We derive the information matrix test, suggested by White (1982), for the normal fixed regressor linear model, and show that the statistic decomposes asymptotically into the sum of three independent quadratic forms. One of these is White's (1980) general test for heteroscedasticity and the remaining two components are quadratic forms in the third and forth powers of the residuals respectively. Our results show that the test will fail to detect serial correlation and never be asympotically optimal against heteroscedasticity, skewness and non-normal kurtosis. The information matrix test is contrasted with the test procedures of Bera and Jarque (1983) and Godfrey and Wickens (1982), who construct a composite statistic form asymptotically optimal and independent tests against particular alternatives. Our results suggest that this alternative strategy is likely to be a more fruitful source of a general regression diagnostic.
