Relationship between degree of efficiency and prediction in stock price changes

TL;DR

This paper empirically examines how the degree of market efficiency, measured by Hurst exponent and approximate entropy, relates to the prediction accuracy of future stock price changes across 27 markets, finding less efficient markets have higher predictability.

Contribution

It introduces the use of Hurst exponent and approximate entropy to quantify market efficiency and demonstrates their relationship with prediction power in stock markets.

Findings

Higher Hurst exponent correlates with better prediction accuracy.

Markets with lower efficiency (lower ApEn) show higher predictability.

Hurst exponent provides more significant predictive information than ApEn.

Abstract

This study investigates empirically whether the degree of stock market efficiency is related to the prediction power of future price change using the indices of twenty seven stock markets. Efficiency refers to weak-form efficient market hypothesis (EMH) in terms of the information of past price changes. The prediction power corresponds to the hit-rate, which is the rate of the consistency between the direction of actual price change and that of predicted one, calculated by the nearest neighbor prediction method (NN method) using the out-of-sample. In this manuscript, the Hurst exponent and the approximate entropy (ApEn) are used as the quantitative measurements of the degree of efficiency. The relationship between the Hurst exponent, reflecting the various time correlation property, and the ApEn value, reflecting the randomness in the time series, shows negative correlation. However, the average prediction power on the direction of future price change has the strongly positive correlation with the Hurst exponent, and the negative correlation with the ApEn. Therefore, the market index with less market efficiency has higher prediction power for future price change than one with higher market efficiency when we analyze the market using the past price change pattern. Furthermore, we show that the Hurst exponent, a measurement of the long-term memory property, provides more significant information in terms of prediction of future price changes than the ApEn and the NN method.