Research Statement

My research is at the intersection of econometrics and finance, with an emphasis on applying newly developed high-frequency econometric procedures to solve the central problems in asset pricing. For example, my job market paper answers a burning question in finance: Are the massive number of asset pricing “anomalies” really unique?. I disentangle this puzzle by pointing out that they are manifested by 1 single factor. Their return differentials are due to their risk exposures to the 1-factor. To understand the massive number of “anomalies” is to understand this 1-factor.
(Download full research statement)

Job Market Paper

An Efficient Factor from Basis “Anomalies”, Download sanity check and replication file

A look-ahead-bias-free, ex-ante mean-variance efficient portfolio from Size, B/M and Momentum “anomalies” has an ex-post Sharpe ratio of 2.3. It captures the non-monotonic benefits from characteristics that are ignored by multi-factors and eliminates 39 out of 42 “unique anomalies”. Based on a wide set of tests, the 1-factor model significantly outperforms and drives out the 11 factors: MKT-Rf, SMB, HML, MOM, RMW, CMA, qME, qIA, qROE, QMJ, LIQ for different combinations of 147 test assets. The efficient factor is priced at the firm-level with more than 12% per year that cannot be explained by existing models.
Analytically, “anomalous” predictabilities are equivalent to 1-factor pricing, regardless of rational/behavioral cause. A projected Stochastic Discount Factor return deduced from the efficient factor is consistent with economic theory. The risk premium is empirically tied not only to consumptions but also to interest rate environment and industrial production.

Working Papers

Good Volatility, Bad Volatility and the Cross-Section of Stock Returns, with Tim Bollerslev and Sophia Zhengzi Li, Submitted

Based on intra-day data for a large cross-section of individual stocks and newly developed econometric procedures, we decompose the realized variation for each of the stocks into separate so-called realized up and down semi-variance measures, or good and bad volatilities, associated with positive and negative high-frequency price increments, respectively. Sorting the individual stocks into portfolios based on their normalized good minus bad volatilities results in economically large and highly statistically significant differences in the subsequent portfolio returns. These differences remain significant after controlling for other firm characteristics and explanatory variables previously associated with the cross-section of expected stock returns. The results also remain intact in double portfolio sorts designed to control for other high-frequency-based realized variation measures. By contrast, the strong negative association between the realized skewness measure and subsequent returns recently documented by Amaya, Christoffersen, Jacobs, and Vasquez (2016) is completely reversed after controlling for the individual stocks relative good minus bad volatility.

Factors and Their Economic Value in Volatility Forecasts, with Lada Kyj

We propose a new family of simple and reliable realized volatility based forecasting models by exploiting the factor structure in the cross-section of volatility, named Factor Auto Regressive (FAR) models. Rather than being isolated, securities are linked with one another through the highly persistent and forecastable common factors. A comprehensive evaluation shows that models exploiting the simple factor structure are able to significantly out-perform (under-perform) the best existing models, for 87% (0%) of securities in all S&P500 constituents and 77% (0%) in the entire TAQ universe. Securities with high betas are expected to, and indeed receive even stronger gains in predictability. For a hypothetical mean-variance investor, the improvement is worth up to 80 bps per year over best existing models. The results stand strong to a long and exhaustive list of robustness checks.

Work In Progress

Measuring Tail Risks with Threshold Quantile Beta, with Yichong Zhang

Using a novel threshold quantile regression method, with automatically determined thresholds, we find strong evidence that asset’s market beta is not constant under different market conditions. Portfolios formed on the threshold quantile beta have significant pricing errors with respect to existing factor models.