manner. Such an approach is made available through application of the Kalman filter with a time varying parameter specification. An example of a time varying parameter model using the Kalman filter can be found in Kim and Nelson (1989).

Our interest here is to develop a robust dynamic trading model for economic sectors using factors identified as significant in the preceding literature. Further, the model we develop assumes time-variation in factor sensitivities to capture changing risk premia over time. Time variation in factor betas is approached using dynamic updating in the Kalman filter. The result is a highly responsive model that significantly outperforms comparable static and rolling parameter specifications. Employing this methodology, we would like a model that is particularly prescient

at business cycle turning points.

passive

models

optimized

over

# Such a model may provide an important hedge against

the

most

recent

economic

regime

or

long

term

samples.

more The

model developed here also appears with the benchmark CAPM.

to

have

important

risk

pricing

properties

when

contrasted

The balance of this paper is organized as follows: In Section 2 a time varying parameter factor model (TVPFM) using lagged economic factors and industry sectors as portfolios is motivated and developed. Section 3 describes the full Bayesian estimation and model selection criteria employed for evaluating the model. Some preliminary indications from the model output are also discussed. Section 4 investigates the behavior of out of sample risk premia on the predicted model sector returns. In Section 5, the potential profitability of a simple trading strategy using the predicted returns of the TVPFM is investigated and discussed. Section 6 concludes.

2.

A Time Varying Parameter Factor Model

2.1.

General Model Specification

We begin with a time series factor model of equity returns. The factors are assumed to be lagged fundamental macroeconomic variables. The return generating process for each portfolio i is expressed as

r_{i , }_{t 1 }

1 t t f

u

_{i,t1 }

(1)

3