Factor Investing in US Sovereign Bond Market: A New Generation of Conditional Carry Strategies with Applications in Asset-Only and Asset-Liability Management

Abstract

This article provides a detailed analysis of the theoretical, statistical, and implementation challenges related to factor investing in the US sovereign bond markets, with a focus on the level factor, which explains for any maturity the largest fraction of differences in bond returns over time. Using a comprehensive database of individual bond returns in the United States covering the 1975–2018 sample period, the authors find that a conditional version of a carry strategy based on a time-varying exposure to the level factor can generate up to 210 bps of excess performance (gross of transaction costs) over the benchmark and a significantly higher Sharpe ratio. Overall, their results suggest that even in a single-issuer universe with highly correlated bond returns, and after accounting for transaction costs, conditional investing strategies based on second-generation return-predicting factors can lead to robust benefits from both asset-only and asset-liability management perspectives.

TOPICS: Analysis of individual factors/risk premia, factor-based models, style investing

Key Findings

• • Conditional carry strategies based on second-generation return-predicting factors generate excess performance with respect to both unconditional carry strategies and conditional carry strategies based on the slope of the yield curve.

• • These strategies remain profitable when implemented at a CUSIP level, even when transaction costs and long-only constraints are accounted for, and show particularly strong levels of outperformance in increasing interest rate environments and in bear equity markets.

• • For liability-driven investors, these conditional carry strategies can be used to generate dynamic active duration bets with respect to the liability benchmark and substantially outperform a strict duration-matching strategy with relatively modest levels of tracking error.