Exchange rate predictability and dynamic Bayesian learning

Abstract

We consider how an investor in the foreign exchange market can exploit predictive information by means of flexible Bayesian inference. Using a variety of vector autoregressive models, the investor is able, each period, to learn about important data features. The developed methodology synthesizes a wide array of established approaches for modeling exchange rate dynamics. In a thorough investigation of monthly exchange rate predictability for 10 countries, we find that using the proposed methodology for dynamic asset allocation achieves substantial economic gains out of sample. In particular, we find evidence for sparsity, fast model switching, and exploitation of the exchange rate cross-section.This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

Real-Time Density Nowcasts of US Inflation: A Model-Combination Approach

“…This sharp distinction implies that the density nowcast assessed as the most accurate gets a very high score, which translates into a substantially higher average score and, in turn, significantly higher weight in the combination, possibly resulting in a fast-switching pattern. 33 Recent research generally views this feature of fast model-switching as desirable, and so a framework that allows for it is viewed favorably (see Beckmann et al, 2020).…”

“…This approach amounts to "learning from past mistakes" and is widely used in the density combination literature due to its simplicity (e.g.,Gerard and Nimark, 2008;Jore, Mitchell, and Vahey, 2010;Kascha and Ravazzolo, 2010; Bjornland et al, 2011;Garratt et al, 2011;Aastveit et al, 2014;Beckmann et al, 2020). While we use the entire expanding history, we also explored the strategy of computing the weights over rolling 12-month periods, to "learn from recent mistakes" in computing the average score.…”

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

“…For instance, going back to the generalized Phillips curve mentioned above, there are a myriad of theories that suggest a link between variables, such as the unemployment rate, T-bill rates, level of economic activity, house prices, and the rate of inflation. Therefore, the practitioner is faced with the situation, where he (she) the purpose of our study is not to list every single publication (or working paper) that applies DMA in one way or another, we can list the following interesting applications: Dangl and Halling (2012), Liu et al (2015), and Naser and Alaali (2018) with regard to predicting aggregate equity returns; Koop and Tole (2013) in the context of forecasting the spot price of carbon permits; Buncic and Moretto (2015), Drachal (2016), and Naser (2016) with regard to predicting commodity prices; Bruyn et al (2015), Beckmann and Schüssler (2016), Byrne et al (2018), and Beckmann et al (2020) in the context of forecasting exchange rates; Gupta et al (2014) with regard to forecasting foreign exchange reserves; Bork and Møller (2015), Risse and Kern (2016), and Wei and Cao (2017) in the context of forecasting house price changes; Aye et al (2015) and Baur et al (2016) with regard to predicting the rate of return on the price of gold; Koop and Korobilis (2011) and Filippo (2015) with regard to forecasting non-U.S. rate of inflation; Byrne et al (2017) with respect to forecasting the term structure of government bond yields; and Wang et al (2016), Liu et al (2017), Nonejad (2017b), and Ma et al (2018) with respect to forecasting equity return and commodity price volatility.…”

Interdependence or contagion: A model switching approach with a focus on Latin America

“…However, our approach allows us to establish the existence and magnitude of DIs and SIs over time. 1 In terms of the contagion literature, our model switching approach relates to studies which use regime switching methods and time-varying parameters (TVP). Regime switching methods can capture crisis and non-crisis regimes without arbitrarily specifying when break dates occur (see, among many others, Gravelle, 2006 as an early example and Casarin et al, 2018 and Chan et al, 2018 for recent examples).…”

“…Our approach incorporates both the abrupt change seen in regime switching models and gradual change seen in TVP models. We achieve this by estimating the degree of model switching and time-variation in parameters at each point in time following Beckmann and Schüssler (2016) and Beckmann et al (2018). Moreover, while we allow for model switching we are not restricted to focussing on two regimes.…”