Contrarian Strategy and Herding Behaviour in the Chinese Stock Market

Contrarian Strategy and Herding Behaviour in the Chinese Stock Market

Qiwei Chen*

Xiuping Hua†

Ying Jiang†

* Department of Economics and Finance, Brunel University, UK.

† Nottingham University Business School China

This version: June, 2015

Correspondence email:

Abstract

This paper investigates the profitability of several types of zero-cost price momentum and contrarian strategies in the Chinese stock market for the 1994-2013 period. Several distinct features of Chinese market are documented. We find that contrarian strategies that use Jegadeesh and Titman's (1993) method with weekly frequency are profitable. However, investment strategies based on the “nearness” to of 52-week high or the recency of the 52-week high are not profitable. Our analysis also shows that contrarian profits are higher during the crisis period of 2008-2012. In addition, the return reversal of the winner and loser portfolios suggests that contrarian profits can be attributed to overreaction. Finally, we also find evidence of herding behaviour in the Chinese market; and the degree of herding behaviour is positively correlated with the profits of contrarian trading strategies.

JEL Classification: G12; G14; G15; G32

Keywords: Momentum; Contrarian; Herding; Overreaction; China

1. Introduction

Over the past few decades, a vast literature has extensively examined the profitability of momentum and contrarian trading strategies in the stock market. Since the seminal paper of Jegadeesh and Titman's (1993, henceforth, JT), many studies have found that a strategy that holds a long position of past winners and a short position of loser stocks (the “momentum” strategy) often beats the market.[1] Other studies have however found that buying losers and short selling winners (the “contrarian” strategy) may also be profitable, as prices tend to revert after periods of extremely high and low returns (De Bondt and Thaler 1985, 1987; Chopra et al, 1992; Baytas and Cakici 1999). Studies have found that contrarian profits can be explained by a three-factor asset pricing model (Fama and French, 1996), as well as by the bid-ask spread bias (Conrad and Kaul, 1993), and liquidity (Cox and Peterson 1994). Turning to the momentum profits, they tend to be associated with several factors including the firm size (see Clare and Thomas 1995, and Zarowin 1990, Hong, Lim and Stein 2000), and trading volume (Connolly and Stivers 2003).

Another strandstream of the literature explains the profitability of momentum and contrarian strategies by price under-/overreaction. According to Barberis et al. (1998) stock prices underreact to earnings announcements, and overreact to consistent patterns of good or bad news. Hong and Stein (1999) model a market populated by two groups of agents, the "newswatchers" and "momentum traders". The market initially underreacts to firm-specific news because information does not always reach newswatchers instantly (hence the profitability of the momentum strategies). This initial underreaction is usually followed by overreaction, as momentum traders seek to make a profit by trend chasing, which inevitably causes prices to overshoot their long-run equilibrium values (hence the profitability of contrarian strategies). Therefore, it is possible to have both profitable momentum and contrarian strategies operating overfor different formation and holding periods.

Since JT, alternative strategies related to the price momentum have been proposed. George and Hwang (2004) (GH henceforth) examine a strategy that consists of buying stocks with a high current price to 52-week high price ratio. They find that those stocks with a high ratio (i.e. a price close to the 52-week high) outperform those with a low ratio. More recently, Bhootra and Hur (2013) (BH henceforth) consider a trading strategy based on the recency of the 52-week high (i.e. the number of days passed since the 52-week high occurred). It is found for the US market that the momentum strategy based on the recency of the 52-week high is profitable, and in addition it outperforms GH's momentum strategy. GH and BH argue that the profitability of their strategies is due to the existence of an anchoring bias (Tversky and Kahneman 1974) and a recency bias respectively. In the presence of such biases, stock prices are not adjusted to their fundamental values, which gives rise to momentum/contrarian profit.

In this paper, we investigate price momentum and contrarian profits in the Chinese stock market usingfor the three methods discussed above (JT, GH and BH) for the period 1994-2013. To the best of our knowledge, this is the first paper to conduct a comparison of these trading strategies for the Chinese market. Given that JT, GH, and BH momentum strategies are proved to be profitable in the US, it would be interesting to know whether they are equally profitable in an emerging market with a different composition of investors (i.e. fewer institutional investors, more individual investors), market size, liquidity etc.

A second contribution of this paper is to investigate the association between the profitability of momentum/contrarian strategies and herding behaviour. Herding is a well-established phenomenon (Grinblatt et al. 1995), including in the Chinese market (Tan et al. 2008; Yao et al. 2014). Because of herding, investors are trading in the same direction, which can lead to price overreaction (Avery and Zemsky 1998; DeLong et al. 1990). This, in turn, means that herding may indirectly lead to the profitability of momentum and contrarian strategies. In this paper we first test for herding using the approach of Chiang and Zheng (2010), before examining whether the profitability of the trading strategies is affected by the magnitude of herding.

Our results can be summarized as follows. First, the JT, BH, and GH momentum strategies are all not profitable. This finding is different from prior studies for developed markets.[2] Second, the only profitable contrarian strategy is one that ranks stocks using the JT approach, based onusing the weekly frequency. The profits of that strategy are enhanced during the financial crisis period of 2008-2012. These results are robust to size and bid-ask bounce effects. Third, we confirm the existence of herding in the Chinese stock market. Herding is particularly strong for stocks with amongst poorly performing stocks. Finally, we find the contrarian profits are higher during high herding periods, which indicates that there is an association between contrarian profits and herding behaviour.

This paper is organized as follows. Section 2 introduces the data and methodology. Section 3 presents the results for the profitability of the trading strategies. Section 4 shows the robustness of the results to Fama-MacBeth regressions. Section 5 tests for herding behaviour, and subsequently shows the link between contrarian profits and herding. The final section concludes.

2. Data and Methodology

2.1 Data

Our data sample consists of all listed "A" stocks of the Shanghai Stock Exchange and Shenzhen Stock Exchange over the period January 1st 1994 to December 31st 2013.[3] The number of stocks increases from 167 at the beginning of the sample period to 2467 at the end of the sample period. In order not to be affected by survivor biases, we do not require a stock to exist during the entire holding period (Brown et al. 1995). We are therefore using an unbalanced panel, and the number of stocks varies over time. Firm-level data is obtained from the Wind Information Co. Ltd database, and these include monthly/weekly/daily split and dividend adjusted closing prices, day-high prices, monthly/weekly market values and yearly book-to-market ratios. Logarithmic rate of returns are calculated using the closing prices for each data frequency.

2.2 Contrarian and momentum trading strategies

We examine the profitability of three types of zero-cost contrarian/momentum trading strategies. These are the strategies proposed by Jagedeesh and Titman (1993) [JT], George and Hwang (2004) [GH], and Bhootra and Hur’s (2013) [BH].

2.2.1 Jagedeesh and Titman (1993)’s strategy

We first consider the approach of JT. For each period t, stocks are ranked in ascending order according to their past performance (i.e. the cumulative return) over the previous J periods. Based on that ranking, two equally weighted portfolios are formed: stocks ranked in the top 10% are assigned to the winner portfolio, and stocks in bottom 10% are assigned to the loser portfolio. We form a zero-cost momentum portfolio by longing the winner portfolio and shorting the loser portfolio, while holding the position for K periods. We call JT momentum strategy the strategy of investing in a momentum portfolio using the JT ranking method. Likewise, a contrarian portfolio can be formed by longing the loser portfolio and shorting the winner portfolio. We call JT contrarian strategy the strategy of investing in a contrarian portfolio using the JT ranking method. For each period t of a (J, K) strategy, the returns to winners/losers are calculated as the equally weighted average of the period t returns from K separate winner/loser portfolios, each formed in one of the J consecutive prior periods t – J to t – 1. The return to the overall strategy is the difference between the return to winners and to losers in period t.

We use both the conventional monthly frequency as well as the weekly frequency. For monthly data, the portfolios are formed at the end of each month; and, for weekly data, the portfolios are formed each Wednesday (if the day is a non-trading day, then the next trading day is used) in order to avoid the weekday seasonalities (e.g. Monday effect or Friday effect). For the length of the formation and holding periods, we choose 1, 3, 6 months for the monthly frequency, and 1, 2, 3 weeks for the weekly frequency. We impose a gap between the formation and the holding periods to mitigate the bid-ask bounce effects. We impose a one-month gap and a one-week gap for the monthly frequency and weekly frequency respectively.

2.2.2 George and Hwang (2004)’s strategy

We then estimate the profitability of the trading strategies proposed by GH, which are based on the “nearness” to the 52-week high price computed over a specific period. The nearness is defined as the ratio of the current stock price to the 52-week high price. This ratio is written as , where is the price of stock i at the end of period t-1 and is the highest price of stock i during the last 52 weeks. A high value indicates that the current price is close to its 52-week high. Stocks are ranked based on the nearness ratio, and, similar to JT, winner and loser portfolios are formed (with top 10% and bottom 10% stocks ranked on the nearness). We then calculate, for weekly and monthly frequency, the return of GH momentum strategies (long the winner portfolio, short the loser) and GH contrarian strategies (long the loser portfolio, short the winner).

2.2.3 Bhootra and Hur’s (2013)’s strategy

Finally, we consider the strategies proposed by BH (2013), which also make use of the 52-week high. However, and unlike GH, stocks are ranked on the basis of the “recency ratio” or RR, which is calculated as follows:

RR=

The recency ratio is between 0 and 1 and is inversely related to the number of days that have passed since the stock hit the 52-week high price. The ratio is high for stocks that have recently reached their 52-week high and low for stocks that whose 52-week price occurred many months ago. Similar to GH and JT, we form the loser and winner portfolios and calculate the return of momentum and contrarian strategies.

The table in Appendix summarizes the ranking criteria for our three strategies. It should be noted that the analysis of the profitability of BH and GH strategies can also provide information about the sources of momentum profits.

2.3 Test of herding

One of the objectives of this paper is to investigate some of the linkages between herding and the profitability of trading strategies. In the presence of herding, prices are temporarily pushed away from their fundamentals, i.e. they overreact. For that reason, it is believed that herding may indirectly lead to both momentum and contrarian profits for different time windows, as prices initially deviate from their fundamental value before ultimately mean revertsing. According to Hong and Stein (1999), these price fluctuations can be explained by the existence of trend chasers who try to make a profit by buying stocks with recent positive returns and short selling the losing ones.

We first test for herding using the herding detection model of Chiang and Zheng (2010), which is a generalised form of the model initially proposed by Chang et al. (2000). In the context of this study, herding is defined as a situation where investors mimic financial gurus or follow the activities of successful traders rather than relying on their own information (see Chiang and Zheng, 2010).

Individual assets differ in their sensitivity to the market return, and rational asset pricing models predict a linear relationship between the market return and the cross-section dispersion of stock returns. In the case of herding, however, investors may suppress their own beliefs and follow the market consensus during periods of large market movements. Therefore, herding predicts that the cross-sectional dispersion of returns should decrease or increase less than proportionally with the market return, as investors are drawn to the consensus of the market (Christie and Huang, 1995). Herding can be detected using the following regression (see, Chiang and Zheng, 2010):

(1)

where CSAD, the cross-sectional absolute deviation, is a measure of return dispersion:

(2)

where N is the number of stocks included in the portfolio, and is the observed stock return of stock i for period t. is the market return , i.e. the cross-sectional average of stock returns in the portfolio for period t. We estimate equation (1) with a Newey-West consistent estimator (Newey West 1987). A negative coefficient for shows that the dispersion of returns is increasinges at a decreasing rate with the market return, which is conventionally interpreted as evidence of herding.

We estimate model (1) using the whole sample to test for market-wide herding. We also estimate the model separately for the winner and loser portfolios to capture possible differences in the degree of herding in the two sub-groups. As is standard in the literature, the herding-detection model is estimated at a in daily frequency, and the portfolios are re-balanced each week (on Wednesdays). Each Wednesday we take the daily returns of the winner and loser portfolios selected for that week, and switch to a new set of portfolios’ daily returns the following Wednesday, so as to obtain a continuous time series of daily returns.

In order to investigate the relationship between momentum/contrarian profits and herding, we divide the sample into high and low herding periods. Because herding behaviour results in abnormally low stock dispersion during periods of large price movements, we propose to define high herding periods as periods where both the value of is above the 30% (or 50%) percentile and is below the 30% (or 50%) percentile. These are periods during which return dispersion is low in spite of the large absolute market return. Likewise, low herding periods are periods where both is above 30% (or 50%) percentile, and is above 30% (or 50%) percentile. Note that the 30% and 50% cutoff points are somewhat arbitrary, but our results would still hold using alternative cutoff points. We then compare momentum/contrarian profits in high and low herding periods to detect a possible association between the trading strategies’ profits and herding.

3. Profitability of trading strategies

In this section we report the profitability of the JT, GH, and BH trading strategies, first for monthly returns and subsequently for weekly returns.

3.1 Trading strategies using monthly returns

Table 1 shows the profitability of the trading strategies using the monthly frequency, for 1, 3 and 6 months holding periods. For JT, we consider formation periods of J=1, 3, and 6 months (JT-1, JT-3, JT-6), as explained above. Portfolios for the GH and BH strategies are based on the end-of-month nearness and recency ratios, respectively. Table 1 reveals that none of the strategies (JT, GH and BH) generates statistically significant momentum or contrarian profits using the monthly frequency. For instance, for strategies with 6-month holding period, the momentum profit for BH is -0.0002, and that for GH is 0.0029, both insignificant. This findings contrast with prior studies showing the profitability of momentum strategies on the US market. However, our results are consistent with Chen et al. (2012), which find no momentum profits for the Chinese stock market for the monthly frequency. This suggests that there is no price underreaction.

[Insert Table 1 here]

Table 1 also shows the returns for the winner and loser portfolios separately. BH (2013) find that the GH momentum profit can be attributed to the extremely low loser portfolio returns, while with the recency ratio measure, the winners contribute more to the overall raw momentum profits. In our sample, however, we find no significant difference between winner and loser portfolios for all momentum strategies.

3.2 Trading strategies using weekly returns