Introduction

The problem of rational expectations (RE) in limited dependent variable (LDV) models was first considered in Chanda and Maddala (1983). Maddala (1993) provided more detailed discussions on expectation formation under different assumptions used by agents as well as different estimation methods. The specification and estimation of such models have further been investigated in Maddala (1990, 1993), Pesaran (1990), Donald and Maddala (1992), Lee (1994), Pesaran and Ruge-Murcia (1996, 1998), and Pesaran and Samiei (1995). Under some general conditions, the RE solution exists and is unique (Pesaran and Samiei (1992a), Donald and Maddala (1992), Lee (1994), and Pesaran and Ruge-Murcia (1996)). The usefulness of such models in empirical studies can be found in Shonkwiler and Maddala (1985) and Holt and Johnson (1989) for agricultural commodities markets with price supports and Pesaran and Samiei (1992a, 1992b) and Edin and Vredin (1993) for models with exchange rate determination under a target zone.

For the estimation of limited-dependent rational expectations (LDRE) models, Pesaran and Samiei (1992a) have suggested the maximum likelihood (ML) method. Estimated models have, so far, assumed serially independent disturbances. Even though dynamics can easily be introduced through the inclusion of observed lagged dependent variables without complication (Pesaran and Samiei (1992b)), neither dynamic structures involving lagged latent dependent variables nor serially correlated disturbances are allowed. As these models are used for time series data, it is of interest to consider the model estimates with serial correlation or latent dynamic structures in addition to the inclusion of observed lagged dependent variables.