Lagged Dependent Variable and Serial Correlation

Including a lagged dependent variable can introduce serial correlation, meaning that the error terms are correlated across time. This violates the assumption of independence required for Ordinary Least Squares (OLS) regression, leading to biased and inconsistent estimates. As a result, the reliability of the model's predictions and inferences is compromised.

The Durbin-Watson test is specifically designed to detect first-order serial correlation in the residuals of regression models. It analyzes the differences between consecutive error terms, providing a statistic that indicates the presence and extent of autocorrelation, making it the most commonly used diagnostic tool for this purpose.

A Durbin-Watson statistic of 2 indicates that there is little to no first-order serial correlation in the residuals of a regression model. Values around 2 suggest that the residuals are independent, while values significantly below or above 2 indicate positive or negative serial correlation, respectively.

In a lagged dependent variable model, including Y_{t-1} as a regressor can lead to biased estimates if the error term u_t is serially correlated. This is because the past values of the dependent variable (Y_{t-1}) can be correlated with the error term, violating the OLS assumption of no correlation between regressors and the error term.

In a lagged dependent variable model with serial correlation, OLS coefficient estimates remain unbiased because OLS does not rely on the assumption of independence of errors. However, the standard errors may be biased, leading to unreliable inference. Thus, while the estimates are unbiased, the associated standard errors are not necessarily accurate.

The Breusch-Godfrey test assesses serial correlation by regressing the residuals from a regression model on both the original regressors and the lagged residuals. This approach allows it to identify not just first-order autocorrelation, but also higher-order autocorrelation, making it a comprehensive tool for diagnosing residual patterns in regression analysis.

When a lagged dependent variable is included in a regression model with serially correlated errors, the bias in the OLS estimator arises from both the strength of the serial correlation and the coefficient of the lagged variable. This interaction affects how accurately the model estimates relationships and can lead to misleading results if not addressed.

When a model includes a lagged dependent variable and exhibits serial correlation in the errors, ordinary least squares (OLS) can produce biased and inconsistent estimates. Instrumental variables (IV) estimation helps to address these issues by using instruments that are correlated with the endogenous explanatory variables but uncorrelated with the error term, leading to more reliable parameter estimates.

In Arellano-Bond dynamic panel estimation, lagged levels of the dependent variable are used as instruments to mitigate endogeneity issues. This approach leverages past values to provide valid instruments, allowing for consistent estimation of the model parameters despite the presence of a lagged dependent variable that may be correlated with the error term.

When serial correlation is present, the OLS estimator remains unbiased for the coefficients of independent variables, including lagged dependent variables. However, the standard errors may be biased, which affects inference. The bias in the OLS coefficients does not decrease with an increasing sample size; instead, the presence of serial correlation can lead to inefficient estimates.

The Durbin-Watson test is designed to detect serial correlation in residuals, but when a lagged dependent variable is included in the model, it complicates the interpretation of the test statistic. The presence of this lagged variable can distort the relationship between residuals, making the test results unreliable for assessing serial correlation.

Ignoring serial correlation in the errors can lead to incorrect estimates of the regression coefficients. In this case, the lagged inflation may be correlated with the error term, causing the estimated coefficient to be biased downward. This occurs because the model fails to account for the persistence of inflation over time, underestimating its effect.