Further Mining the Predictability of Moving Averages: Evidence from the US Stock Market

• DOI: http://doi.org/10.1111/irfi.12166
• Author: Chaoqun Ma,Danyan Wen,Yong Jiang,Gang‐Jin Wang
• Date: 01 June 2019
• Published date: 01 June 2019

Further Mining the Predictability

of Moving Averages: Evidence from

the US Stock Market*

CHAOQUN MA,DANYAN WEN,GANG-JIN WANG AND YONG JIANG

Department of Management Science, Business School of Hunan University,

Changsha, China

ABSTRACT

Most studies on the predictability of moving average (MA) technical analysis

use the discrete (buy/sell) trading recommendations. However, it is possibly

incomplete or unreliable to explore the predictability of MA by only employ-

ing its generated trading signals. To further explore the forecastability of MA,

we study its measurable impact on the stock market returns by using a con-

ventional predictive regression framework. Our empirical study on the US

stock market with respect to more detailed price information finds, (i) that

the proposed predictor, MADP (MA based on daily prices) shows significant

predictability in- and out-of-sample, and significantly outperforms the histor-

ical average (HA) benchmark as well as the MA based on monthly prices,

(ii) that the predictability of MADP centers on the short-term lags (within

the most recent 10 days) and disappears when lags are beyond 20 days, and

(iii) that the economic evaluation of the portfolios based on trading strate-

gies confirms the superior performance of MADP with short-term lags against

the benchmark even though considering transaction costs.

JEL Codes: G17; C53; G11; G12

Accepted: 1 November 2017

I. INTRODUCTION

Stock return forecastability has been considered as a “stylized fact.”While some

researchers employeconomic variables to predict the stockmarket (e.g., Cochrane

2006), more and more studies provide the evidence that technical indicators,

especially the moving average (MA),

1

have strongly predictive power in the stock

market (e.g., Brock et al. 1992; Gençay and Stengos 1998; Gençay 1998b; Lento

and Gradojevic 2007; Neely et al. 2014; Glabadanidis 2015; Marshall et al. 2017;

* The work of CM and DW was supported by NSFC (Grant nos. 71431008 and 71521061). The work

of GJW was supported by NSFC (Grant no. 71501066) and the Hunan Provincial Natural Science

Foundation of China (Grant no. 2017JJ3024).

1 Here the MA is a technical indicator rather than the moving-average (MA) model in time

series.

© 2017 International Review of Finance Ltd. 2017

International Review of Finance, 19:2, 2019: pp. 413–433

DOI: 10.1111/irfi.12166

Mitra et al. 2017; Zakamulin 2017). Brock et al. (1992) seems the first major study

to provide strong support of the predictability of MA trading indicators. Subse-

quently, the predictive power of MA is also confirmed in other financial markets,

like commodity futures (Han et al. 2016a; Lee and Brorsen 2017) and foreign

exchanges (Gençay 1999; Gradojevic and Lento 2015). To enhance the predict-

ability of the MA indicator, much literature considers trading uncertainty, infor-

mation uncertainty, or the combination with other trading signals (Lento and

Gradojevic 2007; Kozyra and Lento 2011; Gradojevic and Gençay 2013; Chen

et al. 2016). Contemporarily, with the popularity of algorithmic trading, lots of

studies employ machine learning techniques, for example, artificial neural net-

works (ANNs) and support vector machine (SVM), to improve the predictability of

markets through the MAtrading signals (Gençay 1996, 1997, 1998a;Kaucic 2010;

Karathanasopoulos et al. 2016 and among others).

Although the studies on the predictability of MA have achieved great success,

most existing literature only focuses on recommending the up or down state of

certain assets. In other words, those studies usually provide buy or sell signals.

However, it is possibly incomplete or unreliable to explore the predictability of

MA without providing its measurable impact on future returns. On the one

hand, further mining the predictability of stock returns helps researchers to

construct more realistic asset pricing models that can better explain the data.

On the other hand, the ability to precisely forecast returns has important impli-

cations for the practitioners to enhance the investment performance. To

directly measure the MA’s impact on stock returns, Zhou and Zhu (2013) pro-

pose an equilibrium model suggesting that it is possible to theoretically link the

predictability of MA to stock returns.

2

However, the empirical evidence of MA’s

direct impact on stock returns has not been thoroughly explored. In this paper,

we provide an empirical study for addressing three problems related to the mea-

surable predictability of MA on stock returns: (i) What is the statistical perfor-

mance of the MA predictors in the in- and out-of-sample tests? (ii) Which part

of the past price information dominates the predictability of MA on stock mar-

ket returns? (iii) Whether there exists and how to assess the economic gains of

the selected predictive models based on trading strategies? To sum up, our goal

here is to measurably explore the predictability of MA both in- and out-of-

sample with respect to more detailed past price information.

By using a conventional predictive regression framework, we investigate the

forecastability of MA on the US stock market data spanning over January

2, 1963 to December 31, 2015. Specifically, we employ more detailed past price

information to construct the predictor MADP (MA based on daily prices) com-

paring with the MAMP (MA based on monthly prices) in the popular bivariate

and principal component predictive regressions. Then, we attempt to evaluate

the out-of-sample predictability with statistical tests between one of the models

we estimate and the historical average (HA) benchmark model. Next, we assess

2 Based on the equilibrium model, Han et al. (2016b) construct a trend factor cross-sectionally

and find that the trend factor performs better than the momentum factor.

© 2017 International Review of Finance Ltd. 2017414

International Review of Finance