Alwyn Hoffman

By Alwyn J. Hoffman

Arbitrage is an approach to portfolio composition that times the selection of investment instruments based on deviations between price and true value, exploiting the expected movement of price toward true value. Pairs trading is an early form of statistical arbitrage, exploiting temporary deviations from equilibrium for instruments that tend to move together. To extend the potential investment, universe statistical arbitrage was extended to include several co-integrated instruments that display mean reverting model errors. A weakness of these methods is the possible breakdown of the observed relationships when the model error becomes a random walk, resulting in trading losses. Partial co-integration addresses this limitation by allowing for model errors that include both random walk and mean-reverting components. As both the mean-reverting and random walk components are non-observable, a Kalman filter approach is employed to estimate these system states. Partial co-integration is shown to outperform market returns and normal co-integration. Higher returns are observed during bear cycles, making it an attractive strategy to combine with bull market strategies. It is furthermore illustrated how partial co-integration can be further improved using convolutional neural networks to predict breakdowns in mean reversion and reinforcement learning to optimize the levels for entering and exiting trades.

Part of the book: Applications and Optimizations of Kalman Filter and Their Variants