Model uncertainty and forecast accuracy

Forecasting trend output in the Euro area

“…The studies that report uncertainty regions around trend output measures or the output gap are usually only ex-post and do not give exact information about how the uncertainty bands are calculated (Duspaquier et al, 1999;Gerlach and Smets, 1999), especially if and how estimation uncertainty is considered. This implies that in contrast to investigations where observable variables are modelled and forecasted (Chatfield, 1996), the forecasting of trend output often includes model uncertainty but neglects estimation uncertainty. That is why estimation uncertainty is emphasized here.…”

Can High‐frequency Trading Strategies Constantly Beat the Market?

“…where Y t represents the actual values and Ŷ τ is the forecasted values. Chatfield (1996) argue that the real test of a forecasting model is its ex ante predictive ability and models with high in-sample explanatory power usually do not have high out-of-sample fit because of model overfitting (Bossaerts and Hillion, 1999). We compute the modified Diebold-Mariano (MDM) test (an extension of the Diebold and Mariano, 1995, test) to verify out-of-sample the statistical significance of the previously discussed analysis.…”

Thompson Sampling: Endogenously Random Behavior in Games and Markets

“…In fact, the ultimate test of a model is considered to be (pseudo) out-of-sample forecasting. (Chatfield, 1996;Stock and Watson,2015, p. 613) This is a common methodology both in microeconomics (e.g. Camerer and Ho (1999)) and macroeconomics (see Clark and McCracken 18 For Thompson sampling the median deviations from Nash equilibrium for the row players (−2.53 • 10 −6 ) and the column players (−1.56•10 −7 ) were insignificantly different from zero (p-value Wilcoxon signed-rank test: 0.6250 for ROW; 0.8457 for COL.) For the Bayesian logit the median deviations from Nash equilibrium for the row players (−0.0426) and the column players (0.0061) were also insignificantly different from zero.…”