Detecting and Correcting Autocorrelation in Time-Series Models of Unemployment Rates

This paper looks at the existence and correction of autocorrelation in a time series regression modelwithin Using Or- dinary Least Squares (OLS) and the Cochrane-Orcutt approach. Initial research showed positive autocorrelation in the residuals, as shown by a Durbin-Watson statistic of 1.21 and notable findings from the Breusch-Godfrey LM test. The Ljung-Box Q-statistics at lesser lags also verified the presence of autocorrelation; low p-values signaled statistical relevance. The initial OLS model failed the residual normality test and had an R² is 0.62. The Cochrane-OrcuttAR(1) approach was used to solve these problems. The revised model's Durbin-Watson score of 2.03 indicates the effective removal of autocorrelation. Moreover, changes in the R² (up to 0.71), AIC (down to 13.48), and residual normalcy were noted. Randomly distributed around zero, the adjusted residuals supported the white noise hypothesis. These results highlight the need of recognizing and fixing autocorrelation in order to increase the dependability and statistic- al validity of regression models.