Applying Genetic Regulatory Networks to Index Trading

Abstract

Abstract. This paper explores the computational power of genetic regulatory network models, and the practicalities of applying these to realworld problems. The specific domain of financial trading is tackled; this is a problem where time-dependent decisions are critical, and as such benefits from the differential gene expression that these networks provide. The results obtained are on par with the best found in the literature, and highlight the applicability of these models to this type of problem.

Dynamic Index Trading Using a Gene Regulatory Network Model

“…The trading methodology is based in previous studies [7,14,12], where a trader issues buy, sell, or do nothing signals for each day of the training or test periods. Starting with a capital of $10000, if a buy signal is issued, 10% of the total funds (initial capital plus earnings) are invested in the index; this position is automatically closed after a ten day period.…”

“…In this work, a model originally presented by Wolfgang Banzhaf [1] is used; it was shown to exhibit similar dynamics to real world GRNs [2], and has been applied to dynamic control problems (such as the pole-balancing benchmark [13] and index trading [12]). …”

“…The concentration of proteins produced by P-genes can then be used as output signals of the system. The approach taken here is the same as in previous studies [12].…”

“…The work presented here follows closely the methodology of previous applications of Grammatical Evolution [14,3] and GRNs [12] to index trading, and uses three datasets, from the UK FTSE 100 index, the Japan Nikkei index, and the German Dax index. To keep the results comparable, all data is drawn from the period between 16/4/1991 and 21/10/1996; Fig.…”

“…In previous studies [3,12], the data was pre-processed prior to evolution, with the raw prices initially transformed into a moving average with a 75 day gap, and then normalised into the range of 0 to 1. However, as the current study focuses on application to real market trading, both of these pre-processing steps are troublesome, as detailed below, and hence were not used.…”