Adaptive models and heavy tails with an application to inflation forecasting

Abstract

This paper introduces an adaptive algorithm for time-varying autoregressive models in the presence of heavy tails. The evolution of the parameters is determined by the score of the conditional distribution, the resulting model is observation-driven and is estimated by classical methods. In particular, we consider time variation in both coefficients and volatility, emphasizing how the two interact with each other. Meaningful restrictions are imposed on the model parameters so as to attain local stationarity and bounded mean values. The model is applied to the analysis of inflation dynamics with the following results: allowing for heavy tails leads to significant improvements in terms of fit and forecast, and the adoption of the Student-t distribution proves to be crucial in order to obtain well calibrated density forecasts. These results are obtained using the US CPI inflation rate and are confirmed by other inflation indicators, as well as for CPI inflation of the other G7 countries.JEL classification: C22, C51, C53, E31.

Modeling and Forecasting Macroeconomic Downside Risk

“…On the other hand, when the asymmetry parameter is zero (red line), the weights display the classic outlier-discounting typical of the observation driven models with Student-t distributions (see, e.g. Harvey and Luati, 2014;Delle Monache and Petrella, 2017). When the distribution is positively (negatively) skewed, i.e.…”

An Overview of Dynamic Model Averaging Techniques in Time‐series Econometrics

“…For instance, with respect to forecasting the rate of inflation, an interesting topic can be to revisit Koop and Korobilis (2012) and evaluate how DMA using a large number of predictors fares against the bounded trend inflation model with stochastic volatility effects suggested in Chan et al . (2013) and the GAS‐based model suggested in Delle Monache and Petrella (2017) using only a few predictors, respectively. For example, one can augment the models suggested in Chan et al .…”

“…For example, one can augment the models suggested in Chan et al . (2013) and Delle Monache and Petrella (2017) with the unemployment rate and evaluate whether DMA based on the same number of predictors as in Koop and Korobilis (2012) is able to outperform these more complex models. Likewise, with respect to density forecast accuracy, instead of the current literature that predominately relies on the log‐score criterion, one can follow studies, such as Groen et al .…”

Integer‐Valued Autoregressive Models With Survival Probability Driven By A Stochastic Recurrence Equation

“…The GAS‐INAR model can be a useful means to fill this gap. In the context of GAS models, Delle Monache and Petrella and Blasques et al proposed a class of AR models that feature time‐varying coefficients driven by the score of the predictive log‐likelihood. The GAS‐INAR model can also be seen as an integer‐valued version of their model.…”