Institutional Information Manipulation and Individual Investors’ Disadvantages: A New Explanation for Momentum Reversal on the Chinese Stock Market

ABSTRACT

In this study, empirical evidence is presented to explain the momentum reversal phenomenon in the Chinese stock market in terms of the manipulation of institutional information. On the institutional “sell” side, we demonstrate that institutional traders send manipulated information to the market using a large volume of buy orders in order to boost the stock price and thus induce trading by retail investors. The reverse is also true on the institutional “buy” side. Thus instead of the traditional view of order flow information—“the more, the merrier”—the authors argue that “more is less” in the case of individual investors on the Chinese stock market. As a result, the empirical results presented in this study provide another feasible explanation for momentum reversal.

KEY WORDS:

• individual investors
• manipulation of institutional information
• momentum reversal

JEL:

• G11
• G12
• G14
• D83

Notes

1. Among others in favor of momentum continuation, Griffin showed a different result, claiming that stocks that experience high (low) volumes have subsequent positive (negative) excess returns.

2. A counterexample was provided by Griffin, Harris, and Topaloglu (2003). They used daily and intraday trading data to demonstrate that strong contemporaneous daily patterns can be explained largely by institutional positive trading following excess stock returns. Our study shows that this may not be the case in the Chinese stock market on a weekly basis.

3. Noise traders’ risk was also elaborated by De Long et al. (1990).

4. A sample of the full dataset is available upon request for nonprofit use only. The data were purchased through the National Science Research Funding project.

5. This method is supported by Easley and O’Hara (1987), Ülkü (2008), Blau, Van Ness, and Van Ness (2009), Chakravarty (2001), and Alexander and Peterson (2007), among others. However, a counter-opinion was given by Barclay and Warner (1993). They argued that informed institutional traders might break up big orders into medium-size and small orders to avoid revealing their true trading direction. This claim is irrelevant for our purpose. If an institutional trader wants to send a “signal” to the market, he needs to do it in a noticeable way. If he uses stealth trading, he cannot attract the attention of individual investors. Thus, we argue that what a retail investor sees in the market is what is released for him/her to see. As a result, we use big orders only here.

6. The prices have been adjusted for stock splits and dividends.

7. Small-cap stocks have RMB 10 billion or less in market value, while large-cap stocks have more than RMB 10 billion in market value.

8. The algorithm begins by first determining turning points$P_t$, which are defined as $P_{t-6}...P_{t-1}⟨P_t⟩P_{t+1}...P_{t+6}$ for peaks and $P_{t-6}...P_{t-1}>P_t<P_{t+1}...P_{t+6}$ for troughs. Then, bull/ bear periods are identified as periods between troughs/peaks and peaks/troughs. Throughout this study, we assume that two market conditions (bull and bear) are sufficient. This algorithm is also adopted by Chauvet and Potter (2000).

9. The serial correlations between concurrent and lagged institutional/individual trading have been tested and remedied using the following procedure. First, we ran cross-sectional autoregression tests on the model to specify the orders of the autoregressive model AR(l), l = 1, 2, 3 (in most cases, we found AR(1) is sufficient) for the error term. If AR(l) is found to be significant, a dynamic panel data model is used with lagged institutional/individual behavior variables on the right-hand side of the model. But if AR(l) is not significant, then the original models are appropriate. We found that serial correlations between concurrent and lagged institutional/individual trading are significant, and thus the models are adjusted accordingly, which affects all the models for H1, only panels C and D for H2, and all models for the robustness tests. In addition, the independent explanatory variables may also have multicollinearity. We use VIF to test this effect and omit variables for which VIF>10. We found that the variables removed by VIF are the same as the ones removed by a simple correlation test. All models are tested following the same procedures.

10. A counterexample result was given by Bayar (2013), in which the author developed a dynamic model to prove that liquidity provided with limit orders could attenuate an adverse selection problem.

Additional information

Funding

This study is funded by the National Natural Science Foundation of China, Grant No. 71473204. The authors thank three anonymous reviewers and English-language editor Debra E. Soled for making this study much better;Fundamental Research Funds for the Central Universities [JBK18FG16,JBK1902054].