Multiperiod-Ahead Predictive Densities and Model Comparison in Dynamic Models
This study proposes a new model comparison method which uses multiperiod-ahead predictive densities. Use of predictive densities allows us to incorporate the uncertainties associated with point forecasts in comparing the forecasting performances of various models. In the special case where one-period-ahead predictive densities are used, the method is equivalent to the Bayesian posterior odds. This new model-comparison method is contrasted to other measures of forecasting performance, such as the mean squared error, which don’t consider the uncertainties associated with point forecasts. To evaluate the multiperiod-ahead predictive densities in dynamic models, this study uses simulation methods.
KeywordsRoot Mean Square Error Training Sample Posterior Density Benchmark Model Future Outcome
Unable to display preview. Download preview PDF.
- Berger, JO, and LR Pericchi (1992), “The Intrinsic Bayes Factor,” Purdue University Department of Statistical ReportGoogle Scholar
- Chow, GC (1973), Multiperiod Predictions from Stochastic Difference Equations by Bayesian Methods,Econometrica, 41, 109–118.Google Scholar
- Friedman, M, and A Schwartz (1991), Alternative Approaches to Analyzing Economic Data,American Economic Review81, 39–49Google Scholar
- Gelfand, AE, DK Dey, and J Chang (1992), “Model Determination using Predictive Distributions with Implementation via Sampling-Based Methods,” Bayesian Statistics4, JM Bernardo, JO Berger, AP Dawid, and AFM Smith, eds., Oxford: Clarendon Press, 147–167Google Scholar
- Geweke, J (1994), “Bayesian Comparison of Econometric Models,” Manuscript.Google Scholar
- Min, C (1995), “Forecasting the Adoptions of New Consumer Durable Products,” Manuscript.Google Scholar
- Thompson, P.A., and R.B. Miller (1986), “Sampling the Future: A Bayesian Approach to Forecasting From Univariate Time Series Models,” Journal of Business & Economic Statistics, 4, 427–436Google Scholar