Inferences in Factor Pricing Models with Many Assets

Massachusetts Institute Of Technology, Cambridge MA

Investigators

Abstract

Factor pricing models, which intend to explain non-trivial expected returns to assets as compensation for risk assumed by asset holders, are important benchmarks that academics and practitioners in financial industry use to understand asset prices and to make investment and portfolio decisions. However, current statistical and econometric techniques for estimating and assessing factor pricing models often produce spurious statistical results. The goal of this research is to develop more reliable new statistical techniques that will work well in an environment that makes use of big data sets on assets returns and that will address the empirically-important issues of observed strong co-movements between asset prices and the weak explanatory power of some risk factors. Given the broad range of applications for factor models, the techniques and insights from this research will also be useful for researchers and policy makers working with macroeconomic data, including Federal Reserves and Central Banks. In addition, a key educational component of this research involves the training of graduate students in the area of econometrics. This research demonstrates the statistical problems of the currently widely employed two-pass Fama-MacBeth procedure and further creates more reliable new statistical methods. The investigator combines several modern econometric and statistical approaches such as large-dimensional asymptotics, drifting-parameter embedding (as in the weak-identification literature) and factor models, in order to better approximate empirically relevant features of commonly-used data sets. Each of these approaches brings its own distinct statistical challenges that the investigator will solve by proposing a new and original estimation procedure. One of the insights from this research is that strong cross-sectional dependence between asset returns such as an approximate factor structure can be exploited to create proxies for dependence and that this can be used to build a correction. The weakness of observed factors is an additional serious complication that the investigator aims to solve. This research will develop new techniques for three distinct tasks: (i) estimation and inference on risk premia; (ii) testing whether the proposed factor pricing model is correctly specified; and (iii) assessment of the quality-of-fit of a pricing model. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.