The role of supervised learning in the decision process to fair trade US municipal debt

Abstract

Determining a fair price and an appropriate timescale to trade municipal debt is a complex decision. This research uses data informatics to explore transaction characteristics and trading activity of investment grade US municipal bonds. Using the relatively recent data stream distributed by the Municipal Securities Rulemaking Board, we provide an institutional summary of market participants and their trading behavior. Subsequently, we focus on a sample of AAA bonds to derive a new methodology to estimate a trade-weighted benchmark municipal yield curve. The methodology integrates the study of ridge regression, artificial neural networks, and support vector regression. We find an enhanced radial basis function artificial neural network outperforms alternate methods used to estimate municipal term structure. This result forms the foundation for establishing a decision theory on optimal municipal bond trading. Using multivariate modeling of a liquidity domain measured across three dependent variables, we investigate the proposed decision theory by estimating weekly production-theoretic bond liquidity returns to scale. Across the three liquidity measures and for almost all weeks investigated, bond trading liquidity is elastic with respect to the modeled factors. This finding leads us to conclude that an optimal trading policy for municipal debt can be implemented on a weekly timescale using the elasticity estimates of bond price, trade size, risk, days-to-maturity, and the macroeconomic influences of labor in the workforce and building activity.