Serial correlation, also known as autocorrelation, describes the relationship between observations on the same variable over different periods of time. This is different from traditional correlation, which compares multiple variables over one period of time. Technical analysts and investors use serial correlation to measure how well past price movements can predict future movements for the same asset, a crucial concept in technical stock market analysis. Because serial correlation is largely dependent on the time interval used, common examples of serial correlation are difficult to qualify. However, one well-known serial correlation among traders is called the "January Effect," whereby returns tend to be larger in January than any other month of the year.
Serial correlation is a function of mean and variance; it can never be absolute and relies heavily on circumstance and interpretation. Even if there was 100% positive correlation, or mean aversion, or 100% negative correlation, or mean reverting, between an asset's price action over time, there is still no law dictating any such correlation must continue. Countless studies have been performed by financial analysts and econometricians to discover serial correlation among price changes in markets, stocks or portfolios, but these have generally yielded insignificant insights.
Serial correlation suggests the returns distributed across observations are not strictly random. Even if the notion that price changes in period A have something to tell traders about price changes in period B is deeply ingrained in the framework of technical analysis, the actual existence and nature of any such correlation is debated among serious statisticians.
Famous studies conducted by Fama (1965), Jennergren and Korsvold (1974) and Cootner (1961) looked at stocks and commodities over time and found very low or insignificant serial correlation. However, long-term studies about whole markets suggest a substantial negative serial correlation, indicating markets tend to reverse themselves over long periods. The first major work in this area was reported by Fama and French in 1988.