This paper investigates three techniques for the estimation of conditional time-dependent betas: (a) a multivariate generalised ARCH approach; (b) a time-varying beta market model approach suggested by Schwert and Seguin (1990); and (c) the Kalman filter technique. These approaches are applied to a sample of returns on Australian industry portfolios over the period 1974-1996. The evidence found in this paper, based on in-sample forecast errors, overwhelmingly supports the Kalman filter approach When out-of-sample forecasts are considered the evidence again finds in favour of the Kalman filter approach.
- Kalman filter
- Time-varying beta