Chinese Stock Market Return Predictability: Adaptive Complete Subset Regressions

• Date: 01 October 2016
• Author: Rui Chen,Xueyong Zhang,Keqi Chen,Min Zhu
• DOI: http://doi.org/10.1111/ajfs.12152
• Published date: 01 October 2016

Chinese Stock Market Return Predictability:

Adaptive Complete Subset Regressions*

Keqi Chen

PBC School of Finance, Tsinghua University

Rui Chen**

School of Finance, Central University of Finance and Economics

Xueyong Zhang

School of Finance, Central University of Finance and Economics

Min Zhu

Business School, Queensland University of Technology

Received 26 January 2016; Accepted 7 September 2016

Abstract

This paper proposes a new combination framework to explore the Chinese stock market

return predictability. While most well-known predictor variables and simple combinations fail

to beat the historical average benchmark, our adaptive complete subset regressions deliver

statistically and economically significant out-of-sample performance. The subset, in which

each regression includes five predictors, produces a significant R2

OS statistic of 8.00% for Jan-

uary 2006 to September 2014. A mean-variance investor who uses the adaptive subset regres-

sions forecasts, instead of the historical average forecasts, can obtain sizable utility gains of

7.60% per annum. The results of our paper suggest that there is significant predictability in

the Chinese aggregate stock market portfolio.

Keywords Chinese stock market; Forecast combination; Out-of-sample predictability

JEL Classification: C53, G11, G12

*The authors would like to thank Kwangwoo Park (the Editor), an anonymous Associate Editor,

and two anonymous referees, Fuwei Jiang, as well as participants at the China Financial Engi-

neering Association 2016 meeting in Urumqi, for helpful comments and suggestions. Keqi Chen

acknowledges support from the Financial Academic Talents Training project of the School of

Finance, Central University of Finance and Economics. Rui Chen acknowledges financial sup-

port from the Program for Innovation Research at the Central University of Finance and Eco-

nomics. Xueyong Zhang acknowledges financial support from the Program for Innovation

Research at the Central University of Finance and Economics, and the annual research program

of the School of Finance, Central University of Finance and Economics. All errors are our own.

**Corresponding author: Rui Chen, School of Finance, Central University of Finance and

Economics, 39 South College Road, Haidian District, Beijing 100081, China. Tel: 86-10-6228-

8607, Fax: +86-10-6228-8607, email: r.chen@cufe.edu.cn.

Asia-Pacific Journal of Financial Studies (2016) 45, 779–804 doi:10.1111/ajfs.12152

©2016 Korean Securities Association 779

1. Introduction

Stock return predictability is of profound importance in many fields of finance

including asset pricing, portfolio allocation and risk management, and hence has

been one of the most researched areas in finance for decades. A range of studies

over the years present empirical evidence both for and against return predictability.

The current literature is predominantly biased toward the view that returns are pre-

dictable (Cochrane, 2008, 2011). However, how to best capture the predictable part

of the returns in terms of selection of predictor variables and a model framework is

still an open question. The current research in this area focuses almost exclusively

on the United States stock market. To advance studies into return predictability, a

wider range of data must be used and, hence, market types. The Chinese stock mar-

ket now ranks the largest among all emerging stock markets and the second-largest

among all national stock markets. Due to its huge economic size and rapid eco-

nomic growth in the past two decades, it makes academic and practical sense to

investigate Chinese market characteristics. Yet very little research has been carried

out to investigate whether stock return predictability exists in the Chinese stock

market and whether the methodology proposed for the United States market could

be applied successfully in the Chinese market. This paper fills a gap in the literature

by providing a comprehensive study on Chinese stock market return predictability.

An acknowledged challenge to predicting returns is instability in both parameter

and model, which is due to weak predictors and a time-varying relationship

between predictors and returns. There are two main channels to deal with the issue

of instability: variable combination and model combination. Variable combination

is in the spirit of combining all potential predictors in the hope of forming a stron-

ger predictive variable despite each predictor being weak. Principle component anal-

ysis (PCA) is a classic tool for this purpose. The second and much more

predominant approach to tackle instability in the literature is model combination.

Rapach et al. (2010) make a compelling argument on the necessity and benefits of

model combination for return forecasting. Rapach et al. use a simple combination

scheme —an equally-weighted combination of univariate regression models, which

are followed by a series of studies on how to best combine models to explore return

predictability. The most notable of these include Baysian model averaging (Dangl

and Halling, 2012) and an equally-weighted combination of multivariate regression

models (Elliott et al., 2013).

There are several interesting studies on Chinese stock market predictability.

Jiang et al. (2011) investigate Chinese stock return predictability, employing the

principal component approach and report positive results based on samples from

January 2002 to June 2009. Huang et al. (2016) employ a state-dependent model

that can generate forecasting gains over both good and bad times using a set of

United States economic variables. Chen et al. (2014) argue that downside market

risk is helpful in forecasting the Chinese stock market. However, the effectiveness of

model combination in the Chinese market has received limited attention. This study

K. Chen et al.

780 ©2016 Korean Securities Association