We examine the small sample properties of tests of rational expectations models. We show using Monte Carlo experiments that these tests can be extremely biased toward rejection for sample sizes typical in applied research. These biases are important when the time series examined are highly autoregressive. We also show that these tests are even more biased with detrended data. We present correct small sample critical values for our canonical problem.
Mankiw, N. Gregory and Shapiro, Matthew D., "Do We Reject Too Often? Small Sample Bias in Tests of Rational Expectations" (1985). Cowles Foundation Discussion Papers. 984.