RDP 2008-02: Combining Multivariate Density Forecasts Using Predictive Criteria 6. Conclusion

In this paper, we have looked at a relatively unexplored area of the forecast combination literature, that of combining multivariate density forecasts. We have used predictive-likelihood scores to combine density forecasts produced by a suite of models consisting of a BVAR, a FAVAR and a DSGE model. The weighting scheme suggests that the DSGE model should be assigned a very low weight in the combined density forecast. Inspecting the probability integral transforms of the models' forecasts suggests that this low weight is due to the fact that over the evaluation sample the DSGE produced density forecasts that were too wide when compared with the actual distribution of observations. Overall, testing the performance of the combined density forecasts returned mixed results, and it is not clear that the combined forecasts are superior to those of the best-performing individual model or an equal-weighting scheme. This may be a result of the short sample available to evaluate the forecasts.