Kalmann filter for pricing illiquid bonds

Specialeforsvar: Louis Krog Kaufmann

Titel: Kalman filter for pricing illiquid bonds

Abstract: This paper presents a Kalman filter model for estimating bond yields developed based on a propietary dataset of historical time-series data for Chinese corporate bonds. By developing a state-space model of the bond market, this allows for parameterization of a Kalman filter model which uses an exogenous proxy model and observed trades to estimate an issuer-wide spread between the proxy model and observed yields. Details of the implementation and adjustments to the standard implementations are discussed to make the model more applicable to the sparse structure of the data. The tuning of the model parameters is explored using maximum likelihood estimation and the autocovariance least squares method. The autocovariance least squares method is a correlation based method, which has no published literature relating to implementation of the method for financial data. The results in terms of reducing estimation error show that the method also has applicability within financial time series data. The Kalman filter produces an estimate of the traded yields which has significantly less error than the base proxy model, and the tuning procedures also provide significant improvements with regards to reducing prediction error.

Vejleder: David Glavind Skovmand
Censor:  Peter Sestoft, IT Universitetet