COMPARING ASSET PRICING MODELS USING QUANTILE REGRESSIONS FOR DISTANCE-BASED METRICS
This thesis compares the performance of ten well-known asset-pricing models for cross-sectional returns of various portfolios from January 1967 to December 2016. We rely on the distance-based metrics as the primary performance measure and use quantile regressions to compare models at a wide range of quantiles of the asset return distribution. The model performance is examined from both statistical and economic perspectives. We find that the Fama and French (2018) six-factor model reliably outperforms other competing models in pricing the selected portfolios. In particular, both the momentum factor and the value factor are necessary in asset-pricing models to explain the return variations in different quantiles. We also find that the performance of Barilla and Shanken (2018) six-factor model exhibits strong explanatory power in medium to high quantiles, despite some existing findings that their model performs poorly in OLS regressions. Overall, we show that the distance-based metrics coupled with quantile regressions provide a consistent and robust model-comparison methodology that largely enhances the existing OLS-based statistical measures.