My main fields of research are macroeconometrics and financial econometrics, with a particular interest in forecasting, risk management, and machine learning. Research Statement can be provided upon request.
Superior Predictive Ability in Unstable Environments
with an Application to Downside Risk Forecasts
This paper introduces the Fluctuant-SPA (FSPA) test, a methodology for evaluating superior predictive ability (SPA) in unstable environments. The highlight of the FSPA test is that it allows to detect superior predictive ability when it is time varying. We showcase our test using a comprehensive assessment of downside risk forecasts to the U.S. economy over a 45-year span, considering a large collection of forecast methodologies. A number of findings emerge from the empirical application. First, there is substantial heterogeneity in forecasting performance across time. Second, the quantile regression equipped with a financial conditions index -- a major benchmark in this literature -- outperforms its alternatives after the Global Financial Crisis (2007-2009), yet it is surpassed in the periods leading up to the crisis. These local findings contrast with recent global assessments where this benchmark was found inferior to several alternatives. Overall, the empirical application showcases that the FSPA is particularly useful for forecast evaluation in unstable environments.
*Earlier versions of this paper were presented at 2023 BSE Summer Forum (Workshop on Macroeconomics and Policy Evaluation; poster session), Barcelona, Spain; 3rd International Econometrics PhD Conference, The Econometric Institute (EI) at Erasmus University Rotterdam; 44th International Symposium on Forecasting (ISF), Dijon, France; So.Fi.E. Summer School ``Monitoring and Forecasting Macroeconomic and Financial Risk", Brussels, Belgium.
Does Anything Beat a Factor Model?
Comparing Predictive Accuracy in Large Panels of Macroeconomic Time Series
with Christian Brownlees (UPF and BSE) and Eduardo Fonseca Mendes (FGV)
We assess the evidence of superior predictive ability of the factor model against a comprehensive set of alternative methods for macroeconomic forecasting using the FRED database. For this purpose we introduce a testing procedure that allows to test the null of superior predictive ability of a benchmark method uniformly over both a set of alternative methods and a set of series. Results show that the factor model is not outperformed when forecasting the policy relevant variables at short horizons. When considering longer forecasting horizon we have enough evidence against its dominance and simple methods drive the results. Moreover, heterogeneous results emerge when forecasting different categories of the FRED database. Our results highlight the importance of carrying out uniform testing when assessing predictive ability over a collection of series.
*Earlier versions of this paper presented at 29th Finance Forum, the Annual Meeting of the Spanish Finance Association (AEFIN), Santiago de Compostela, Spain; 5th Annual Workshop on Financial Econometrics, Örebro, Sweden; and 27th Meeting of Young Economists (SMYE 2023), Turin, Italy.
Forecasting Intra-daily Volume in Large Panels of Assets
with Christian Brownlees (UPF and BSE), Gaëelle Le Fol (Paris Dauphine University) and Serge Darolles (Paris Dauphine University)
We propose an intra-daily volume prediction model for large panels of assets. The model that combines factor models with large-dimensional regularized regressions, applying this methodology to the intra-daily volume of large European stocks listed in the STOXX 600 Index across different intra-daily frequencies. The analysis shows that the proposed methodology allows to improve intra-daily volume forecasting over a number of benchmarks and, in particular, it outperforms univariate benchmarks commonly used in the literature.
Do Brazilian Municipalities Respond Asymmetrically to a Common Macroeconomic Shock?
with Bruno R. Delalibera (Universitat de Barcelona) and Victor Rodrigues (Insper)