Estimating Long Memory Time-Series-Cross-Section Data

This paper extends the MD (multiple differenced) methodology of Tsay (2006) to estimate a class of time-series-cross-section (TSCS) models consisting of stationary or nonstationary long memory regressors and errors, while allowing for correlations and heteroskedasticities in both cross-section and time dimensions. Interestingly, the regression coefficients of these models still can be easily tested with the MD-based approach using the critical values from the standard normal distribution. Under various combinations of long memory processes and cross-section dimensions, the finite sample performance of the MD-based method is promising even though the time span is only 20. We then apply this method to reexamine the data of Hicks and Swank (1992). The testing results are more in line with the findings in Beck and Katz (1995) whereby the evidence for positive voter turnout effects in Hicks and Swank (1992) is no more highly statistically significant when the number of differencing is greater than or equal to 1.