Combining the predictive ability of factorial analysis and transfer functions for vat revenue forecasting

Resumen

We propose a two-step methodology combining factorial analysis and a dynamic regression model to produce a valid forecast for the Spanish VAT revenue. Instead of using final consumption expenditure as the only explanatory variable in a transfer function, we propose a set of indicators covering different areas of the economy (General, Construction, Labour Market and Service Activity Indicators). The idea is to enforce parsimony and to avoid multicollinearity with little information loss by performing principal component analysis and regressing not on the full set of indicators but rather on the first two principal components. We apply the proposed method to quarterly data beginning in 1995 and ending in 2014, providing out of sample estimations for the four quarters of 2015 and the first quarter of 2016.