Introduction

The method of Instrumental Variables (IV) was first formalized by Reiersol (1941) and Geary (1948, 1949) in the context of the errors-in-variables (confluence analysis) using time series data. The IV method was introduced as an alternative to least-squares in an attempt to deal with the inconsistency of the ordinary least-squares (OLS) estimator. Durbin (1954) systematized and extended the IV method to a general linear relationship where the explanatory variables are correlated with the error term. He also conjectured a direct relationship between the IV estimator and the limited information maximum likelihood (LIML) estimator in the context of the simultaneous equations model (SEM). This conjecture was formally confirmed by Sargan (1958) who extended the scope of the IV method by relating it to maximum likelihood methods, establishing the IV as a general estimation method in econometrics. Since then the IV method has been increasingly used both as a unifying concept in econometric estimation (see Hendry, 1976; Hausman, 1975; Bowden and Turkington, 1984; inter alia), as well as a computationally convenient method (see Brundy and Jorgenson, 1974; Hansen and Singleton, 1982). More recently, the IV method has been extended in a number of ways to unify estimation and testing procedures in a non-maximum likelihood framework (see White, 1982, 1984), misspecification testing (see Pagan, 1984), as well as unit-root testing (see Hall, 1989). Arguably, the most important extension has been the Generalized Method of Moments (GMM) proposed by Hansen (1982); see also Newey (1985).