Cross-Section Error Dependence in Panel Quantile Regressions

Seminar abstract

We argue that a factor structure of the disturbances in panel quantile regression [QR] models may have a biasing effect on the QR estimator even if factors and loadings are strictly exogenous. Therefore, cross-sectional dependence is an indicator of misspecification in panel quantile regressions. We furthermore propose a test for panel QR misspecification based on this argument. The proposed test is a version of the familiar Breusch-Pagan test based on residuals from individual-unit quantile regression estimation at the quantile of interest. To this end, we distinguish between pooled slope coefficient estimation and individual-unit estimation. The test possesses a standard normal limiting distribution under joint N,T asymptotics with restrictions on the relative rate at which N goes off to infinity. A finite-sample correction improves the applicability of the test for larger cross-sectional dimension N. We illustrate the usefulness of the proposed test with an application to housing prices.

How to join this seminar

This seminar is free to attend and will take place online.

Please contact the organiser for details on accessing the event.

Speaker bio

Matei Demetrescu is a professor in statistics and econometrics, and his main research area are in panel and time series econometrics, financial econometrics and econometric methods. He has published peer-reviewed articles in internationally renowned scientific journals, including the Journal of Econometrics, the Journal of Applied Econometrics, the Journal of Business and Economics Statistics and Econometric Theory.

His research work is centred on macroeconomic and financial forecasting, both from an applied and a methodological point of view.
Among others, he has been developing methods to robustify various estimation and test procedures in econometric forecasting to stylized facts of the data such as uncertain persistence or time-varying volatility, also in higher-dimensional contexts such as large-N, large-T panel data sets.

Recent research includes monitoring quantitative risk forecasts, but also the use of bootstrap methods to allow for predictability testing when the predictors are of uncertain persistence or to allow for forecast comparisons when the forecasts are noisy.