Forecasting Inflation Using Dynamic Model Averaging*

Abstract

We forecast quarterly US inflation based on the generalized Phillips curve using econometric methods that incorporate dynamic model averaging. These methods not only allow for coefficients to change over time, but also allow for the entire forecasting model to change over time. We find that dynamic model averaging leads to substantial forecasting improvements over simple benchmark regressions and more sophisticated approaches such as those using time varying coefficient models. We also provide evidence on which sets of predictors are relevant for forecasting in each period.

Yield curve forecast combinations based on bond portfolio performance

“…This environment gives rise to at least two possible economic decisions: (i) use available information of time t−1 to select the model with the highest chance of being the best-performing model in period t; or (ii) use a combination of forecasts, where combination weights are based on the chances of each candidate model to deliver the highest return next period. The first alternative is related to what Koop and Korobilis (2012) labeled as DMS, while the second option is related to the DMA (Raftery et al 2010).…”

“…Additionally, Pesaran and Timmermann (2000) document how the best models to forecast stock returns change over time, and argue that it does not seem plausible that the real-time search for return predictability will be conducted using only one model. Koop and Korobilis (2012) use the DMA approach of Raftery et al (2010) to forecast inflation, and argue that it might be better to use only the model with highest predictive probability in t as the period-t + 1 forecasting model. The rationale for trimming forecast is to avoid giving weights to models that perform poorly and add noise to forecasts of period t (Granger & Jeon, 2004).…”

“…These probabilities are then used to determine the forecast combination weights. The dynamic model selection (DMS) approach proposed by Koop and Korobilis (2012) is based on the approach of Raftery et al, but instead of combining models according to their predictive probabilities it simply selects the model with highest predictive probability as the forecasting model for period t+1.…”

“…First, we follow Raftery et al (2010) and Koop and Korobilis (2012), and implement DMA and DMS using predictive likelihoods obtained from each individual forecast model. This approach can be straightforwardly replicated in our setting since it is possible to write down the likelihood function of all models considered in our empirical exercise.…”

Hierarchical Shrinkage in Time-Varying Parameter Models

“…Adding time variation in parameters may exacerbate this problem, suggesting that shrinkage may be useful with TVP models. However, there have been relatively few papers which attempt to ensure shrinkage in TVP models (exceptions include Korobilis, 2009 andKoop, Leon-Gonzalez andStrachan, 2009). …”