ABSTRACT
Autocorrelation is a time-series problem that occurs frequently with economic data. Second-order serial correlation is much less frequently encountered. As you might imagine, second-order serial correlation is when the current error term is related to the two prior error terms: tttt eee HUU 2211 11.2 Consequences of Serial Correlation Serial correlation is a direct violation of assumption 3 of the Classical Linear Regression Model: The error terms are not related to one another: E[ui uj] = 0 for all i j. Recall that assumption 3 was required to prove that the estimators are best, but not unbiased. Therefore, in the presence of autocorrelation the structural parameters are not best. In addition, the residual statistics from the regression are biased:AE SER, S.E.(B's), T-ratios, F, R2 - - + + + The signs under each statistic indicate the direction of the bias. Notice that each statistic is biased in a way that makes it appear better. For instance, the standard error of the regression (SER) indicates a good fit when it is low and it will be biased downward in the presence of autocorrelation. The consequences of serial correlation are the same as those of heteroskedasticity with one small difference. In the presence of serial correlation, the direction of the bias in the residual statistics is known. With heteroskedasticity, the direction of the bias is unknown. Since the standard errors and t-ratios are biased, hypothesis tests must be carried out with extreme caution in the presence of serial correlation. TYPE I errors will be more likely because the standard errors are biased downward.
