Buy-low and sell-high investment strategies (Q2847244)

Assuming that the price \(X\) of a stock evolves in accordance with the SDE NEWLINE\[NEWLINE dX_t=b(X_t)dt+\sigma(X_t)dW_t, \qquad X_0=x>0, NEWLINE\]NEWLINE consider an investor faced with the problem of sequentially buying and selling one unit of the stock. The process \(Y\) describing the net holdings of the stock is then a finite variation (pure jump) \(\{0,1\}\)-valued process starting at \(Y_0=y\). Upon buying (resp. selling) the stock the investor pays (resp. earns) the price \(X_t+c_b\) (resp. \(X_t-c_s\)). The objective of the investor is that of maximizing the asymptotic expected, net present value of all future payments arising from \(Y\). The authors study the HJB equation associated with this problem and characterize the existence of a (unique) solution in Lemma 3.2. In Theorem 3.3, the main result of the paper, a complete and explicit characterisation of the solution is obtained. The relative simplicity and intuitive form of the solution are a remarkable feature of this problem. The authors deduce a number of clarifying examples. If, e.g., \(0\) is a lower boundary point for the process (when the diffusion is, say, a geometric Brownian motion) then the existence of fixed transaction costs leads to the conclusion that it is never optimal to buy the stock and the optimal strategy consists of liquidating the initial holding (if positive) as soon as the process crosses some exogenous threshold.