Fiscal Readjustments in the US: A Non-linear Time Series Analysis

We analyze the fiscal adjustment process in the US using a multivariate threshold Vector Error Regression Model (VECM). We find that the shift from a single equation to multivariate setting adds value both in terms of our economic understanding of the fiscal adjustment process in the US and the forecasting performance of non-linear models. First, we find evidence that fiscal authorities will intervene to reduce real per capita deficit only when it reaches a certain threshold and that the fiscal adjustment process takes place primarily by cutting government expenditure rather than increasing tax revenues. Second, the out-of-sample density forecast and probability forecasts results suggest that a shift from a univariate AR model specification to a multivariate model improves forecast performance. We also find that the forecasting performance of both linear and non-linear VECM is similar for long horizons (e.g. two years ahead).