This version of the paper is from 1994.
Abstract
Recent empirical investigations have found a decrease over time in the estimated first order autocorrelation of stock index returns. It has been suggested that this change in autocorrelation is caused by the introduction of new financial markets, such as options and futures on the index. (Froot & Perold (1990)).
This paper carries out formal hypothesis tests for changes in index autocorrelation, and evaluates the relationship between the estimates of autoregressive coefficients with those of other second moments. The methods used are tests for structural changes in time series GMM estimation. The small sample characteristics of the test statistics are investigated by Monte Carlo, and the tests are carried out on US stock market data for the period 1976 to 1989.
The results show that if we impose a constant conditional variance, we can reject a null of no change in autoregressive coefficients of the S{\&}P 500 index. If we allow for a changing conditional variance, we no longer reject a null of no change. Hence, correcting for changes in the variance, we can not claim there has been a change in the autoregressive relationship. To explore the causes of these results, we also looked at cross-correlations between different--sized indices, and found that there has been a change in the lead-lag relationship between the portfolio of largest stocks and other size-based portfolios.
The paper is available as a pdf file.