Markov Breaks in Regression Models

Aaron Smith
2012 Journal of Time Series Econometrics  
This article develops a new Markov breaks (MB) model for forecasting and making inference in regression models with stochastic breaks. The MB model permits an arbitrarily large number of abrupt breaks in the regression coefficients and error variance, but it maintains a low-dimensional state space, and therefore it is computationally straightforward. I compare the model to competing breaks models and show that it outperforms them in a Monte Carlo experiment. I employ the MB model to assess the
more » ... odel to assess the efficacy of the conditional CAPM in pricing US stock returns. Both in and out of sample and for all portfolios under study, the MB model fits monthly stock returns significantly better than several alternative models. Using the estimated MB model to capture time-varying alphas and betas, I show that the momentum effect has persisted with a constant pricing error since 1927, the size effect has been prominent only in short bursts, and the book-tomarket effect has been strong since 1980 after being less prominent in the preceding 30 years. . I am grateful to Robert Engle for comments and discussions that were crucial in the development of this manuscript.
doi:10.1515/1941-1928.1111 fatcat:dxgtjr3qvve2vnrm4pt5ovytla