This website uses cookies.

By clicking the "Accept" button or continuing to browse our site, you agree to first-party and session-only cookies being stored on your device to enhance site navigation and analyze site performance and traffic. For more information on our use of cookies, please see our Privacy Policy.

« Back to Results

New Methods for the Cross Section of Expected Returns

Paper Session

Sunday, Jan. 6, 2019 10:15 AM - 12:15 PM

Hilton Atlanta, Grand Ballroom D
Hosted By: American Finance Association
  • Chair: Michael Weber, University of Chicago

Large Sample Estimators of the Stochastic Discount Factor

Soohun Kim
,
Georgia Institute of Technology
Robert Korajczyk
,
Northwestern University

Abstract

We propose estimators of the stochastic discount factor (SDF) using large cross-section of individual stocks. We suggest a correction for the small sample bias in a standard GMM estimator induced by having a finite time series and show how to use the correction in exploiting unbalanced panels of individual stock
returns. Our estimators can utilize both a prespecified set of traded or non-traded factors implied by a specific asset pricing model and latent factors estimated by multivariate statistical methods. The estimators perform well in simulations designed to mimic the U.S. equity markets. We apply our SDF estimators to the 10,112 individual stock price dynamics in the U.S. over 50 years from 1976 to 2016, and identify which factors in popular asset pricing models command a price of risk.

Empirical Asset Pricing via Machine Learning

Shihao Gu
,
University of Chicago
Bryan Kelly
,
Yale University
Dacheng Xiu
,
University of Chicago

Abstract

We synthesize the eld of machine learning with the canonical problem of empirical asset pricing: Measuring asset risk premia. In the familiar empirical setting of cross section and time series stock return prediction, we perform a comparative analysis of methods in the machine learning repertoire, including generalized additive models, dimension reduction tools, boosted regression trees, random forests, and neural networks. At the broadest level, we nd that machine learning has great promise for describing asset price behavior. Our implementation establishes a new standard for accuracy in measuring risk premia summarized by unprecedented high out-ofsample return prediction R2. We identify the best performing methods (trees and neural nets) and trace their predictive gains to allowance of nonlinear predictor interactions that are missed by other methods. Lastly, we nd that all methods agree on the same small set of dominant predictive signals that includes variations on momentum, liquidity, and volatility. Improved risk premia measurement through machine learning can simplify the investigation into economic mechanisms of asset pricing and justies its growing role in innovative nancial technologies.

What Firm Characteristics Drive United States Stock Returns?

Yufeng Han
,
University of North Carolina-Charlotte
Ai He
,
Emory University
David Rapach
,
Saint Louis University
Guofu Zhou
,
Washington University-St. Louis

Abstract

We employ a forecast combination approach to analyze the ability of 94 firm characteristics from Green, Hand, and Zhang (2017) to predict US stock returns. Using machine learning tools to pool forecasts, we find that most of the firm characteristics matter over time--and approximately 30 matter on average at each point in time--for forecasting value-weighted cross-sectional returns before and after 2003, a year when Green, Hand, and Zhang (2017) detect a major structural break. By processing the information in a plethora of predictors in a manner that alleviates overfitting, our combination approach provides economically significant out-of-sample forecasts of cross-sectional returns consistently over time.

Break Risk

Simon Smith
,
University of Southern California
Allan Timmermann
,
University of California-San Diego

Abstract

We propose a new approach to forecasting stock returns in the presence of structural breaks that simultaneously affect the parameters of multiple portfolios. Exploiting information in the cross-section increases our ability to identify breaks in return prediction models and enables us to detect breaks more rapidly in real time, thereby allowing the parameters of the predictive return regression to be updated with little delay. Empirically, we find that accounting for breaks in panel return models allows us to generate out-of-sample return forecasts that are significantly more accurate than existing forecasts along both statistical and economical measures of performance. Moreover, we find that firms whose equity risk premium processes are most affected by breaks earn significantly higher average returns than firms with lower break exposure, suggesting that ``breaks'' is a risk factor that is priced in the cross-section. Finally, we find that the majority of breaks in equity premiums can be closely tied to breaks in the dividend growth process.
Discussant(s)
Svetlana Bryzgalova
,
Stanford University
Alberto Rossi
,
University of Maryland
Jonathan Lewellen
,
Dartmouth College
Andreas Neuhierl
,
University of Notre Dame
JEL Classifications
  • G1 - General Financial Markets