On measuring serial correlation: Implications for finance and accounting research studies

Accounting and finance studies that measure serial correlation implicitly make two assumptions. One, the studies assume that the sample estimate of the autocorrelation coefficient is unbiased. The assumption is intuitively appealing, but incorrect. This article provides a measure of the size of the bias. Two, the studies assume that the target of the time series is constant over time. However, over a long period target values may change. This article models the general case in which not only do random shocks affect actual values, but also random changes affect target values.