State-Dependent Mispricing in the Japanese Bond Market

Abstract

This paper investigates mispricing in the Japanese Government Bond (JGB) market under varying liquidity conditions between 2015 and 2025. We develop a noise measure, defined as the root mean squared error between observed yields and those generated by a fitted Nelson–Siegel–Svensson (NSS) yield curve model. The NSS model provides a smooth and parsimonious approximation of the yield curve, where deviations from fitted values are interpreted as idiosyncratic pricing errors. Following previous literature the noise measure is identified as a proxy for liquidity, implying that periods of elevated noise is caused by a lack of arbitrage capital in the JGB market. By analysing the noise measure we find that periods of elevated fitting errors align with certain episodes of market stress, while lower noise levels correspond to more stable, liquid conditions. To identify shifts in market liquidity, we employ a Markov Regime Switching framework using noise measure as the response variable, and the lag of the noise as a state predictor. The model identifies two regimes: a high liquidity regime (low noise) and a low liquidity regime (high noise). Our results show the existence of persistent high-noise that coincide with periods of market dislocation, such as Covid-19 and globally rising interest rates. By relating the noise measure to other market variables we find significant relationships that help explain some of the variation in the noise. Further extending the research, the inclusion of the Markov Regime Switching model enhances both the analytical depth and empirical scope of the thesis. With the reason being, that it allows for a more refined understanding of bond market dynamics that are not captured by conventional linear time-series models. The results from the Markov model show that several explanatory variables display stronger relationships and increased explanatory power during low-and high liquidity regimes, underlining the fact that liquidity conditions significantly influence the behaviour of key market indicators. By accounting for non-linear dynamics and regime-dependent behaviour, the Markov model improves our ability to detect and interpret subtle structural changes in the JGB market, and therefore offers insights that would likely have remained hidden under a static framework. By utilising the Markov Regime Switching Model, this thesis contributes to the literature by bridging advanced yield curve modelling with dynamic regime classification, thereby improving the empirical understanding of how bond market liquidity evolves over time in a low-rate environment.