Key points are not available for this paper at this time.
The use of residuals from the structural equations in a simultaneous‐ equations model can lead to misleading measures of association and to invalid diagnostic statistics for heterosedasticity and functional form misspecifications. In an important paper, Pagan and Hall (1983) suggest a few principals upon which asymptotically valid tests can be constructed, while the issue of appropriate measures of goodness‐of‐fit for IV regressions is addressed in Pesaran and Smith (1994). This paper is concerned with the optimal construction of diagnostics for IV regression, and examines the finite‐sample properties of several tests for functional form misspecifications and heteroscedasticity.
Pesaran et al. (Sat,) studied this question.