Search

In the first chapter of my thesis, I study the asset pricing implications of being able to optimally early exercise a plain-vanilla put option, contrasting the expected returns of equivalent American and European put options. Standard pricing models with stochastic volatility and asset-value jumps suggest that the expected return spread between them is positive, can be economically sizable, and widens with a higher optimal early exercise probability, as induced through a higher moneyness, shorter time-to-maturity, or lower underlying-asset volatility. Studying single-stock American put options and equivalent synthetic European options formed from applying put-call parity to American call options on zero-dividend stocks, my empirical work supports the theoretical predictions. My results, therefore, indicate that the early exercise feature can have a strong effect on option returns. In the second chapter, I introduce a dynamic trading strategy based on a theoretical proposition of Shreve (2004). Many studies report that American option investors often exercise their positions suboptimally late. Yet, when that can happen in case of puts, there is an arbitrage opportunity in perfect markets, mentioned in Shreve (2004), exploitable by longing the asset-and-riskfree-asset portfolio replicating the put and shorting the put. Using early exercise data, I show that the arbitrage strategy also earns a highly significant mean return with low risk in real single-stock put markets, in which exactly replicating options is impossible. In line with theory, the strategy performs particularly well on high strike-price puts in high interest-rate regimes. It further performs well on short time-to-maturity puts on low volatility stocks, consistent with evidence that investors do not correctly incorporate those characteristics into their exercise decisions. The strategy survives accounting for trading and short-selling costs, at least when executed on liquid assets. In the third chapter, I revisit the value-weighted stock return predictability of Black-Scholes (1973) option implied volatility spreads. Studies so far have explained this predictability using investors' informed trading activities in options ahead of the stock market and/or frictions in the underlying stock. Nevertheless, for single-stock American options, I show that the ability of implied volatility spreads to predict cross sectional stock returns is primarily driven by the friction-induced optimal early exercise of put options that is not accounted for in calculating implied volatility. The contribution of other factors to the predictive ability of implied volatility spreads are largely insignificant. Further evidence suggests that the predictability cannot be solely explained by the trading activities of informed option investors.

This thesis broadly covers three different topics in empirical cross-sectional asset pricing and consists of three papers. The first paper prices the cross-sectional delta-hedged option and straddle returns in a consumption-based asset pricing model. Delta-hedged options are particularly sensitive to the underlying asset's volatility, which is in turn determined by the fundamental consumption volatility. The strong connection between delta-hedged options and consumption volatility provides us with powerful test assets to identify the consumption volatility premium and hence the preference of the representative agent. As indicated by our results, exposures to consumption growth, expected consumption growth, and consumption volatility are all significantly priced in the cross-section of delta-hedged option and straddle returns. Consumption growth and expected consumption growth command positive risk premiums, whereas consumption volatility commands a negative risk premium, suggesting that investors prefer early resolution of uncertainty. Our results further suggest that consumption risk exposures provide rational foundations for well-known relations between option moneyness or idiosyncratic underlying-stock volatility and the cross-section of delta-hedged option or straddle returns. The second paper relies on a hazard-model prediction of failure as proxy for firm-level distress risk. The paper discovers a significantly negative relation between firm-level distress risk and the cross-section of corporate bond returns, which is analogous to the often negative relation between distress risk and stock returns found in prior studies ("distress anomaly"). Our finding casts doubts on theories arguing that the distress anomaly arises due to shareholders shifting financial risk onto debtholders in distress. In accordance, proxy variables suggested by such theories do not condition the distress risk-bond return relation. Theories suggesting that distressed firms own valuable disinvestment options and thus have a low levered asset risk are more promising to explain the anomaly, with some of the proxy variables suggested by these theories conditioning the former relation. The third paper evaluates the prediction performance of machine learning methods in predicting the cross-sectional bond returns out-of-sample. Recent studies show that machine learning methods, especially neural networks, perform well in predicting the cross-sectional stock returns when the number of predictors is large. Prior research indicate that bond returns can be predicted by not only macroeconomic factors, bond market factors, and bond-level characteristics, but also stock market factors and stock-level characteristics. Therefore, the number of predictors in the bond market is even larger than that in the stock market, and the advantage of machine learning will be more pronounced in forecasting bond returns. In this work, I show that machine learning methods perform much better than the simple linear model in predicting bond returns out-of-sample.