Working papers

This paper strives to identify value-based systematic investment opportunities in the U.S. corporate bond market through the joint construction of a bond valuation model and a return factor model. The valuation model explains the cross-sectional variation of bond yields with a flexible local linear functional form in bond risk characteristics. The return factor model embeds the residual yield from the valuation model as a mispricing factor, while accounting for stronger co-movements between bonds from the same industry, similar rating classes, and similar duration segments, as well as differential market pricing for bond return risk, liquidity cost, and optionality exposure. The slope coefficient on the mispricing factor captures the ex post excess return on the value-investing portfolio that targets a unit exposure to the identified mispricing opportunities while being neutral to all systematic risk exposures. 

Investors are averse to risk but love optionality. When a security's embedded optionality increases with its risk level, the entanglement,  combined with the opposite investor preferences, can generate seemingly abnormal market pricing behaviors. This paper frames the bond and stock return behavior within a structural framework and disentangles their directional risk exposure from their optionality exposure via a joint stock-bond return factor model. The factor portfolio targeting a unit exposure to market risk but zero exposure to optionality generates a significantly positive average excess return, consistent with investor risk aversion. By contrast, the factor portfolio targeting a unit exposure to optionality but without directional exposure to firm value variation generates a significantly negative average excess return, reflecting investor penchant for optionality. The separation of risk from optionality sheds light on the distress puzzle in the stock and bond market and helps explain the bet-against-beta and volatility premiums in the stock market.


According to Thorndike (2012)'s analysis of historical firm performance over the past 50 years, the best-performing CEOs are the best capital allocators. This paper argues that a firm's capital structure decision is fundamentally a captial allocation decision based on its risk-return tradeoff prospect, not much different from classic mean-variance investment analysis. The paper constructs mean-variance ratio forecasts based on company return-on-asset histories and shows that the forecasts explain a large portion of the cross-sectional company leverage variation. The leverage predicted by the mean-variance ratio forecast maximizes company value. Once the mean-variance ratio forecast is accounted for, contributions from other commonly identified variables become small. Furthermore, some of the additional explained variations do not constitute value-maximizing leverage target variations, but rather variations away from the target.

The paper identities the historical variance term structure as a key conditioning variable that differentiates the different phases of a company's information cycle and shows that stock variance dynamics vary strongly through the different phases of the cycle. To predict stock variance over a large and ever-changing universe, the paper replaces time-series dynamics specification per each name with a cross-sectional forecasting relation at each date and develop a two-dimensional conditional pooling estimation that balances the needs for reducing estimation errors and capturing dynamics variation across the cycle. We quantify the economic significance of the approach through an option investment analysis and highlight the classic asset pricing relation variations across the information cycle.

The model is constructed in parallel to classic stock return factor models, and decomposes a company's relative valuation into the product of a set of cross-sectionally standardized firm characteristics and the common market pricing coefficients that measure the common value contribution of each characteristic across all companies at a given point in time.

Option prices commonly differ from the Black-Scholes formula along two dimensions: implied volatilities vary by strike price (volatility smiles) and maturity. We account for both using Gram-Charlier expansions to approximate the conditional distribution of the logarithm of the price of the underlying security. In this setting, volatility is approximately a quadratic function of moneyness, a result we use to infer skewness and kurtosis from volatility smiles.