Mean square error for the Leland-Lott hedging strategy: convex pay-offs (Q650775): Difference between revisions

A method due to \textit{H. Leland} [``Option pricing and replication with transactions costs'', J. Finance 40, 1283--1301 (1985)] is among a few practical recipes how to price options under transaction costs. The portfolio strategy suggested by Leland generates the value process \[ V^n_t=\widehat C(0,S_0)+ \int^t_0 \sum^n_{i=1} H^n_{t_{i-1}} 1_{]t_{i-1},t_i]}(u)\,dS_u- \sum_{0< t_i< t} k_n S_{t_i}|H^n_{t_i}- H^n_{t_{i-1}}|. \] Here, \(S_t= S_0\exp(\sigma W_t- \sigma^2 t/2)\), \(0\leq t\leq 1\), is a Black-Scholes model (under the martingale measure) with maturity \(T= 1\). Given an option with pay-off \(G(S_1)\) (\(G\) being e.g. continuous with polynomial growth) the Black-Scholes arbitrage price at time \(t\) is \[ C(t,x)= E[G(x\exp(\xi\sigma(1- t)^{1/2}- \sigma^2(1- t)/2))],\quad 0\leq t\leq 1,\quad x> 0, \] where \(\xi\sim N(0,1)\). In the expression for \(V^n_t\), the investor holds \(H^n_{t_{i-1}}=\widehat C_x(t_{i-1}, S_{t_{i-1}})\) units of stock during the time interval \(]t_{i-1}, t_i]\), \(t_{i-1}= (i- 1)/n\), and \(\widehat C(t,x)\) arises from \(C(t,x)\) by replacing \(\sigma\) by \(\widehat\sigma> \sigma\), given by \(\widehat\sigma^2= \sigma^2+ \sigma k_0(8/\pi)^{1/2}\). Here, \(k_n= k_0n^{-1/2}\) is the transaction cost coefficient. In the case of a call option (i.e. \(V_1:= G(S_1):= (S_1- K)^+)\) Leland claimed (without proof) that \(V^n_1\to V_1\) in probability. In the present paper it is shown that \(V^n_1\to V_1\) even in \(L^2\), and, more precisely, \[ E[(V^n_1- V_1)^2]= A_1 n^{-1}+ o(n^{-1})\quad (n\to\infty),\tag{\(*\)} \] where the coefficient \(A_1\) can be explicitly calculated as a function of \(S_0\), \(\sigma\), \(K\), and \(k_0\). Under additional assumptions, the same asymptotics as \((*)\) holds for portfolio strategies with non-uniform grids and more general pay-off functions, and the authors even show that the distribution of the process \(Y^n:= n^{1/2}(V^n-\widehat C(t, S_t))\) converges in the Skorokhod space \({\mathcal D}[0,1]\) weakly to the distribution of the process \[ Y_t= \int^t_0 F(t,S_t)\,dW_t', \] where \(W'\) is an independent Wiener process, and \(F(t,x)\) is a function which can be explicitly defined.