Some Results on Cointegration with Random Coefficients in the Error Correction Form : Estimation and Testing

This study considers a time-series model with random coefficients in cointegration. The estimation problem can be solved by maximizing the log-likelihood. Asymptotic distributions of the least squares and maximum likelihood estimates are considered. The randomness of the cointegration vector is checked by a score-based test approach. The test statistic converges asymptotically to a functional of Brownian processes. An empirical application to two cointegrated series, federal fund rate and 90-day treasury bill rate is considered