STATE DEPENDENT MODELS OF STOCK RETURNS

We analyze the nonlinear time series properties of weekly stock returns using a state dependent model. The data exhibit little evidence of state dependency or even time dependency. They do, however, have some temporal movement in their autoregressive coefficients in the period 1962 through 1975 and fit linear autoregressive models from 1975 through 1987. We conclude that previous findings of nonlinearities in stock returns are not due to nonlinearities in the conditional mean process.