Abstract:Time varying model parameters offer tremendous flexibility while requiring more sophisticated learning methods. We discuss on-line estimation of time varying DLM parameters by means of a dynamic mixture model composed of constant parameter DLMs. For time series with low signal-to-noise ratios, we propose a novel method of constructing model priors. We calculate model likelihoods by comparing forecast distributions with observed values. We utilize computationally efficient moment matching Gaussians to approximate exact mixtures of path dependent posterior densities. The effectiveness of our approach is illustrated by extracting insightful time varying parameters for an ETF returns model in a period spanning the 2008 financial crisis. We conclude by demonstrating the superior performance of time varying mixture models against constant parameter DLMs in a statistical arbitrage application.
BACKGROUND
Linear ModelsLinear models are utilitarian work horses in many domains of application. A model's linear relationship between a regression vector F t and an observed response Y t is expressed through coefficients of a regression parameter vector θ. Allowing an error of fit term ε t , a linear regression model takes the form:where Y is a column vector of individual observations Y t , F is a matrix with column vectors F t corresponding to individual regression vectors, and ε a column vector of individual errors ε t . The vector Y and the matrix F are observed. The ordinary least squares ("OLS") estimateθ of the regression parameter vector θ is (Johnson and Wichern, 2002)
Stock returns exampleIn modeling the returns of an individual stock, we might believe that a stock's return is roughly a linear function of market return, industry return, and stock specific return. This could be expressed as a linear model in the form of (1) as follows:where r represents the stock's return, r M is the market return, r I is the industry return, α is a stock specific return component, β M is the sensitivity of the stock to market return, and β I is the sensitivity of the stock to it's industry return.