Mean reversal for stochastic hybrid systems (Q1003546)

The author considers two finite state discrete parameter Markov chains \(X\) and \(Y\) with \(\pm 1\) win or lose payoff subject to transition between states. A new process is obtained by choosing at random by flipping a fair coin either \(X\) or \(Y\) at each time step. It is shown that a so-called mean reversal can arise, that is, even if the cumulative expected payoffs for \(X\) and \(Y\) are decreasing in time, the expected payoff for the randomized process can become increasing in time. This result generalizes the idea of combining two losing games into a winning one, known as Parrando's Paradox.