Predictability, long memory and non linear dynamics of stock returns

Abstract

The presence of long-range dependence and nonlinear dynamics in stock returns is examined using data from the Athens Stock Exchange. The authors apply (among other well-known time series techniques) the Rescaled Range (R/S) Statistic AND THE modified R/S statistic, as being more appropriate for tracking short and long memory in the stock market and as more robust to other alternative methods. The results support evidence of short memory in all single stock returns in the sample used but there is an agreement between the two types of R/S statistics regarding existence of strong long memory in the squared returns. The findings are in agreement with recent empirical evidence investigating long-range and short-range statistical dependence in other stock exchanges apart from the Greek stock market. Furthermore, nonlinear dynamics are supported in the results from evidence of strong conditional autoregressive evolution in the stock return path. The authors consider these results as indicative of a rapidly developing and strengthening market and not a symptom of inefficiencies in the Greek capital market.