In the context of econometrics, what is the meaning of no autocorrelation of residuals?

Answer 1

Hello,

This is a great question and an important issue to understand. When data is such that the current observation is linearly correlated to the past observations, that data is said to exhibit autocorrelation. In time series data, autocorrelation is correlation between observations of the same dataset at different points in time. The need for distinct time series models stems in part from the autocorrelation present in time series data.

When modeling data, one intuitive way to identify autocorrelation is to look for patterns in the residuals. For example, if there appears to be persistence in the residuals -- like negative residuals tend to stay negative -- there is likely autocorrelation. This is one of the many reasons that residual plots can be helpful for model diagnostics.

Identifying autocorrelation is important because it has modeling consequences. Ordinary least squares assumes that there is no autocorrelation in the error terms of a series. When autocorrelation is present :

1. OLS estimators are valid. (Estimates Will be consistent)
2. Traditional OLS standard errors are smaller than true standard error and inference tests are no longer valid. (Estimates are inefficient)

Some resources may refer to autocorrelation and serial correlation as the same thing. However, often in econometrics we use the term autocorrelation when referring to time series data and serial correlation when referring to cross-sectional data.