Chance Discovery in Stock Index Option and Futures Arbitrage

Abstract

The prices of the option and futures of a stock both reflect the market's expectation of futures changes of the stock's price. Their prices normally align with each other within a limited window. When they do not, arbitrage opportunities arise: an investor who spots the misalignment will be able to buy (sell) options on the one hand, and sell (buy) futures on the other and make risk-free profits. Historical data suggest that option and futures prices on the LIFFE Market do not align occasionally. Arbitrage chances are rare. Besides, they last for seconds only before the market adjusts itself. The challenge is not only to discover such chances, but to discover them ahead of other arbitragers. In the past, we have introduced EDDIE as a genetic programming tool for forecasting. This paper describes EDDIE-ARB, a specialization of EDDIE, for forecasting arbitrage opportunities. As a tool, EDDIE-ARB was designed to enable economists and computer scientists to work together to identify relevant independent variables. Trained on historical data, EDDIE-ARB was capable of discovering rules with high precision. Tested on out-of-sample data, EDDIE-ARB out-performed a naive ex ante rule, which reacted only when misalignments were detected. This establishes EDDIE-ARB as a promising tool for arbitrage chances discovery. It also demonstrates how EDDIE brings domain experts and computer scientists together.

Investigating the effect of different GP algorithms on the non-stationary behavior of financial markets

“…The simple GP was inspired by a financial forecasting tool, EDDIE [18], [19], which applies genetic programming to evolve a population of financial advisors, or, alternatively, a population of market-timing strategies, which guide investors on when to buy or hold. These market timing strategies are formulated as decision trees, which, when combined with the use of GP, are referred to as Genetic Decision Trees (GDTs).…”

“…The effectiveness of this constrained fitness function has been discussed in details in [19]. The constraint is denoted by R, which consists of two elements represented by percentage, given by…”

Evolving Decision Rules to Discover Patterns in Financial Data Sets

“…Finally, to complement our analysis, we have used the standard GP for the first example, a specialized GP proposed by Tsang et al (2005) for the second example and C.5 for the remaining example to compare with the results found by ECR. For the first example (which is an extremely imbalanced data set), our approach outperformed the traditional GP.…”

“…To detect important movements in financial stock prices, our approach is inspired by two previous works: EDDIE Tsang et al (1998), Li (2001), , Tsang et al (2005) and Repository Method Almanza and Tsang (2006b), Almanza and Tsang (2006a). EDDIE is a financial forecasting tool that trains a GP using a set of examples.…”

“…However, it is fair to say that the approach proposed by Tsang et al (2005), called EDDIE-ARB, achieved a precision of 100% and a recall of 42%. Nevertheless, our approach (ECR) achieved a precision of 96% with the same recall reported by Tsang and co-workers.…”