A Performance Comparison of Large-n Factor Estimators
Tsinghua University - PBC School of Finance
London School of Economics & Political Science (LSE) - Department of Accounting and Finance
Robert A. Korajczyk
Northwestern University - Kellogg School of Management
February 25, 2016
We evaluate the performance of various methods for estimating factor returns in an approximate factor model. Differences across estimators are most pronounced when there is cross-sectional heteroskedasticity, or when cross-sectional sample sizes, n, are below 4,000 assets. Estimators incorporating either cross-sectional or time-series heteroskedasticity out-perform the other estimators when those types of heteroskedasticity are present. With both cross-sectional and time-series heteroskedasticity and an unbalanced panel, the methods of Connor and Korajczyk (1988) and Jones (2001) provide the most accurate factor return estimates. The estimator of Stock and Watson (1998) provides less accurate estimates than the other estimators for n less than 2,000 (for the first factor) or 4,000 (for higher-order factors).
Number of Pages in PDF File: 28
Keywords: Factor Model, Asymptotic Principal Components, Large-Scale Factor Model
JEL Classification: G1, G12, C15, C23
Date posted: January 18, 2014 ; Last revised: February 26, 2016
© 2016 Social Science Electronic Publishing, Inc. All Rights Reserved.
This page was processed by apollobot1 in 0.234 seconds