Complexity analysis of the stock market

Abstract

We study the complexity of the stock market by constructing $\epsilon$-machines of Standard and Poor's 500 index from February 1983 to April 2006 and by measuring the statistical complexities. It is found that both the statistical complexity and the number of causal states of constructed $\epsilon$-machines have decreased for last twenty years and that the average memory length needed to predict the future optimally has become shorter. These results support that the information is delivered to the economic agents and applied to the market prices more rapidly in year 2006 than in year 1983.