Flow toxicity of high‐frequency trading and its impact on price volatility: Evidence from the KOSPI 200 futures market

• DOI: http://doi.org/10.1002/fut.22062
• Author: Wooyeon Kim,Jangkoo Kang,Kyung Yoon Kwon
• Date: 01 February 2020
• Published date: 01 February 2020

J Futures Markets. 2020;40:164–191.wileyonlinelibrary.com/journal/fut164

|

© 2019 Wiley Periodicals, Inc.

Received: 9 September 2019

|

Accepted: 17 September 2019

DOI: 10.1002/fut.22062

RESEARCH ARTICLE

Flow toxicity of high‐frequency trading and its impact

on price volatility: Evidence from the KOSPI 200

futures market

Jangkoo Kang

1

|

Kyung Yoon Kwon

2

|

Wooyeon Kim

3

1

Graduate School of Finance and

Accounting, College of Business, Korea

Advanced Institute of Science and

Technology, Seoul, Korea

2

Department of Accounting and Finance,

Strathclyde Business School, University of

Strathclyde, Glasgow, United Kingdom

3

College of Business, Korea Advanced

Institute of Science and Technology,

Seoul, Korea

Correspondence

Wooyeon Kim, College of Business, Korea

Advanced Institute of Science and

Technology, 85, Hoegi‐ro, Dongdaemoon‐

gu, Seoul 02455, South Korea.

Email: qdragon326@kaist.ac.kr

Abstrast

We examine the relation between high‐frequency trading, flow toxicity, and

short‐term volatility during both normal and stressful periods. Using transac-

tion data for the Korea Composite Stock Price Index 200 (KOSPI 200) futures,

we find the Volume‐Synchronized Probability of Informed Trading (VPIN)

useful in measuring flow toxicity as it predicts short‐term volatility effectively.

We further show that high‐frequency trading is negatively related to VPIN and

short‐term volatility during normal times but has a positive association during

stressful periods. Finally, we advocate the use of bulk‐volume classification

(BVC) by presenting evidence that the initiator identified by BVC trades at more

favorable prices than the true trade initiator.

KEYWORDS

high‐frequency trading, order flow toxicity, short‐term price volatility, volume‐synchronized

probability of informed trading

JEL CLASSIFICATION

G14

1

|

INTRODUCTION

Market making promotes efficiency by facilitating the process of price discovery, which is the central function of a

market. In the late 20th century, when floor markets were still pervasive, exchange specialists (e.g., designated market

makers on the NYSE and NASDAQ) ensured market making. However, after exchanges adopted the structure of an

electronic limit order book and with regulatory changes, bids and asks from endogenous liquidity providers (ELPs), or

proprietary market makers, replaced those of exchange specialists. Moreover, with the emergence of algorithmic

trading in modern markets, high‐frequency traders (HFTs) are capturing a considerable part of market making by

generating orders from complex computer algorithms.

These new traders reveal different trading patterns compared with traditional market makers,

1

which has raised

concerns that they may harm market health. In particular, HFTs may exploit their advantage of superior speed to

increase the level of adverse selection in a market by generating toxic order flow and, in turn, amplify systemic market

risks. Specifically, if HFTs elevate the level of adverse selection, and if large orders enter the market during such

1

For example, Kirilenko, Kyle, Samadi, and Tuzun (2017) show that inventory changes of HFTs are positively associated with contemporaneous price changes, whereas those of market makers are

negatively associated with contemporaneous price changes in the 1‐s clock‐time interval in the E‐mini S&P 500 futures market. In the Korea Composite Stock Price Index 200 (KOSPI 200) futures

market, Kang, Kang, and Kim (2018) indicate that HFTs reveal directional trading during extreme price movements, whereas other types of traders, including market makers, absorb volume

imbalances created by HFTs. Therefore, the shift to high‐frequency markets, after the emergence of HFTs, changes the shape of liquidity provision/demand and market making.

periods, a liquidity‐driven market crash can occur even in the absence of fundamental shocks since liquidity providers

reduce or liquidate their positions and exit the market. In view of HFTs with high‐speed order submission and

cancellation now dominating liquidity provision, the market can collapse within a few minutes.

2

From this perspective, it is vital to examine how high‐frequency trading relates to market‐wide order flow toxicity.

However, the theoretical standpoints are mixed. On the one hand, HFTs make markets using their speed advantage to

avoid the risk of “being picked off ”by informed traders and provide liquidity skillfully, which leads to the likelihood of

other investors finding them as trading counterparties. In addition, as pointed out in Brogaard, Hendershott, and

Riordan (2014), HFTs facilitate the process of price discovery, which reduces information asymmetry between informed

traders and slow liquidity providers. Collectively, they tend to decrease order flow toxicity. On the other hand, using

their speed advantage, HFTs can pick off slow traders so that they increase adverse selection and the level of market‐

wide flow toxicity. According to Kang et al. (2018), HFTs, who are not obliged to stabilize the markets during stressful

periods, actually trade in the same direction of extreme price movements, implying that their order flow could be highly

toxic.

In addition, understanding the impact of HFTs on price volatility is crucial to researchers, practitioners, and policy

makers, as studied in numerous research works.

3

Recent research (Brogaard et al., 2018; Kang et al., 2018; Kirilenko

et al., 2017) indicates a difference in the impact of HFTs on price volatility during normal versus stressful times. Thus, it

is critical to examine how HFTs affect price volatility, especially under stressful times, such as in the aftermath of the

2010 Flash Crash in the United States (US).

This study examines the relation between high‐frequency trading, order flow toxicity, and short‐term price volatility

during both normal and stressful times in the KOSPI 200 futures market. Toward this end, following Easley, López de

Prado, and O'Hara (2012a), we construct the Volume‐Synchronized Probability of Informed Trading (VPIN) for

measuring order flow toxicity. Easley et al. (2012a) propose VPIN as a measure of order flow toxicity, and assert that it is

a useful predictor of short‐term, toxicity‐induced volatility in the US futures markets.

4

Although the VPIN is adopted as

a successful proxy of order flow toxicity,

5

there is criticism about its application.

6

Especially, Pöppe, Moos, and

Schiereck (2016) indicate that VPIN may not be robust to the choice for a trade classification scheme between tick rule‐

based and bulk‐volume classifications (BVCs).

7

Hence, we first investigate if the VPIN metric is applicable in the KOSPI

200 futures market, which is one of the most active derivative markets in the world, and analyze which classification

algorithm, that is, the true initiator versus BVC, is better suited to capture the underlying information and is, therefore,

more suitable to calculate VPIN.

We utilize high‐quality data that encompass all the transaction records for the KOSPI 200 index futures from

January 2010 to June 2014. Our data set has the following advantages. First, it has encrypted account information in the

bid and ask side for each transaction, which allows us to classify an account as HFT or non‐HFT (nHFT) based on its

pure trading activities. Second, we further classify HFTs into foreign, individual, and institutional based on the investor

2

A typical example is the 2010 Flash Crash. On May 6, 2010, the US financial markets sufferedone of the most turbulent periods when the price of the E‐mini S&P 500 stock index futures and related

index prices collapsed and recovered in 36 min; the Dow Jones Industrial Average plunged nearly 1,000 points within several minutes and rebounded about 70% of the drop by market close. According

to the 2010 joint report issued by the Commodity Futures Trading Commission and the US Securities and Exchange Commission, the Flash Crash was triggered by large sell orders in the E‐mini S&P

500 future contracts against a backdrop of unusually high volatility and illiquidity. Kirilenko et al. (2017) show that, unlike traditional market makers, HFTs did not alter their trading strategy during

the “down”phase but tried to short accumulate contracts during the “up”phase, which was possible to further accelerate the crash.

3

Theoretically, Cartea and Penalva (2012) present a model that shows that HFTs increase price volatility. Jarrow and Protter (2012) show that HFTs can create a self‐induced mispricing that exploits

against slow traders. However, many empirical studies, including Brogaard (2010), Hasbrouck and Saar (2013), and Hagströmer and Nordén (2013), report that HFTs are helpful to reduce price

volatility in the US equity markets. However, Boehmer, Fong, and Wu (2015) reach different conclusions for international equity markets justifying that HFTs have a positive relation with price

volatility.

4

In contrast to the original PIN measure, it is straightforward to calculate and update in real time by construction, which is easily implementable for traders and regulators in a high‐frequency market

environment. Easley, López de Prado, and O’Hara (2011) demonstrate that their VPIN metric reached its all‐time historical high right before the Flash Crash and gave a warning signal for possible

market turbulence in the E‐mini S&P 500 futures market.

5

To name a few, Chordia, Hu, Subrahmanyam, and Tong (2017) propose that the volatility of order flow (VOIB), which is the similar to VPIN, measures information asymmetry and predicts stock

returns in the cross section in the US stock markets. Low, Li, and Marsh (2018) support the applicability of VPINin international equity markets. Cheung, Chou, and Lei (2015) use mandatory call

events (MCEs) of the callable bull/bear contracts (CBBC) in the Hong Kong Stock Exchange to show that a high level of VPIN indicates high market risk around MCEs. Bhattacharya and Chakrabarti

(2014) study the evolution of adverse selection in the initial public offering (IPO) aftermarket by adopting VPIN as a proxy for adverse selection.

6

For instance, Andersen and Bondarenko (2014a, 2014b) refute the findings of Easley et al. (2012a). They argue that VPIN is mechanically related to the underlying trading intensity, and its

predictability is subsumed by trading intensity and realized volatility in the E‐mini S&P 500 futures market. Abad, Massot, and Pascual (2018) show that VPIN is limited to forecast large intraday price

changes leading to single‐stock circuit breakers in the Spanish stock market.

7

The most common classification in market microstructure is the M. C. Lee and Ready (1991) algorithm. It classifies buy volume and sell volume trade‐by‐trade based on the proximity to the prevailing

quote except the midpoint. Its variations with slight changes and better performances are also introduced in subsequent research, such as Ellis, Michaely, andO'Hara (2000) and Chakrabarty, Li,

Nguyen, and Van Ness (2007). However, as pointed out by O’Hara (2015), such tick rule‐based classifications become more problematic in high‐frequency trading in the following points: (a) The

difficulty to infer the prevailing best bid and offer (BBO) due to varying latencies between the market information system and market centers and high‐order cancellation/resubmission rates, and (b)

the trading norm that traders with information do not necessarily cross the spread but use passive orders to execute trades at favorable prices with order‐splitting behaviors. Easley, de Prado, and

O'Hara (2016) find that BVC better discerns information‐based trading than tick rule in high‐frequency markets, and Easley et al. (2012a) advocate the use of BVC for calculating VPIN.

KANG ET AL.

|

165