Fixed effects with interpersonal and intertemporal covariance

The paper considers the estimation of a fixed effects time series-cross section model where errors have both unspecified interpersonal and intertemporal covariance. Efficient estimators in the form of GLS are suggested, which can be implemented on the data in their actual form or in deviations from time-means. As an empirical example, the determinants of new residential construction activity in the New England states of the US during the 80's are investigated. Results show substantial sensitivity to changes in the interest rate and the unemployment rate, whilst responses to changes in income are more muted. Long term factors related to the region are also influential.