Estimating and Testing Cross-Sectional Asset Pricing Models: A Robust IV Econometric Technique

Misspecified models, noisy betas, and weak instruments are well-known problems in finance and can lead to poor test performance. In this paper, we introduce a new technique for estimating and testing cross-sectional asset pricing models that addresses these problems. We apply our technique to three popular cross-sectional asset pricing models: CAPM, the Fama-French three-factor model, and the CampbellVuolteenaho "good beta-bad beta" model. The estimates of these three models illustrate several desirable properties of our new technique: 1) The asymptotic size of our tests is correct, so our technique does not overreject; 2) Our tests are robust to many sources of weak identification, including misspecified models, measurement error, and weak instruments; 3) Our technique provides an automatic alert for weak instruments; 4) Our tests have considerable statistical power. For example, the tests have sufficient power to reject the three cross-sectional asset pricing models; 5) Using our technique, model rejections can be informative (i.e., they provide constructive information about what is missing from an asset pricing model). In addition to illustrating our technique, we obtain substantive results that may be of wide interest. First, we find evidence that is consistent with the Lewellen-Nagel-Shanken critique (i.e., that the tight factor structure of size and book-to-market portfolios means that many tests have poor power). Second, we find evidence that returns are linked to both Fama-French betas and Campbell-Vuolteenaho betas. Third, we find that characteristics (e.g., size and momentum) contain additional information about returns that is not fully captured by any of the three asset pricing models.