Elsevier

Journal of Empirical Finance

Volume 59, December 2020, Pages 257-277
Journal of Empirical Finance

Does program trading contribute to excess comovement of stock returns?

https://doi.org/10.1016/j.jempfin.2020.11.001Get rights and content

Highlights

  • Program trading is highly persistent, showing the features of habitat investing.

  • There exists excessive return comovement in stocks preferred by program traders.

  • The identified return comovement is disconnected with fundamentals.

  • Habitat investing theory well explains program-trading-induced return comovement.

Abstract

Daily returns of stocks with high program trading comove more with each other but less with others. This significant comovement is disconnected with market movements and news of fundamentals and becomes stronger when market uncertainty is higher. It can be explained by neither the hypotheses of gradual information diffusion and liquidity provision nor the effects of quantitative trading signals, earnings announcements and index fund trading. Its non-fundamental nature is further demonstrated by the observation of program trading stimulating return reversals. Underlying this comovement is the high persistence of program trading. Our findings support the theory of habitat investing and demonstrate program trading creates a distinct source of excess return comovement.

Introduction

Stock return comovement lies at the heart of modern portfolio theory and is extremely important for investors making decisions on portfolio allocation and risk management. Following the traditional view that stock price equates to the present value of future cash flows, comovement in stock returns is perceived as the result of commonality in firm fundamentals or change in common discount rates. However, this simple view is not supported by empirical evidence. Shiller (1989) argues that return comovement between the U.S. and U.K. stocks is too large to be explained by their comovement in dividends. Pindyck and Rotemberg (1993) find that the returns of companies conducting unrelated business comove beyond common variation in discount rates. Contemporary literature focuses on the return comovement in excess to variation in common fundamentals, and the sources of such comovement include S&P 500 index inclusion and deletion (Barberis et al., 2005; hereafter BSW), overweighting of some constituents in the Nikkei 225 Index (Greenwood, 2007), change in firm headquarters location (Pirinsky and Wang, 2006), systematic retail trading (Kumar and Lee, 2006), reclassification of index labels (Boyer, 2011), common mutual fund ownership (Antón and Polk, 2014), and stock splits (Green and Hwang, 2009). These empirical findings are generally consistent with the theories developed by Barberis and Shleifer (2003) and BSW, which predict that correlated excess demands drive stock returns to comove beyond their common fundamentals. However, most existing studies take an indirect approach by identifying non-fundamental characteristics of events and then examining the effects of such events on return comovement.1 While such an empirical design is generally viewed as successful in isolating non-fundamental comovement, it provides little guidance about who these non-fundamental traders are and the underlying channel through which excess demand rises. A recent study by Chen et al. (2016) points out that some non-fundamental events used in the literature may coincide with changes in fundamentals. Excess comovement among stocks following index additions and stock splits even disappears when revisited.

This study documents empirical evidence of excess comovement caused by the trading activities of a group of non-fundamental traders—program traders. The New York Stock Exchange (NYSE) defines program trading as: (1) simultaneous purchase or sale of 15 or more stocks as part of a coordinated trading strategy; or (2) index arbitrage.2 The fact that program trading only trades stock baskets makes it unlikely driven by fundamentals,3 which provides an ideal and direct setting for testing excess return comovement.

Since its emergence in the mid-1980s, program trading has been widely employed by institutional investors to minimize their trading costs and implement trading strategies that involve multiple stocks. It has increased dramatically in recent years and its trading volume accounts for about 22% of the total trading volume over our sample period from 2006 to 2015. Program trading covers almost all types of computerized trading that involves a purchase or sale of 15 or more stocks originated from a coordinated strategy.4 The rise of computerized trading has triggered intense debate about its impact on market stability. Using exogenous variations in High Frequency Trading (HFT) in the Japanese and European equity markets, Jain et al. (2016) and Malceniece et al. (2019) show that HFT facilitates speedier incorporation of fundamental information and increases systemic risk. However, little is known about the impact of program trading on non-fundamental return comovement. The flash crash of 6 May 2010 in the U.S. represents one of the most dramatic events in the history of financial markets and is unlikely driven by changes in fundamentals. Some market practitioners blame program trading as the culprit, questioning whether it adversely impacts on market quality.5 We study program trading from a new perspective by revealing its role in stimulating non-fundamental comovement in stock returns.

Our study is motivated by the habitat investing view of excess comovement proposed by BSW, which starts from the observation that many investors choose to trade only a subset of all available securities, for reasons such as transaction costs and/or lack of information. We find that stocks with a high (low) program trading participation ratio in a quarter tend to hold the ratio high (low) in the following four quarters or longer. The persistence in program trading at the individual stock level shows typical characteristics of habitat investing.

To test whether habitat investing leads to excess comovement among stocks preferred by program traders, we adopt a two-step process similar to the one adopted by Koch et al. (2016). In the first step, we investigate how the daily return of an individual stock comoves with the portfolio return of high program trading stocks, using either the bivariate approach of BSW or residual returns to control for market fundamentals. In the second step, we estimate the cross-sectional relationship between the comovement and the program trading of the current or previous quarter. We find that program trading exerts a significant and positive effect on excess comovement measured by both approaches, and the effect remains strong after controlling for stock characteristics. The bivariate approach also indicates stocks preferred by program traders comoving less with the rest of the market. These findings are consistent with the habitat investing view of BSW and Greenwood (2007) that after controlling for common fundamentals, the return of a stock comoves more with other stocks within its habitat but less with stocks outside its habitat.

We adopt alternative econometric specifications to confirm our findings’ robustness to errors in variables. By using instrumental-variable regressions to address endogeneity concerns, we establish a causal effect of program trading on return comovement, which is substantiated by our finding from a market-wide test that program trading Granger causes return comovement but not vice versa.

To further test the non-fundamental nature of the identified return comovement, we analyze the impact of macroeconomic, industry and firm-specific information and none of them is able to explain the comovement produced by program trading. Additionally, we examine whether program trading stimulates return reversals because stock returns driven by non-fundamentals should reverse in subsequent periods. We are able to confirm that program trading results in a common reversal component in stock returns. These findings support the notion that the identified return comovement is excessive and non-fundamental.

Although program trading exhibits characteristics of habitat investing, gradual information diffusion can also generate excess comovement in stock returns (BSW; Chordia and Swaminathan, 2000). As program traders are primarily sophisticated institutions, stocks preferred by them may incorporate market information quicker than other stocks, resulting in the identified return comovement. An implication of this hypothesis is that the returns of stocks with high program trading should positively lead their counterparts with low program trading. We test this prediction by running a vector autoregression on daily portfolio returns of stocks with high and low program trading, and our results do not support the hypothesis of gradual information diffusion.

Liquidity provision offers another potential explanation for the identified return comovement if program traders serve as liquidity providers. The hypothesis of liquidity provision predicts a stronger relationship between program trading and return comovement during market declines (Hameed et al., 2010). The relationship is also asymmetric between periods with large negative and positive market returns (Ang and Chen, 2002). However, we do not find evidence to support these implications and predictions. On the other hand, if program traders trade stocks within their preferred habitat, their correlated order flows tend to have greater price impact when the market is more turbulent. There is evidence consistent with this conjecture of habitat investing, as the relationship between program trading and return comovement during high-volatility and crisis periods is stronger.

Quantitative trading signals and earnings announcements offer two potential channels to link program trading and return comovement (Stambaugh et al., 2012, Patton and Verardo, 2012). We use stock mispricing for trading signals and find the participation of program trading in overpriced stocks is lower rather than higher. The participation rate is also lower in months when stocks announce their earnings. The effect of program trading on return comovement is stronger (weaker) for overpriced (underpriced) stocks. However, their relationship remains highly significant regardless of a stock’ mispricing level and whether it has an earnings announcement or not. This implies that the identified return comovement is not driven by trading signals or earnings announcements.

Program trading is widely adopted by index fund managers (e.g., Exchange-Traded Funds (ETFs) and index mutual funds) to accommodate retail investment flows. Hendershott and Seasholes (2009) find that after a stock is added to the S&P 500 index, its order flow from program trading starts to comove more with program trading order flow of other S&P 500 stocks. Therefore, index fund trading activities are likely to contribute to our results, especially if fund managers prefer and dominate program trading. Nevertheless, the effect of program trading on the identified comovement remains robust after controlling for ETF flows and index returns, indicating program trading as a distinct driver of excess comovement among stocks within their habitat.

The rest of this paper is organized as follows. Section 2 reviews the relevant literature. Section 3 describes our data and variable constructions. Our baseline analysis is presented in Section 4, followed by extended analysis to further demonstrate the non-fundamental nature of the comovement and to test alternative explanations in Section 5. Section 6 concludes the paper.

Section snippets

Literature review

Program trading involves baskets of securities and is often considered as non-informational in literature. Subrahmanyam’s (1991) theoretical model predicts that portfolio trading is less likely to be information-driven compared to trading individual securities. Hendershott and Seasholes (2009) show that program trading tends to make losses, while Kadan et al. (2018) argue it is not motivated by news. In practice, sell-side brokers maintain separate trading desks and charge lower commission fees

Data and variable measurement

We obtain historical data of program trading from the NYSE’s data product of ProTrac, which contains information about daily program trading volumes of all NYSE stocks from 2006Q3 to 2015Q4. We collect daily return and volume data of all NYSE common stocks (share codes 10 and 11) from CRSP, and balance sheet items from COMPUSTAT. Data of analyst coverage and institutional ownership are collected from I/B/E/S and SEC Form 13F.

The main explanatory variable, Program Trading Participation (PTP) of

Persistence of program trading activities

The habitat investing theory of BSW argues that many investors prefer to trade only a subset of all available stocks, and hence contribute to excess return comovement. To verify that some stocks are indeed a preferred habitat to program traders, we first examine the persistence of program trading. We sort stocks into quintiles according to PTP in a quarter and then report the averages of PTP in the current and four subsequent quarters in Panel A of Table 3. For all PTP quintiles, PTP is quite

Extended analysis

Our analysis so far suggests a strong common return component exists among stocks with high program trading. In this section, we elaborate further on the non-fundamental nature of the identified return comovement. We also test some alternative explanations for this comovement.

Conclusion

Program trading is highly persistent and some stocks are more likely to be in the investment habitat of program traders. We provide strong evidence showing that return comovement in stocks preferred by program traders is excessive. It cannot be explained by macroeconomic, industry, and firm-specific information, consistent with its non-fundamental nature. Program trading leads to a common reversal component in subsequent periods, adding systemic noise to stock returns. We also demonstrate that

CRediT authorship contribution statement

Mingyi Li: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Resources, Software, Validation, Writing - original draft, Writing - review & editing. Xiangkang Yin: Conceptualization, Funding acquisition, Investigation, Methodology, Project administration , Resources, Supervision, Validation, Writing - original draft, Writing - review & editing. Jing Zhao: Conceptualization, Funding acquisition, Investigation, Methodology, Project

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

References (53)

  • GreenT. Clifton et al.

    Price-based return comovement

    J. Financ. Econ.

    (2009)
  • HagströmerBjörn et al.

    The diversity of high-frequency traders

    J. Financial Mark.

    (2013)
  • HasbrouckJoel et al.

    Common factors in prices, order flows, and liquidity

    J. Financ. Econ.

    (2001)
  • JainPankaj K. et al.

    Does high-frequency trading increase systemic risk?

    J. Financial Mark.

    (2016)
  • MalcenieceLaura et al.

    High frequency trading and comovement in financial markets

    J. Financ. Econ.

    (2019)
  • StambaughRobert F. et al.

    The short of it: Investor sentiment and anomalies

    J. Financ. Econ.

    (2012)
  • AntónMiguel et al.

    Connected stocks

    J. Finance

    (2014)
  • BoehmerEkkehart et al.

    Order flow and prices

    (2008)
  • BoyerBrian H.

    Style-related comovement: fundamentals or labels?

    J. Finance

    (2011)
  • BrogaardJonathan et al.

    High-frequency trading and price discovery

    Rev. Financ. Stud.

    (2014)
  • ChordiaTarun et al.

    Trading volume and cross-autocorrelations in stock returns

    J. Finance

    (2000)
  • DaZhi et al.

    Exchange traded funds and asset return correlations

    Eur. Financial Manag.

    (2018)
  • FurbushDean

    Program trading and price movement: evidence from the 1987 market crash

    Financ. Manage.

    (1989)
  • FurbushDean

    Program trading

  • GowIan D. et al.

    Correcting for cross-sectional and time-series dependence in accounting research

    Account. Rev.

    (2010)
  • GrangerClive W.J.

    Investigating the causal relations by econometric models and cross-spectral methods

    Econometrica

    (1969)
  • Cited by (0)

    We are grateful to Tarun Chordia, Huu Duong, Philip Gharghori, Yuting Gong, Alok Kumar, Peter Pham, Avanidhar Subrahmanyam, and participants at the 9th Financial Markets and Corporate Governance Conference, the 26th Conference on the Theories and Practices of Securities and Financial Markets, and the Research Symposium on Capital Market Research for their highly valued comments. This project is supported by the Discovery Projects funding provided by the Australian Research Council (DP140100113).

    View full text