Time Series Data and Autocorrelation
Samprit Chatterjee, Jeffrey S. Simonoff · 2012
This chapter discusses some issues related to building regression models for time series data. It first talks about the effects of autocorrelation if it is ignored. Then, the chapter examines several approaches to identifying autocorrelation, which range from one requiring strong assumptions to one related to a simple graphical examination of residuals that requires virtually no assumptions. The chapter considers several relatively simple approaches to account for common forms of autocorrelation, including trends and seasonal effects, and explores how values from previous time periods can be used to enrich a regression model and account for autocorrelation. It describes three tests for detecting autocorrelation that range from strongly parametric to nonparametric: the Durbin-Watson statistic, autocorrelation function (ACF), and residual plots and the runs test. The chapter concludes with discussion of a more sophisticated approach to handling autocorrelation that moves past ordinary least squares estimation to estimation designed for time series data. Controlled Vocabulary Terms Durbin-Watson statistic; Durbin–Watson statistic; partial residual plot; runs test; time series