Technical Analysis Profitability Without Data Snooping Bias: Evidence from Chinese Stock Market

• Author: Fuwei Jiang,Guokai Song,Guoshi Tong
• Published date: 01 March 2019
• Date: 01 March 2019
• DOI: http://doi.org/10.1111/irfi.12161

Technical Analysis Profitability

Without Data Snooping Bias:

Evidence from Chinese Stock

Market*

FUWEI JIANG

†

,GUOSHI TONG

‡

AND GUOKAI SONG

†

†

School of Finance, Central University of Finance and Economics, Beijing, China and

‡

Hanqing Advanced Institute of Economics and Finance, Renmin University of China,

Beijing, China

ABSTRACT

We perform a comprehensive analysis on the profitability of a large number

of technical analysis based trading rules in Chinese stock market. To counter

data snooping bias, we employ a stepwise superior predictive ability test to

identify genuinely profitable trading rules among more than 28,000 techni-

cal signals. Using 19 years of daily data on Chinese aggregate stock market

return, we find substantial evidence on the profitability of technical trading

rules measured by either the market timing ability or Sharpe ratio gain. Our

results on the profitability of technical rules hold during different subperiods

and remain valid under the presence of transaction costs.

JEL Codes: C12; C52; G14

Accepted: 15 September 2017

I. INTRODUCTION

Technical analysis involves a set of trading rules that make buy and sell deci-

sions on the underlying security based on its past price patterns, trading vol-

umes and potentially other public information. Technical analysis is widely

adopted among security analysts, covered in financial news media and easily

accessible to most investors through various trading platforms. From a theoreti-

cal perspective, the use of technical analysis to forecast future return is justified

in a number of equilibrium models that feature heterogeneity in investors’

* We gratefully acknowledge the helpful comments from the Editor, an anonymous referee, and

feedbacks from Huafeng Chen, Po-Hsuan Hsu, Guofu Zhou, and seminar participants at Renmin

University and Central University of Finance and Economics. We also thank Prof. Po-Hsuan Hsu for

sharing with us a series of MATLAB and Fortran codes on superior predictive ability tests. This article

is supported by the National Natural Science Foundation of China (Nos. 71602198, 71572052), Bei-

jing Natural Science Foundation (No. 9174045), and the Program for Innovation Research in Central

University of Finance and Economics.

© 2017 International Review of Finance Ltd. 2017

International Review of Finance, 19:1, 2019: pp. 191–206

DOI: 10.1111/irfi.12161

access and response to information or feedback trading (e.g., Cespa and Vives

2012; Han et al. 2016). Empirically, however, evidence tends to be inconclusive.

While previous studies have found profitability of using technical analysis tools

in aggregate stock markets (e.g., Brock et al. 1992; Gencay 1998b; Lo

et al. 2000), in foreign exchange and bond markets (e.g., Gencay 1999; Goh

et al. 2013; Neely et al. 2014) and for individual stocks and portfolios

(e.g., Glabadanidis 2015, 2017), an influential work by Sullivan et al. (1999)

documented the absence of out-of-sample profitability and thus raised concerns

on the genuine predictiveness of technical analysis signals.

In this article, we add to the above studies by examining the profitability of

technical analysis trading rules using Chinese aggregate stock market data.

Complementary to many existing works that focus on various moving average

rules, we perform a comprehensive investigation on a large number of com-

monly used technical signals. In particular, we consider five categories of tech-

nical indicators: channel break rules; filter rules; moving average rules;

oscillator rules and support resistance rules. These five categories combined

with a range of plausible parameters setting provide us with 28,909 distinct

technical trading rules, which encompass the 2049 moving average rules con-

sidered in Brock et al. (1992) and 7846 trading rules in Sullivan et al. (1999).

We then evaluate jointly all these trading rules against the performance metrics

of market timing ability and Sharpe ratio gain against a buy and hold

benchmark.

We emphasize that since our empirical analysis relies on a single time series

data to test the profitability of a large number of trading rules, data snooping

bias becomes a concern.

1

To address this concern, we employ a stepwise supe-

rior predictive ability (step-SPA) test following a series of methodological studies

by White (2000), Romano and Wolf (2005), Hansen (2005) and Hsu

et al. (2010). This up-to-date inference procedure extends White’s reality check

test and allows to identify all genuinely profitable trading rules among the large

number of technical signals considered while controlling for data

snooping bias.

We apply this Step SPA test to gauge the predictiveness of technical analysis

in Chinese aggregate stock market. Using 19 years (1997–2015) of daily data on

all A-share index, we find that although the majority of technical rules consid-

ered seem to be profitable based on traditional t-test, less than 1% are genuinely

significant while eliminating data snooping bias. In particular, we identify

170 and 54 significantly profitable trading rules, respectively, under the metrics

of market timing ability and Sharpe ratio gain. These numbers drop to

144 (142) and 54 when we consider a one way transaction cost of 0.25%

(0.5%). We then summarize the parameters setting of all the identified rules

and list the specific forms of the top 10 strategies accounting for transaction

1 Since searching among competing trading rules implicitly involve multiple hypothesis testing

using a single data-set, the likelihood of incorrectly rejecting at least one of the null hypothe-

sis (Type I error) will increase.

© 2017 International Review of Finance Ltd. 2017192

International Review of Finance