VPIN, Jump Dynamics and Inventory Announcements in Energy Futures Markets

• Date: 01 June 2017
• Author: George H. K. Wang,Hui Zheng,Johan Bjursell
• Published date: 01 June 2017
• DOI: http://doi.org/10.1002/fut.21839

VPIN, Jump Dynamics and Inventory

Announcements in Energy

Futures Markets

Johan Bjursell ,* George H. K. Wang, and Hui Zheng

The Volume-Synchronized Probability of Informed Trading (VPIN) metric is proposed by

Easley et al. (2011, 2012) (Journal of Portfolio Management, 37:118–128; Review of Financial

Studies, 25:1457–1493) as a real-time measure of order flow toxicity in an electronic trading

market. This study examines the performance of VPIN around inventory announcements and

price jumps in crude oil and natural gas futures markets with a sample period from January

2009 to May 2015. We obtain several interesting results: (i) VPIN increases significantly

around inventory announcements with price jumps as well as at jumps not associated with any

scheduled announcements. (ii) VPIN does not peak prior to the events but shortly after. (iii) A

minor variation of VPIN based on exponential smoothing significantly improves the early

warning signal property of VPIN, and this estimate of toxicity returns faster to the pre-event

level. © 2017 Wiley Periodicals, Inc. Jrl Fut Mark 37:542–577, 2017

1. INTRODUCTION

High frequency trading (HFT) accounts for a major portion of trading volume in the U.S. equity

and futures markets. In electronic limit order markets, there are no designated market markers,

and liquidity arises endogenously from the orders submitted by HFT and non-HFT market

participants. Technological advances in computation and communication allow HFT traders to

play a crucial role in liquidity supply and demand in the trading environment. For example,

Hendershott, Jones, and Menkveld (2011) present empirical evidence that algorithmic trading

improves liquidity for large stocks; and Hasbrouck and Saar (2013) analyze low-latency activity

and find that HFT improves market quality measures such as liquidity in the limit order book.

Brogaard, Hendershott, and Riordan (2014) provide evidence that HFT trading accelerates price

efficiency and the provision of liquidity at stressed times such as during the most volatile days.

This literature focuses primarily on normal market conditions.

Johan Bjursell is at Credit Suisse,Tokyo, Japan. George H. K. Wang is at Finance Area, School of Business,

George MasonUniversity, Fairfax, Virginia.Hui Zheng is a Faculty of Business and Economics,Department of

Finance,University of Sydney, Sydney, Australia.The authors wish to thank Robert I. Webb (the editor)for his

encouragementand suggestions, andto the discussants and conference participantsfor their helpful comments

and suggestionsat the 2016 Derivative Markets Conferenceat Auckland, New Zealand, August, 2016, andthe

12th Annual Conference of Asia-Pacific Association of Derivativesat Busan, Korea, August, 2016. The ideas

expressed in this article arethose of the authors and do not reflect the view of Credit Swiss or its staff.

JEL Classification: G14, G12

*Correspondence author, Credit Suisse, Tokyo, 106-6024, Japan. Tel: þ81 3 4550 7110, Fax: þ81 3 4550 9844.

e-mail: johan.bjursell@gmail.com

Received December 2016; Accepted December 2016

The Journal of Futures Markets, Vol. 37, No. 6, 542–577 (2017)

© 2017 Wiley Periodicals, Inc.

Published online 2 March 2017 in Wiley Online Library (wileyonlinelibrary.com).

DOI: 10.1002/fut.21839

It has been recognized that when HFT participants have significant exposure to large

downside market moves and if the toxicity increases, they may become liquidity consumers

rather than providers or even abandon market-making activities. This will result in illiquid

markets and induce an increase in short term price volatility.

1

Easley, de Prado, and O’Hara

(2011, 2012) present the Volume-Synchronized Probability of Informed Trading (VPIN)

metric as a real-time indicator for measuring “order flow toxicity”faced by market makers in

HFT trading environments.

2

The order flow is regarded as toxic when market makers face

strong adverse selection risk. They may be unaware of when such market conditions arise

resulting in providing liquidity at a loss. Hence, market markers’estimate of the real time-

varying toxicity level becomes a crucial factor to managing their liquidity provision. VPIN is a

timely new innovation developed to meet the demand to measure the order flow toxicity for

market makers, exchanges, and regulators. Easley et al. (2012) successfully demonstrate that

VPIN reached the highest level of order flow toxicity in E-mini futures contracts 2 hours prior

to the so-called flash crash on May 6, 2010. They also provide evidence that VPIN achieved

very high levels (the cumulative distribution function (CDF) of VPIN was equal or greater

than 0.9) on May 5, 2011, when speculators unwound their large speculative positions in

WTI crude oil futures. The unwinding of massive positions led them to seek liquidity, and as

market makers realized that the selling pressure was persistent, they started to withdraw,

which in turn increased the high level of order flow toxicity. Andersen and Bondarenko

(2014, 2015), conversely, document in their empirical investigation that VPIN is a poor

predicator of short-run volatility, and that VPIN did not reach an all-time high prior to the

flash crash on May 6 but rather following the event. They suggest that the predictive power of

VPIN is mainly due to a mechanical relationship with underlying trading intensity. In a

rejoinder, Easley, de Prado, and O’Hara (2014) point out there is confusion with the analysis

Andersen and Bondarenko (2014) carry out explaining the contradictory conclusion.

Wu, Bethel, Gu, Leinweber, and Ruebel (2013) analyze five and half years of data from

the 100 most liquid futures contractstraded worldwide in major exchanges. Their test results

confirm that VPIN is a strong predictor of liquidity-induced volatility. With selection of

parameter choices, the false positiverates are about 7% averaged over all futures contracts in

their data set. When the CDF of VPIN rises above 0.99, the volatility in the subsequenttime

windows is higher than 93% on average. Using 120 stocks in NASDAQ for 2008 and 2009,

Yildiz, Ness, and Ness (2013) document that the order flow toxicity in volume bucket t-1 is

positively relatedto the volatility in bucket teven after controlling for trade intensity variables.

Cheung,Chou, and Lei (2015) study the behavior of VPIN aroundthe mandatory call events of

callable bull/bear option contracts at the Hong Kong Option Exchange. They conclude that

high VPIN around mandatory call events indicates the existence of large volume imbalances.

In short, there is an ongoing debate on the predictive power of VPIN and future

liquidity-induced short-run volatility.

3

In general, most previous literature assesses the

usefulness of VPIN as a signal for order flow toxicity at selected single trading day events such

as the May 6, 2010, market flash crash.

4

Observers of energy futures markets have long noted that energy futures prices are very

volatile and often exhibit jumps (price spikes) at inventory news releases. The theory of

storage (see Brennan (1958), Kaldor (1939), Telser (1958), Working (1948, 1949) and

others) demonstrates that the level of inventory is one of the major factors determining spot

1

The market flash crash on May 6, 2010, is an example (see Kirilenko et al. (2015) and Easley et al. (2012)).

2

The theoretical development on concepts from PIN (probability of informed trading) to intraday VPIN and the

computational advantage of VPIN over PIN in the high frequency world may be found in Easley et al. (2012) and

Abad and Yague (2012).

3

For other empirical works related to using VPIN refer to Wei, Gerace, and Frino (2013) and others.

4

See Easley et al. (2012).

VPIN, Jump Dynamics, and Inventory Announcements 543

and futures prices of consumption-based commodities.

5

Volatility behavior of energy futures

prices has been investigated by Mu (2007), Chan, Wang, and Yang, (2010), and others. Mu

(2007) finds that extreme weather conditions and low inventories are important factors

affecting natural gas futures volatility within a single equation model with a GARCH error

process. Chan et al. (2010) study the common jump dynamics in natural gas futures and spot

markets within a bivariate autoregressive jump intensity GARCH framework. They

particularly examine the role of weather as a short-run demand factor and inventory as a

short-run supply factor in explaining price spikes and time varying volatility in spot and

futures returns.

Previous studies examining price behavior and volatility surrounding inventory

announcements of energy stocks include Linn and Zhu (2004), Gay, Simkins, and Turac

(2009), Bjursell, Gentle, and Wang (2015). Linn and Zhu (2004) report an increase in

volatility before and after the release of inventory reports by the Energy Information

Administration. Gay et al. (2009) demonstrate that a one percent unexpected increase in

natural gas inventory results in approximately 1% drop in the natural gas price. Furthermore,

they provide evidence that prices react most strongly to forecasts of analysts with better prior

forecast accuracy. Bjursell et al. (2015) apply nonparametric methods to identify jumps in

futures prices and intraday jumps surrounding inventory announcements of crude oil,

heating oil, and natural gas contracts traded on the New York Mercantile Exchange with

sample period of the intraday data covering January 1990 to January 2008. They obtain

several interesting empirical results. (i) Large volatility days are often associated with large

jump components, and large jump components are often associated with the Energy

Information Administration’s inventory announcement dates and other important news

related to energy markets. (ii) The volatility jump component is less persistent than the

continuous sample path component. (iii) Volatility and trading volume are higher on days

with a jump at the inventory announcement than on days without a jump at the

announcement. Furthermore, the intraday volatility returns to normal faster following

inventory announcements with jumps than after announcements without jumps. Based on

previous results, we can expect that the order flow becomes more toxic due to high volatility

and trading volume during inventory announcement periods. Therefore, we have an ideal

empirical test setting for examining the performance of VPIN as a real-time indicator of order

flow toxicity and early warning indicator for market turbulence around repetitive scheduled

information and liquidity events.

The major purposes of this study are twofold: (i) we examine the behavior of VPIN

around inventory announcements with price jumps (scheduled events) and price jumps not

associated with scheduled events, in crude oil, and natural gas futures markets during a

recent sample period spanning from January 1, 2009 to May 31, 2015; and (ii) we propose a

minor variation of VPIN by applying exponential smoothing in the last stage of the

calculation. We believe this variation will increase the sensitivity of VPIN to capture recent

order flow toxicity. We obtain several interesting results:

1. VPIN estimates increase significantly around the inventory announcements period with price

jumps (scheduled events) as well as at jumps not associated with any inventory announcements

(unscheduled events).

6

5

Crude oil and natural gas are classified as consumption-based commodities. Furthermore, we should mention that

convenience yield has an inverse relationship with level of inventory. (See Hull (2015) for a discussion of the role of

convenience yields in pricing consumption-based commodities.)

6

Unscheduled events refer to jumps which cannot be associated with any event listed in Bloomberg’s economic

calendar.

544 Bjursell, Wang, and Zheng