Abstract: This paper proposes a new least absolute shrinkage and selection operator (lasso) for estimating panel vector autoregressive (PVAR) models. By allowing for interdependencies and heterogeneities across cross-sectional units, typically the number of parameters of PVAR models is too large to estimate using ordinary least squares. The penalized regression this paper introduces ensures the feasibility of the estimation by specifying a shrinkage penalty that contains time series and cross section characteristics; thereby, accounting for the inherent panel structure within the data. Furthermore, using the weighted sum of squared residuals as the loss function enables the lasso for PVAR models to take into account correlations between cross-sectional units in the penalized regression. Given large and sparse models, simulation results point towards advantages of using lasso for PVARs over OLS, standard lasso techniques as well as Bayesian estimators in terms of mean squared errors and forecast accuracy. Empirical forecasting applications with up to ten countries and four variables support these findings.
presented at (selected): EEA-ESEM 2017, Lisbon, Portugal; Barcelona GSE Summer Forum Time Series Econometrics and Applications for Macroeconomics and Finance, Barcelona, Spain; Econometrics Seminar, University of Sydney, Australia; Econometrics Seminar, University of Melbourne, Australia; Spring Meeting of Young Economist 2017, Halle, Germany; VfS Annual Conference 2017, Vienna, Austria
Abstract: The paper introduces the stochastic search variable selection for panel vector autoregressive models (SSVSP). The proposed selection prior allows for a data-based restriction search ensuring the estimation-feasibility. The SSVSP differentiates between domestic and foreign variables, thereby allowing a flexible panel structure and extending Koop and Korobilis's S4 to a restriction search on single elements. Absent a matrix structure for restrictions, a Monte Carlo simulation shows that SSVSP outperforms S4 in terms of deviation from the true values. Furthermore, a forecast exercise for G7 countries demonstrates that forecast performance improves for SSVSP focusing on sparsity in form of no dynamic interdependencies.
presented at (selected): IAAE 2016 Annual Conference, Milan, Italy; European Seminar on Bayesian Econometrics, Venice, Italy; CFE, Seville, Spain; BAM RG Seminar, University of Melbourne, Australia; VfS Annual Conference 2016, Augsburg, Germany
Abstract: This paper assesses the macroeconomic effects of international monetary policy transmission for the United States, the United Kingdom and the Euro area. We use a Bayesian Proxy Panel SVAR in order to capture international linkages and to trace the dynamic responses of the macroeconomic variables. A specific selection prior incorporating the panel dimension allows the estimation of the large number of parameters in the PVAR model. The monetary policy shocks of the three regions are identified via changes in daily future contracts around policy announcement dates. We use changes in stock prices as second proxies combined with sign restrictions to disentangle central bank information shocks from the monetary policy surprises.
presented at (selected): ESOBE 2019, St Andrews, Scotland; CFE, London, England; FU Berlin Research Seminar