Journal of Econometrics 108 (2002) 365–393 www.elsevier.com/locate/econbase The problem of near-multicollinearity revisited: erratic vs systematic volatility Aris Spanos a ; ∗ , Anya McGuirk b a Department of Economics, Virginia Tech, 3016 Pamplin Hall (0316), Blacksburg, VA 24061, USA b Department of Agricultural and Applied Economics, Department of Statistics, Virginia Tech, Blacksburg, VA 24061, USA Received 20 July 2000; received in revised form 29 October 2001; accepted 13 November 2001 This paper is Dedicated to the memory of Paul Driscoll, a wonderful friend and colleague Abstract The main argument of the paper is that the traditional discussion of near-multicollinearity ‘boils down’ to two rather dierent issues which are often conated: (a) a structural issue (high correlation among regressors) which, under certain conditions, gives rise to systematic volatility, and (b) a numerical issue (the regressor data matrix (X T X) is ill-conditioned) which gives rise to erratic volatility. We call into question the traditional account concerning the eects of increasing the correlation among the regressors, and we put forward a revised account of systematic volatility. The main conclusion is that the precision of the coecient estimators and the associated t -ratios do not necessarily decrease as the correlation among regressors increases. We also question the traditional methods of detecting erratic volatility and propose norm bounds for quantifying the potential problem. c  2002 Elsevier Science B.V. All rights reserved. JEL classication: C1; C2; C4 Keywords: Near-multicollinearity; Ill-conditioning; Condition number; Norm bounds; Erratic volatility; Systematic volatility; Statistical parameterization 1. Introduction Multicollinearity constitutes one of the primary empirical modeling problems pertain- ing to the linear regression model (LRM); see Johnston (1984), Judge et al. (1988), Greene (1997), Kennedy (1998), Wooldridge (1999). Even so, the manifestations of near-multicollinearity seem unclear, and there is no generally accepted way to detect * Corresponding author. Fax: +1-540-231-5097. E-mail addresses: aris@vt.edu (A. Spanos), meguirk@vt.edu (A. McGuirk). 0304-4076/02/$-see front matter c  2002 Elsevier Science B.V. All rights reserved. PII:S0304-4076(01)00144-0