not imagine a simpler relation between risk premium and risk. Suppose we let denote the instantaneous lognormal volatility of the numeraire portfolio, i.e. 2 dLt = 2 dt Lt Then the market price of Brownian risk is t . Our goal is now to determine this volatility, which represents both risk and reward.
12 / 30
Using Longs numeraire portfolio, we replace Rosss restrictions on the form of preferences with our restrictions on the form of beliefs. More precisely, we suppose that the prices of some given set of assets are all driven by a univariate time-homogenous bounded diusion process, X . Letting L denote the value of the numeraire portfolio for these assets, we furthermore assume that L is also driven by X and t and that (X , L) is a bivariate time homogenous diusion. We show that these assumptions determine the real world dynamics of all assets in the given set.
13 / 30
Our Assumptions
We assume no arbitrage for some nite set of assets which includes a money market account (MMA). As a result, there exists a risk-neutral measure Q under which prices deated by the MMA evolve as martingales. Under Q, the driver X is a time homogeneous bounded diusion: dXt = b(Xt )dt + a(Xt )dWt , t [0, T ].
where X0 ( , u) and where W is standard Brownian motion under Q. dLt = r (Xt )dt + (Xt )dWt , Lt
We also assume that under Q, the value L of the numeraire portfolio solves: t [0, T ].
We know the functions b(x), a(x), and r (x) but not (x). How to nd it?
14 / 30
where L(x, t) is a positive function of x R and time t [0, T ]. Applying Its formula, the volatility of L is: o (x) 1 L(x, t)a(x) = a(x) ln L(x, t). L(x, t) x x
ln L(x, t) =
Exponentiating implies that the value of the numeraire portfolio separates multiplicatively into a positive function () of the driver X and a positive function p() of time t: L(x, t) = (x)p(t), where: (x) = e
Carr/Yu (NYU Courant)
x (y ) dy a(y )
Separation of Variables
The numeraire portfolio value function L(x, t) must solve a PDE to be self-nancing: a2 (x) 2 L(x, t) + L(x, t) + b(x) L(x, t) = r (x)L(x, t). t 2 x 2 x On the other hand, the last slide shows that this value separates as: L(x, t) = (x)p(t). Using Bernoullis classical separation of variables argument, we know that: p(t) = p(0)e t , and that: a2 (x) (x) + b(x) (x) r (x)(x) = (x), 2 x [ , u].
16 / 30
Here (x) and are unknown. This can be regarded as a regular Sturm Liouville problem. From Sturm Liouville theory, we know that there exists an eigenvalue 0 , smaller than all of the other eigenvalues, and an associated positive eigenfunction, (x) which is unique up to positive scaling. All of the eigenfunctions associated to the other eigenvalues switch signs at least once. One can numerically solve for both the smallest eigenvalue 0 and its associated positive eigenfunction, (x). The positive eigenfunction (x) is unique up to positive scaling.
17 / 30
As a result, the volatility of the numeraire portfolio is uniquely determined as: (x) = a(x) ln (x), x [ , u]. x Mission accomplished! Lets see what the market believes.
18 / 30
where B is a standard Brownian motion under the real world probability measure P. In equilibrium, the risk premium of the numeraire portfolio is simply 2 (x). Since we have determined (x) on the last slide, the risk premium of the numeraire portfolio has also been uniquely determined. The market price of Brownian risk is simply (Xt ). The function (x) is now known but what about the dynamics of X ?
19 / 30
From Girsanovs theorem, the dynamics of the driver X under the real world probability measure P are: dXt = [b(Xt ) + (Xt )a(Xt )]dt + a(Xt )dBt , t 0,
where recall B is a standard Brownian motion under the real world probability measure P. Hence, we know the real world dynamics of the driver X . We still have to determine the real world transition density of the driver X .
20 / 30
dP dQ
is:
r (Xt )dt
T 0
r (Xt )dt
r (Xt )dt
dQ denote the assumed known Arrow Debreu state (XT ) 0 T e dA. (X0 )
dP =
(XT ) 0 T e e (X0 )
T 0
r (Xt )dt
dQ =
As we know the function (y ) , the positive function e T , and the Arrow (x) Debreu state pricing density dA, we know dP, the real-world transition PDF of X .
21 / 30
where for x ( , u), t [0, T ], Si (x, t) solves the following linear PDE: a2 (x) 2 Si (x, t) + S (x, t) + b(x) Si (x, t) = r (x)Si (x, t), 2 i t 2 x x subject to appropriate boundary and terminal conditions. If Sit > 0, then the SDE at the top can be expressed as: dSit = r (Xt )+(Xt ) ln Si (Xt , t)a2 (Xt )]dt+ ln Si (Xt , t)a(Xt )dBt , t 0. Sit x x In equilibrium, the instantaneous risk premium is just d ln Si , ln L t , i.e. the increment of the quadratic covariation of returns on Si with returns on L.
Carr/Yu (NYU Courant) Risk, Return, and Ross Recovery September 13, 2012 22 / 30
Our results thus far apply only to diusions on bounded domains. Hence, our results thus far cant be used to determine whether one can uniquely determine P in models like Black Scholes (1973) and Cox Ingersoll Ross (1985) where the diusing state variable X lives on an unbounded domain such as (0, ). We dont yet know the general theory here, but we do know two interesting examples of it.
23 / 30
Suppose only one stock option trades and from its observed market price, we learn the numerical value of . All of our previous assumptions are in place except now we have allowed the diusing state variable X to live on the unbounded domain (0, ).
24 / 30
In the BS model with zero rates, a2 (x) = 2 x 2 , b(x) = r (x) = 0, = 0, and u = so we want a positive function (x) and a scalar solving the Euler ODE: 2 x 2 (x) = (x), x (0, ). 2 In the class of twice dierentiable functions, there are are an uncountably innite number of eigenpairs (, ) with positive. This result implies Ross cant recover here because there are too many candidates for the value of the numeraire portfolio. However, all of the positive functions (x) are not square integrable. If we insist on this condition as well, then there are no candidates for the value of the numeraire portfolio. Ross cant recover again, but for a dierent reason.
Carr/Yu (NYU Courant) Risk, Return, and Ross Recovery September 13, 2012 25 / 30
x ( , u).
Here, a2 (x) = 1, b(x) = 0, r (x) = x2 , = , and u = so we want a positive function (x) and a scalar solving the linear ODE: 1 x2 (x) (x) = (x), 2 2
Carr/Yu (NYU Courant) Risk, Return, and Ross Recovery
x (, ).
September 13, 2012 26 / 30
This is a famous problem in quantum mechanics called quantum harmonic oscillator. Suppose we restrict the function space to be square integrable functions on the whole real line. Then the spectrum is discrete with lowest eigenvalue 0 = 1 . The associated eigenfunction (A.K.A. ground state) is the Gaussian 2 function 2 (x) = e x /2 , x (, ). The eigenfunction is clearly positive and moreover it is unique up to positive scaling, since the other eigenfunctions have the form: n (x) = hn (x)e x
2
/2
, n = 1, 2 . . . ,
x (, ),
where hn (x) is the n-th Hermite polynomial dened by the Rodrigues formula: 2 2 n hn (x) = e x /2 Dx e x /2 , n = 1, 2, . . . .
Carr/Yu (NYU Courant) Risk, Return, and Ross Recovery September 13, 2012 27 / 30
/2 t/2
x (, ), t 0.
Under P, Girsanovs theorem implies that the state variable X becomes the following Ornstein Uhlenbeck process: dXt = Xt dt + dBt , where B is a standard Brownian motion. Since we still have rt = r (Xt ) = 2t , the short rate is still a CIR process, but with larger mean reversion under P: drt = 1 2rt 2 dt + 2rt dBt , t 0.
X2
t 0,
In this example, Ross recovery fully succeeded and moreover we used his model!
Carr/Yu (NYU Courant) Risk, Return, and Ross Recovery September 13, 2012 28 / 30
Summary
We highlighted Rosss Theorem 1 and propose an alternative preference-free way to derive the same nancial conclusion. Our approach is based on imposing time homogeneity on the real world dynamics of the numeraire portfolio when it is driven by a bounded time homogeneous diusion. We showed how separation of variables allows us to separate beliefs from preferences.
29 / 30
Extensions
Extend this work to two or more driving state variables. Explore sucient conditions under which Ross recovery is possible on unbounded domains. Explore the extent to which Rosss conclusions survive when the driving process X is generalized into a semi-martingale. Explore what restrictions on P can be obtained when markets are incomplete. Implement and test.
30 / 30