ROBUST OPTION REPLICATION FOR A BLACK-SCHOLES MODEL EXTENDED WITH NONDETERMINISTIC TRENDS

Statistical analysis on various stocks reveals long range dependence behavior of the stock prices that is not consistent with the classical Black and Scholes model. This memory or nondeterministic trend behavior is often seen as a reflection of market sentiments and causes that the historical volatility estimator becomes unreliable in practice. We propose an extension of the Black and Scholes model by adding a term to the original Wiener term involving a smoother process which accounts for these effects. The problem of arbitrage will be discussed. Using a generalized stochastic integration theory [8], we show that it is possible to construct a self financing replicating portfolio for a European option without any further knowledge of the extension and that, as a consequence, the classical concept of volatility needs to be re-interpreted.


Introduction
The Black and Scholes model for the price S of a stock, given by the stochastic differential equation dS #Stdt + (rStdWt, (1) where # is the return rate, cr the volatility and W a Wiener process, is widely accepted as a tool for the valuation of contingent claims (options) on the underlying stock [3]. The presence of long range dependence in the structure of data of stock prices from financial markets suggests, however, that the Black and Scholes model is not entirely realistic [6] and has led to proposals some years ago that the Wiener process in (1) should be replaced by a fractional Brownian motion [4,8]. Fractional 1 Brownian motion (fBm) B h with Hurst index < h < 1, introduced by Mandelbrot and Van Ness [9] to model long range dependence, is a zero mean Gaussian process with covariance function Fh(t,s): (t 2h + s 2h-It--s Ih). (2) The fBm version of the Black and Scholes model is then dS #Stdt + rBStdBht. ( However, as we will see later on, fractional Brownian motion is not a semi-martingale and there is no equivalent martingale measure, so by general results this implies almost that there must be arbitrage. In fact, Rogers [10] has shown that the model (3) admits arbitrage opportunities by constructing an arbitrage explicitly using the specific nature of the fBm. Unfortunately the arbitrage strategy in [10] is quite technical and not easy to carry out in practice. In the present paper we will argue along other lines that the model (3) cannot stand and that both a Wiener process and an additional process Z to model long range dependence behavior are required for an appropriate stock price model. In particular, we propose the model dS #Stdt -t-rStdW -t-StdZt, (4) where W is a standard Wiener process and Z is a continuous process of square variation zero, which are not necessarily independent. For technical reasons2, however, we assume that Z can be split up by two continuous, square zero variation processes Z (1) and Z (2) as Z Z (1) --Z (2) such that Z (1) is adapted to W and Z (2) is independent of W. As such Z is a smoother process than the Wiener process, but the distribution of Z is considered to be completely unknown. From the additional assumptions it follows that W is a martingale with respect to the filtration generated by {W,Z(2)} and that Z is adapted to this filtration.
In Section 3 we discuss the question of arbitrage, but we will set aside this problem for a moment and consider as an example for Z the process Z-Z (2) [8,10]. From this it follows that B h is not a semi-martingale and the use of fractional Brownian motion B h or a more general process Z with zero quadratic variation in a stochastic differential equation requires a different concept of stochastic integral since stochastic calculus based on semi-martingale integrators is not applicable. In this respect, we could use non-probabilistic pathwise integration methods of Fhllmer [5] and Bick, Willinger [2]. Also S.J. Lin [8] defined a stochastic integral with respect to a continuous process Z with zero quadratic variation for integrands of the form (Yt, zt), where : 2__, is a smooth enough function, Y is an arbitrary continuous semi-martingale on a filtered probability space and Z is adapted to this filtration. We will reformulate Lin's definition in Section 2 and give some extensions of his ideas on stochastic differential equations, including an It-like formula for solutions of these equations. In Section 4 we will construct a self financing replicating portfolio for a European option claim and will show that the initial value of this portfolio can be valuated in a way similar to the Black and Scholes theory and depends only on the coefficient r of the Wiener process in (4) and, surprisingly, not on the specific nature of the process Z.
However, in Section 5 we will see that the volatility r can now no longer be regarded as the deviation of the stock return log St + 1 but rather that r 2 is merely the rate of S the second variation of the process logs t.

Stochastic Integrals for Integrands with Zero Quadratic Variation
We start by recapitulating Lin's definition of a stochastic integral with respect to a continuous process with zero quadratic variation, such as a fractional Brownian motion. After adding some measurability requirements and changing the notation in [8] slightly, we have Definition 1-Let Y be a continuous semi-martingale on a filtered probability space and let Z be a continuous process with zero quadratic variation adapted to this filtration. Given :2_, if there exists a (I)G C2(2---) such that Oz(y,z)- Lin showed that the stochastic integral defined in this way can be regarded as a limit in probability of Riemann sums,  [8], we will see that under stable conditions the stochastic differential equation (SDE) dX a(Xt)dt + b(Xt)dW + c(Xt)dZ with X 0 x0, (5) where W is a Wiener process and Z is a square zero variation process which satisfies the same conditions as in (4), has a solution X of the form X (Ut, Zt) for a certain semi-martingale U which is to be determined and a function such that the composition function c o is like the function in Definition 1.  Xt (Ut, Zt) g(Ut + Zt + Po) (6) where g(Po) Xo" Proof: The conditions are such that the differentials in (5) are properly defined, so all we have to do is to replace (I) by in Definition 1 and everything works out straightforwardly. A is a process of finite variation (even differentiable) and adapted to W. It is clear that this situation can be considered as a special case of the model (4), where Z is proportional to A. At this point it is an interesting question whether the arbitrage for the model (4), for instance, in the case r-0 and Z-rB Bh, independent of W.
However, in the next section we will show that it is always possible to replicate, or hedge, a European option by a self financing portfolio without having any further knowledge of the process Z! We consider this as an important fact for the following reasons.
(i) If Z is such that the model (4) is mathematically arbitrage free, then the value of this portfolio at any time point before maturity is equal to the value of the European option at that time point in the usual "no arbitrage theory." (ii) There is lot of practical evidence that markets are not always in equilibrium and allow for arbitrage opportunities for a very short time due to the fact that these opportunities cannot be seen immediately. See [11]. In this situation the stock price model (4) may not be arbitrage free in the strict mathematical sense, but still will be in practice because market participants need time to discover an arbitrage opportunity due to the unknown distribution of Z, at least at the beginning. Once dealers get hold of the distribution of Z and an arbitrage strategy is seen, they will try to carry it out, but, then this will influence the stock price evolution in such a way that the possibility of arbitrage disappears again. In the model (4) this change will be reflected by a change in the distribution of Z after the discovery of the arbitrage. As time goes on there may arise a new arbitrage opportunity which will, however, disappear again after its discovery, and so on.

JOHN G.M. SCHOENMAKERS and PETER E. KLOEDEN
In this more general situation which allows for "bubble arbitrage opportunities" we will see that the European option can still be replicated almost surely by the same self-financing strategy which thus can be regarded as a robust strategy with respect to unknown smoother perturbations of the standard Black and Scholes model.

Replicating A European Option
Here we will show how the pay-off of a European contingent claim (option) can be replicated by a self-financing portfolio when a stock price follows an SDE (4) and where the process Z may reflect long range dependence. It is somewhat surprising that we do not need to know anything more about the specific nature of Z. (i) If r > O, then there exists a function Ct(. and a self-financing portfolio with value C g(t, cot) at a time instant t < T prior to T and terminal value Ct(T, ST)--g(ST) at the maturity time T such that the function C is completely determined by the risk free interest rate r, the volatility coefficient r of the Wiener process and the maturity time T. In particular, r'a'T(t, oet) is given by the standard Black and Scholes formula C a (9) where the process t,s is the solution of the SDE dS r rSrdr + rSrdWr, S s.
(10) Remarks: 1) It is not true in general here that there exists an equivalent measure [P* such that the process e-rtS is a martingale with respect to P* and as in the standard theory for option pricing. This is due to the fact that it is not possible to change a process which is not a semi-martingale into a semi-martingale by an equivalent measure transformation.
2) If the process Z has finite variation and if there exists an equivalent measure P* such that the P* distribution of Wt+#-rt is equal to the distribution of the Wiener process W, then it is easy to see that e-rto is a martingale with respect to [P* and that the present result also follows from the standard theory for option pricing. From Girsanov's theorem it can be seen that, if Z is a to W adapted process with almost sure continuous differentiable sample paths, such a * exists. Moreover, the semi-martingale with long range depen- The first term here is just the infinitesimal return of the stock, while the second term is the infinitesimal return of the risk-free investment. From these considerations, we see that the portfolio V is self-financing and replicates the pay-off value of the contingent claim with probability 1.

Conclusions, A Different Interpretation of Volatility
It is remarkable that the price of any contingent claim depends only on r, the coefficient of the Wiener term in the stock price model (4), and not on the specific nature of the process Z. Consequently, a Wiener component in the model (4) (4) it follows that the squared implied volatility of a stock, which must St+l be in accordance with (11), is substantially different from the variation of log st which in turn can be estimated from a sample of the stationary, in general dependent, S n + 1 2,.. of identical distributed Gaussian random variables, sequence log s n-0, 1, n where the time points t n are supposed to be equally spaced. Indeed, this discrepancy is observed from actual data of stock prices, see for example [7]. Our generalized Black-Scholes model (4) provides an explanation, at least partially. In order to detect abnormalities in the stock market we need to compare the squared implied volatility of a particular stock with the rate of the second variation of the process logs of the stock. Thus we have to observe a particular stock during a not necessarily very long time interval It, t + T] on a very detailed time scale t t o <... < t N t + T and compare the squared implied volatility with the estimator for the second variation Finally, we note that there are several extensions of the Black and Scholes model studied in the literature, for instance models where the risk-free interest rate is time dependent or where the volatility depends on S and t explicitly [7]. It is not difficult to show that one can also extend several of these models by including a smoother process Z to account for long range dependence behavior and that similar conclusions can be made concerning the pricing of European options and the concept of volatility.