Differencing as a Test of Specification
University of Rochester, Rochester, NY 14627
G. William Schwert
University of Rochester, Rochester, NY 14627
and National Bureau of Economic Research
Halbert White
University of California, San Diego, CA
International Economic Review, 23 (October 1982) 535-552
This paper formalizes arguments presented in Plosser and Schwert [1978]
by constructing a test of the hypothesis that a linear time series
regression equation is well-specified. The test is based on a comparison
of the least squares estimators obtained from the differenced and
undifferenced regressions. The resulting specification test is similar to
those proposed by Hausman [1978]. Monte Carlo experiments show that the
differencing test has favorable size and power properties against errors-in-variables
and omitted variables compared with Wu's [1973] instrumental
variables test, Ramsey's RESET test, and the Durbin-Watson test.
Return to Publications Page