We explore the use of technical indicators to forecast corporate bond returns with various machine learning models. We show that technical indicators yield statistically significant and economically meaningful results, consistently outperforming bond characteristics. Although bond characteristics possess predictive power for bond returns, they do not provide incremental value beyond technical indicators across all bonds. Additionally, machine learning models do not offer substantial improvements over the benchmark linear model. These results underscore the significance of technical indicators in the corporate bond market.

We investigate the impact of scheduled macroeconomic news announcements on the U.S. Treasury market’s efficiency. Using intraday data and controlling for microstructure noise, we employ a robust method to construct market inefficiency measures. We find that the U.S. Treasury market is less efficient in the five-minute interval before news arrival. Our findings are robust for different sample periods, macroeconomic news announcements, and market inefficiency measures. We find that investor heterogeneity provides a possible explanation for the decreased market efficiency before scheduled news announcements.

We evaluate the efficacy of price discovery in the round-the-clock U.S. Treasury market. Using a comprehensive intraday database, we explore informational role of trades over the 24-hour day. We find that information asymmetry is generally highest in the preopen period and lowest in the postclose period. Information asymmetry in the overnight period is comparable to that in the regular trading period. However, on days with macroeconomic announcements, information asymmetry peaks shortly after the news release at 8:30. Moreover, information asymmetry is higher on Monday morning and higher immediately before than after the open of U.S. Treasury futures trading. Although volume is low after hours and trading cost is relatively high, overnight trading generates significant price discovery. Results suggest that overnight trading activity is an important part of the Treasury price discovery process.

This paper studies the pricing of liquidity risk in the cross section of corporate bonds for the period from January 1994 to March 2009. The average return on bonds with high sensitivities to aggregate liquidity exceeds that for bonds with low sensitivities by about 4% annually. The positive relation between expected corporate bond returns and liquidity beta is robust to the effects of default and term betas, liquidity level, and other bond characteristics, as well as to different model specifications, test methodologies, and a variety of liquidity measures. The results suggest that liquidity risk is an important determinant of expected corporate bond returns.

This paper examines the predictability of corporate bond returns using the transaction-based index data for the period from October 1, 2002 to December 31, 2010. We find evidence of significant serial and cross-serial dependence in daily investment-grade and high-yield bond returns. The serial dependence exhibits a complex nonlinear structure. Both investment-grade and high-yield bond returns can be predicted by past stock market returns in-sample and out-of-sample, and the predictive relation is much stronger between stocks and high-yield bonds. By contrast, there is little evidence that stock returns can be predicted by past bond returns. These findings are robust to various model specifications and test methods, and provide important implications for modeling the term structure of defaultable bonds.

Using the daily data of Chinese 7-day repo rates from January 1, 1997 to December 31, 2008, this paper tests a variety of popular spot rate models, including single-factor diffusion, GARCH, Markov regime-switching and jump-diffusion models. We document that Chinese spot rates are subject to both market forces and administrative forces. GARCH, regime-switching and jump-diffusion models capture some important features of the dynamics of Chinese spot rates, but all models under study are overwhelmingly rejected. We further explore possible sources of model misspecification using diagnostic tests.