Modeling Bond-Trading Behavior Using a Zero-Inflated Multivariate Poisson

• Activity Number: 174
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Business and Economic Statistics Section
• Author(s): Bonnie Kathryn Ray*+ and Sarah Thomas and Katherine Bennett Ensor
• Companies: IBM T.J. Watson Research Center and Rice University and Rice University
• Keywords: Multivariate Poisson ; ZIP ; MCEM ; systemic risk ; bond liquidity

Abstract

We present an observation-driven, doubly inflated multivariate ZIP model to characterize portfolios of bonds based on transaction volumes or counts of price change deltas over time. Marginal inflation on each individual series allows us to capture idiosyncratic features, while joint zero inflation allows us to capture potential systemic risk issues present in the portfolio. We present the form of this new, doubly inflated model, describe an MCEM algorithm to fit the model, and apply the results to series of hourly volumes and price changes for bonds traded on the US market during 2009. If time permits, we also discuss a clustering algorithm that can be used to identify similar sets of bonds on the basis of their liquidity, helpful, for example, in validating bond ratings or building a diverse portfolio of bonds.