Identifying Sources of Autocorrelation

Autocorrelation, or serial correlation, occurs when observations in a time series are correlated with each other based on time lags. In regression analysis, particularly with time series data, autocorrelation in the error terms violates the assumption of independent errors, leading to biased standard errors and inefficient estimates.

Analysis of Potential Sources

Let's examine each potential source:

• I. Omitted explanatory variables: If relevant variables that influence the dependent variable are excluded from the model, their effects might be absorbed into the error term. If these omitted variables exhibit a time-series pattern (e.g., they are themselves autocorrelated), the error term will likely become autocorrelated. This is a common cause.
• II. Interpolation in statistical observation: When data points are estimated (interpolated) between actual observations, the process of interpolation can introduce or mimic serial dependence, especially if the underlying data has temporal structure. This can lead to autocorrelation in the observed residuals.
• III. Mis-specification of the true random term ‘v’: The error term ($u_t$) in time series models often follows a structure like $u_t = \rho u_{t-1} + v_t$, where $v_t$ is assumed to be white noise. If the actual process governing the random term is different from the one specified in the model (e.g., assuming white noise when it's actually autoregressive), the residuals will exhibit autocorrelation due to this mis-specification.
• IV. Economic variables to move together over time: While variables moving together (contemporaneous correlation or multicollinearity) is important, it doesn't directly cause the *error terms* to be correlated *over time*. It might contribute to omitted variable issues if not properly modeled, but it's not a direct source of autocorrelation itself.

Conclusion on Autocorrelation Sources

Based on the analysis, omitted variables (I), issues arising from interpolation (II), and incorrect specification of the error term's structure (III) are direct sources of autocorrelation. Therefore, items I, II, and III are valid sources.