Since the 1980s, prospect theory has been considered as the most successful descriptive theory for decision making. In this paper, we examine the predictive power of prospect theory in the U.S. corporate bond market. The empirical evidence shows that prospect theory has significant predictive power for corporate bond returns, especially for junk bond returns. Unlike the findings for the stock market, the loss aversion component plays the most important role in predicting corporate bond returns. The probability weighting component also plays a predictive role for junk bonds, but not for investment-grade bonds.

Introduction

Kahneman and Tversky (1979) propose prospect theory and Tversky and Kahneman (1992) modify it into the known cumulative prospect theory. Since then, prospect theory has been widely considered to be the most successful descriptive theory for decision making. A number of studies have reported that prospect theory can explain some financial phenomena, such as the long-run underperformance of initial public offerings (IPOs) (Ma and Shen, 2003), skewness of stock returns (Barberis and Huang, 2008), inflation perceptions (Brachinger, 2008), stock option pricing (Gurevich et al., 2009), stock momentum (Menkhoff and Schmeling, 2006), and etc. Benartzi and Thaler (1995) propose that investors evaluate the stock/bond market by calculating the prospect theory value of its historical return distributions. Barberis et al. (2016) propose a stock-level analog of Benartzi and Thaler’s (1995) market-level model. Their empirical evidence shows that prospect theory based on a stock’s historical return distributions has significant predictive power for subsequent stock returns. Although it has been shown that prospect theory has predictive power for stock returns, there is a paucity of studies implementing prospect theory in the corporate bond market. This raises a natural question: Does prospect theory have predictive power in the corporate bond market? Furthermore, unlike the stock market, institutional investors hold most of the investment-grade corporate bonds, while individual investors play an important role in junk bonds.1 The different investors’ structure between the investmentgrade and junk bonds provides an ideal setting to test whether institutional investors make decisions differently from individual investors. In this paper, we develop our hypotheses around these questions and test them by using both investment-grade and junk bonds. The essential concept of prospect theory is called ‘‘mental accounting’’ (Kahneman and Tversky, 1984; Thaler, 1985). It involves the mental process of an investor in both coding and evaluating financial assets. For the process of coding a financial asset, the historical return distribution is the most convenient, easily accessible, and intuitive information that investors can get when they make investment decisions. As proposed by Benartzi and Thaler (1995), investors evaluate a corporate bond’s historical returns through a mental mechanism that is captured by prospect theory. To explain the rationality of this, we start by introducing two closely related concepts in cognitive psychology: System 1 and System 2 thinking processes. In these two concepts of cognitive psychology, people follow a two-step mental procedure when making decisions (Kahneman and Frederick, 2002). Stanovich and West (2000) named them as System 1 and System 2 thinking processes, respectively. System 1 thinking is quick with little consciousness, and is basically pretty primitive. In contrast, System 2 thinking is slower, more reflective, and acquire proficiency and skills. In the first step, people use System 1 thinking to make a quick, intuitive judgment to a problem. There is little effort exerted in this step and it does not require the use of learned rules and skills. In the second step, people use System 2 thinking to evaluate the judgment made in the first step and accept, revise, or override it. Frederick (2005) uses an example of when people solve the cognitive reflection test problem: If it takes 5 machines 5 min to make 5 widgets, how long would it take 100 machines to make 100 widgets? An intuitive answer to this problem would be 100 min. The thinking process that leads to this intuitive answer is System 1 thinking. A reevaluation of the answer will then recognize that the original 5 machines are enough to make 100 widgets in 100 min. A more careful and elaborative calculation will then lead to the correct answer of 5 min. The thinking process that reevaluates System 1 thinking and may override the incorrect answer from System 1 thinking with the correct answer is System 2 thinking.