Title: An econometric analysis of regression models with sorted variables
Authors: Sha Meng - University of Liverpool (United Kingdom) [presenting]
Abstract: The quintile portfolio-level analysis (QPA) is widely used in the empirical finance literature to investigate the predictive power of a tested risk factor for assets returns. The QPA assumes that the individual stock return is a linear combination of controlling risk factors and tested risk factors. The hypothesis testing uses the cross-section of stock returns to create stock portfolios that have different sensitivities to the tested risk factor and estimates regressions for sorted portfolios and then calculates $T$ statistics. However, very few studies have investigated the econometric properties and finite sample performance of QPA test. A model framework is constructed for the data generating process and analyses the size and power of QPA. Also, the standard QPA procedure is modified and a direct test procedure is proposed to directly test the assumption in panel data regressions. The size and size-corrected power of QPA and direct test are compared using the same datasets, and results show that sizes of the two tests are similar and there are no size distortions, whereas, the size-adjusted power of QPA test increases more slowly comparing with that of the direct test, which means that the direct test is much more powerful than QPA.