Expected Versus Realized Income Expectations: A Test of the Rational Expectations Hypothesis

We analyze answers to household survey questions on whether the respondents' household income has changed in the past twelve months, and on whether the respondents expect their household income to change in the next twelve months. Both questions are answered on a discrete five points scale. The data are an unbalanced panel of ten consecutive annual waves. Using cross-tabulations of expected and realized changes, we first test the "best-case" hypothesis. This hypothesis implies, under two different nonparametric assumptions on how respondents form their predictions, that respondents have rational expectations, that there are no common unexpected shocks, and that reported expectations are best predictions of future outcomes. We find that the best case hypothesis is rejected: for all years, too many respondents who predict an income fall, ex post report that their household income has not changed. We then construct a bivariate ordered probit random effects panel data model, in which we explain both expectations and realizations from background variables such as age, education level, and labour market status, and from the one year lagged expectation and realization. We show that the hypothesis of rational expectations implies certain restrictions on the parameters in the two equations of this model. The model is estimated by simulated maximum likelihood using the Geweke-Hajivassilou-Keane (GHK) method. The hypothesis of rational expectations is rejected. The hypotheses that expectations are adaptive or naive can be tested in a similar way, and are also rejected.