Financial conditions and density forecasts for US Output and inflation

The authors reassess the predictive power of financial indicators for output and inflation in the US by studying predictive densities generated by set of linear and nonlinear forecasting models.  They argue that, if the linkage between financial and real economy is state-dependent as implied by standard models with financial frictions, predictive densities should reveal aspects of the co-movements between financial and macroeconomic variables that are ignored by construction in an ordinary (central) forecasting exercise.  The authors study the performance of linear and nonlinear (Threshold and Markov-Switching) VARs estimated on a monthly US dataset including various commonly-used financial indicators.  We obtain three important results.  First, adding financial indicators to an otherwise standard VAR improves both central forecasts and predictive distributions for output, but the improvement is more substantial for the latter.  Even in a linear model, financial indicators are more useful in predicting 'tails', or deviations of output and inflation from their expected paths, than 'means', namely the expected paths themselves.  Second, nonlinear models with financial indicators tend to generate noisier central forecasts than their linear counterparts, but they clearly outperform them in predicting distributions.  This is mainly because nonlinear models predict the likelihood of recessionary episodes more accurately.  Third, the discrepancies between models are themselves predictable: a Bayesian forecaster can formulate a reasonable real-time guess on which model is likely to be more accurate in the near future.