Microstructure-Efficient Estimation of Sovereign Yield Curves in Less-Liquid Markets

Abstract We supply microstructure foundations for liquidity-weighted Nelson–Siegel–Svensson curve fitting in less-liquid sovereign bond markets. A yield-space version of the Glosten and Milgrom dealer problem delivers an optimal bond weight proportional to turnover, a Ho and Stoll inventory channel delivers a weight proportional to outstanding amount, and the hybrid of the two is governed by a single dominance ratio that is identifiable from data. We derive consistency, asymptotic normality and Aitken efficiency of the resulting estimator as a nonlinear M-estimator under conditional heteroskedasticity, and we characterise identification of the Svensson curvature parameter as a coverage condition on the maturity grid, which turns two practitioner patches for thin panels into corollaries. A Monte Carlo experiment and a Polish primary-dealer bond panel of more than 5,000 daily cross-sections show that equal weighting is uniformly the least accurate scheme, with the hybrid weight reducing out-of-sample yield-prediction error by about one basis point in median, and a downstream check traces the same order of difference into Adrian, Crump and Moench term-premium estimates.