Tel. +34 93 542 1942
Personal web page
Available for Interviews at
European Job Market for Economists (EEA), December 18-19, Rotterdam, The Netherlands
Allied Social Science Associations (ASSA), January 3-5, San Diego, US
Econometrics. Time Series Econometrics. Forecasting. Empirical Macroeconomics.
"Detecting Density Forecast Breakdowns" (Job Market Paper)
Awarded the IV Marcelo Reyes Award in the IXt Workshop in Time Series Econometrics, Zaragoza, 2019
This paper proposes a method to analyze the relationship between models’ in-sample fit and their out-of- sample density forecasting performance: it develops a formal test to capture density forecast breakdowns (DFBs), situations in which the out-of-sample density forecast performance is significantly worse than its anticipated performance. ’Model’ is understood in a wide sense, including both parametric and non-parametric models. For parametric models, the proposed test allows for model misspecification and takes into account parameter estimation uncertainty. For non-parametric models, conditions under which the estimation uncertainty is asymptotically irrelevant are provided. The proposed test accommodates conditional density forecast models, which nest unconditional density forecast models, and robust versions of this test are provided for practical use. Monte Carlo results indicate that the test has good size properties in moderately large samples and has power against changes in mean and variance, as well as changes in distribution type. The empirical study finds that: (i) DFBs occur sporadically in the lower quantiles of the one-quarter-ahead and one-year-ahead predictive conditional densities of real GDP growth in the US modelled with current financial and economic conditions and generalized skewed-t distributed errors; and (ii) DFBs occur in the one- day-ahead predictive densities for the S&P500, considering GARCH(1,1) and GARCH-t(1,1).
“Identification and Estimation of Parameter Instability in a High Dimensional Approximate Factor Model"
In modeling large panels of data, ignoring parameter instabilities may lead to severe consequences, such as inconsistent estimation and poor forecasting performance. This paper develops a new method for identifying and estimating structural breaks that occur at unknown common dates in the factor loadings in a high dimensional factor model that has an unknown number of latent factors. The methodology is based on the fact that the sum of the number of pseudo factors in the pre- and post-spit subsamples will be minimal if the sample is split at the structural break. It is then shown that the appropriate transformation of such criteria, which is based on the eigenvalue ratios of the covariance matrices of the pre- and post-split subsamples, provides consistent estimations of structural breaks in large panels. Importantly, the new method is robust to structural changes in the volatility of factors (i.e., the second moment of factors), and it can be easily extended to estimate multiple structural breaks. The results of the simulation study show that the new method works well in estimating moderately large breaks under different data-generating processes, and it compares favorably to an existing method (Baltagi et al. (2017)). In the empirical study, the new method is applied to a large panel of sectoral inflation rates. The results indicate two structural breaks in the factor loadings, which corresponded to the 1973 oil price shock and the 2008 financial crisis, respectively. In addition, combining this new method with the existing methods in the literature, a structural change in the volatility of factors is identified as occurring in January 1991.
Research Papers in Progress
“Heteroskedasticity and Autocorrelation Robust Inference in High Dimensional Linear Models” (with Geert Mesters)
“Financial Cycle Measurement with Mixed Frequency Data”
"VAR-based Granger-causality Test in the Presence of Instabilities'' (with Barbara Rossi)
Stata Journal. Forthcoming