JOT

The Voices of Influence | iijournals.com

W W W. I I J OT. C O M SUMMER2015VOLUME10NUMBER3

Behind Stock Price Movement:

Supply and Demand in Market
Microstructure and Market
Influence
JINGLE LIU AND SANGHYUN PARK Provided by
 Behind Stock Price Movement:
Supply and Demand in Market
Microstructure and Market
Influence
JINGLE LIU AND SANGHYUN PARK

T
JINGLE LIU he determination of a secu- In this article, we approach this problem
is a quantitative researcher rity’s price is mainly driven by through the study of explanatory factors for
of algorithmic trading at
the balance between supply of short-term stock price movement in the U.S.
Bloomberg Tradebook
LLC in New York, NY. and demand for that security’s equity market by exploiting the relationship
[email protected] liquidity in the marketplace. A limit order between liquidity provisions and taking
book where buy/sell orders are displayed activities at the market-microstructure level
SANGHYUN PARK ref lects the interests of market participants by quantitatively measuring central limit
is a quantitative researcher and, consequently, contains information order book imbalance and trade imbalance.
of algorithmic trading at
Bloomberg Tradebook
about market consensus. Therefore, the Furthermore, we investigate the macro-level
LLC in New York, NY. study of how market participants trade and effect of whole market movement on indi-
[email protected] quote is important to the understanding of vidual stock prices.
how a security’s price is decided. Evans and Most previous studies were based
Lyons [1999] used signed transaction data on trade and quote (TAQ) data of a single
on the foreign exchange dealer market and exchange, such as NYSE (Chordia et al.
showed that the imbalance explains the Forex [2002]), Paris Bourse (Bouchard et al. [2002];
returns quite well. Bouchard [2002] empiri- Hopman [2007]), or the Tokyo Stock
cally studied statistical properties of limit Exchange (Cont et al. [2010]). In the past
order books of three liquid stocks on the decade, global equity markets have seen a
Paris Bourse. Chordia et al. [2002] explored proliferation of electronic trading venues and
the relationship between the aggregated a consequent fragmentation of order f low.
daily order imbalance and returns. Farmer Modeling the consolidated central limit
et al. [2004] showed that large price f luctua- order book dynamics has become necessary
tions are driven by liquidity f luctuations— to capture private and public information
variations in the market’s ability to absorb of quoting and trading events in order to
new orders. Hopman [2007] found that explain the stock price return. We investigate
imbalance between uninformed buy/sell the correlation between stock price changes
pressure explains most of the stock price and the central limit order book by combing
changes on the Paris Bourse, looking at a level I data from all 12 lit exchanges in the
much longer time horizon than that of our U.S. equity market.
study. Cont et al. [2014] studied price change One aspect that previous market impact
caused by order f low imbalance, which is models rarely addressed is dark pool trading
defined as the imbalance between supply and versus lit exchange trading. Trading in dark
demand at the best bid and ask prices. pools has less direct impact on the displayed

SUMMER 2015 THE JOURNAL OF T RADING

liquidity in the order book than does trading in lit Second, we show the summary statistics of stock uni-
exchanges. Because of the different natures of those verse and model variables. Third, we show the estima-
two trading venue types and the different roles dark tion of model parameters using high-frequency time
pools play for various stocks, we separated trades in lit series of stock order book and market return; we illus-
exchanges from trades in dark pools in the model and trate the model using data from the U.S. equity market.
found that this separation can make the model better Last, we explore the explanatory power of each factor
explain the price movement. across different tickers and time intervals and analyze
Furthermore, we develop an intuitive and empir- the dependence of the model’s explanatory power on
ical framework to model the supply–demand behavior the bid–ask spread.
on a micro level by computing order book quote imbal-
ance and trade imbalance. This model also makes a MODEL
linkage between price movement of individual stocks
and that of the whole market on the macro level. Only The dynamics of stock price movement are com-
a few studies have been done on this micro-to-macro plex in nature. Multiple buy and sell orders are posted
perspective of stock price dynamics. on central limit order books with double auction pro-
We systematically studied the inf luences of these cesses going on at both the bid and ask sides. The actual
driving forces on stock price change by investigating price of a stock, typically measured by the mid quote,
how micro and macro driving forces affect stock price varies based on contemporaneous liquidity supply and
change differently for various types of stocks and various demand for the stock itself as well as the inf luence of the
time intervals. On the aspect of stock spread, our study entire market movement.
indicates that the quote imbalance has higher explana- Higher demand for and lower supply of sell (buy)
tory power for stocks with tighter spread measured by limit orders will drive up (down) the price, whereas
number of ticks compared with stocks with a wider lower demand for and higher supply of buy (sell) limit
spread. On the aspect of time interval, we also show orders will depress (move up) the price. Quantifying
that as the time interval increases from 30 seconds to 1 these effects has been the key to explaining stock price
hour, the inf luence of the trade imbalance and market changes. For a micro-level perspective of an order
movement on stock price grows higher while the quote book, the supply of liquidity could be measured by the
imbalance becomes less. These results reveal how the number of buy/sell limit orders posted in an order book;
information values of those elements vary within the the demand of liquidity could be measured by buyer-
dimensions of price and time. initiated market orders and seller-initiated market
This model not only sheds light on fundamental orders. Limit orders can only be posted in lit exchanges;
elements that drive price movement for a broad range dark pools do not have displayed liquidity. But trades
of stocks and time intervals but also suggests practical can occur in both lit and dark venues. Trades in lit
solutions for traders or brokers using trading algo- exchanges typically involve the removal of liquidity
rithms to optimize order execution performance by on the quote, while trades in dark pools do not have
intelligently controlling their footprints in the public such direct effect. Because of the different natures of
marketplace and avoiding being picked up by preda- those two types of trading venues, it is reasonable to
tory traders. In portfolio management, a more accurate treat trades in lit exchanges differently from those done
measure of expected trading cost is a vital component of in dark pools. To formulate a mathematically precise
effective portfolio strategy implementation and trading description of supply–demand imbalance, we break
performance monitoring (Almgren and Chriss [2000]; down the imbalance into quotes imbalance Q, lit trade
Engle and Ferstenberg [2007]; Kritzman et al. [2007]; imbalance LT, and dark trade imbalance DT.
Obizhaeva and Wang [2013]; Liu and Phadnis [2013]; The quote imbalance Q is defined as net limit
Alfonsi et al. [2012]). posting order f low—which is the difference between
The article is organized as follows: First, we net added buy limit orders QB and net added sell limit
describe a stylized multivariate linear model for the orders QS . For quote imbalance calculation, only limit
relationship between stock price return and stock limit orders on best bid–ask prices are being taken into consid-
order book dynamics along with the market movement. eration. QB is newly added limit orders minus cancelled

BEHIND STOCK P RICE MOVEMENT: SUPPLY AND DEMAND IN M ARKET M ICROSTRUCTURE AND M ARKET I NFLUENCE SUMMER 2015
limit orders on best bid, while QS is newly added limit STATISTICS OF MODEL VARIABLES
orders on the ask QS,+ minus cancelled limit orders on
best ask QS,−. The multivariate linear model described earlier was
applied to high-frequency order-level data for 42 U.S.
Q = QB − QS = (QB,+ − QB,− ) − (QS,+ − QS,− ) (1) tickers with diverse profiles of price, spread, liquidity
and volatility. The summary statistics of average daily
The lit trade imbalance LT is defined as volumes, average bid–ask spreads, average prices and
volatilities of the stock universe here are shown in
LT = LTB − LTS (2) Exhibit 1. As a measure of liquidity, average daily vol-
DT = DTB − DTS (3) umes range from 18,367 shares to 49,434,976 shares.
Volume-weighted average prices range from $2.647
where LT B , DT B are buyer-initiated trades in lit to $1,142.558. As a major component of trading cost,
exchanges and dark pools, respectively, and LTS , DTS bid–ask spreads of the sample universe range from 1 cent,
are seller-initiated trades in lit exchanges and dark pools, which is the smallest tick size in the U.S. market, to
respectively. The rule to classify the sign of the trade is $1.49—more than 100 times the smallest tick size. Most-
based on whether the trade is above or below the mid volatile stocks and least-volatile stocks have volatilities of
quote price at the time of trade. If trade price is above 116.48 and 9.13, respectively. The diversity of the stock
mid quote, it is classified as a buyer-initiated trade. If universe allows us to explore model dynamics in various
trade price is below mid quote, it is classified as a seller- aspects and extract insights on price movement.
initiated trade. If trade price is the same as mid quote, it We used Bloomberg-managed B-PIPE market
is not counted when calculating trade imbalance. data, which contain top-level quote updates from all
In addition to micro-level supply–demand imbal- U.S. exchanges and trades in lit/dark venues, to compute
ance in the order book, we also include the market all independent and dependent variables. We investi-
movement in the model as a macro-level factor that gated the model fitting for time intervals from 30 sec-
ref lects the inf luence of the industry-wide movement onds to 1 hour. The time period of the data set used
caused by central bank decision(s), government policy, in this study was from 4/21/2014 to 5/21/2014. Only
geopolitical events and so on. Some part of the stock continuous trading data between 9:30 and 16:00 ET
price movement that is not explainable by quote and were used to fit the model; auction prints and off-hour
trade imbalance can be attributed to this exogenous trading were excluded to remove other effects, such as
market effect. In our study, the S&P 500 Index is used company earnings releases, that are not in the scope of
to gauge the U.S. equity market movement. this study.
To explain the stock price change rs (t) (in $) for The independent and dependent variables were
given time interval t, we developed a multivariate linear computed for each time interval separately before the
model as follows: regression test was performed. For example, all the
rs (t) = β0 + β1 Q(t) + β2 LT(t)
+ β3 DT(t) + β4rm (t) (4)
EXHIBIT 1
where Q(t), LT(t), DT(t) and rm (t) are stock order book Summary Statistics of Average Daily Volumes,
quote imbalance, lit trade imbalance, dark trade imbal- Average Bid–Ask Spreads, Average Prices and
ance and market return in time interval t, respectively, Volatilities of 42 U.S. Securities
and are calculated using Equations (1)–(3). Market
movement rm is calculated as log return of the S&P 500
Index during time interval t. βi, i = 0, 1 … 4 are coeffi-
cients for the linear model. β0, β4 have units of $ and β1,
β2, β3 have units of $/shares. The reason to use a linear
model is that it is simple, intuitive, easily understood
and robust. The empirical results of applying this model
will be discussed in the following sections.

SUMMER 2015 THE JOURNAL OF T RADING

EXHIBIT 2 quote and trade data between 4/21/2014
Summary Statistics of Independent and Dependent Variables and 5/21/2014 were grouped in five-minute
for Six U.S. Stocks for Five-Minute Time Intervals intervals, such as 9:30–9:35, 9:35–9:49 and
15:55–16:00. Each full trading day has 78
intervals, with one dependent variable and
four independent variables for each interval.
To remove outliers, the intervals containing
the top 1% and bottom 1% of price changes,
quote imbalances and trade imbalances were
removed from the data sample. As an illustra-
tion of data, summary statistics of model vari-
ables for six stocks are shown in Exhibit 2.
The skewness levels of most variable dis-
tributions are between –1 and +1, which
means that variables are quite symmetrically
distributed. The fact that means of variable
distributions are small compared with the
standard deviation suggests that the center of
variable distribution is close to zero, as would
be expected in a balanced market.

ESTIMATION OF MODEL
COEFFICIENTS

The fitted model coefficients for time

intervals of five minutes for 42 tickers are
shown in Exhibit 3. Results show that the
percentage of tickers whose absolute t-sta-
tistics of β0, β1, β2 , β3 and β4 are larger than
1.97—which corresponds to the critical two-
tailed t value for 95% confidence level with
degree of freedom of around 1,700—are as
follows: 26.2%, 100.0%, 100.0%, 61.9% and
95.2%, respectively. Therefore, quote imbal-
ance, lit trade balance and market move-
ment are quite significant for most tickers
tested.
For time intervals of five minutes, the
coefficient of lit trade imbalance β2 is typically
larger than the coefficient of quote imbal-
ance β1. The explanation: market participants
consider that trades contain more information
than quote updates, that is, the same amount
of traded shares would create more market
impact than the impact created by the same
amount of quote change. For example, for
AMZN in time intervals of five minutes,
Note: Around 1,700 observations were included after outlier removal.
1,000 shares traded on the offer side would,

BEHIND STOCK P RICE MOVEMENT: SUPPLY AND DEMAND IN M ARKET M ICROSTRUCTURE AND M ARKET I NFLUENCE SUMMER 2015
EXHIBIT 3
Estimation and t-Statistics of Model Coefficients and Model Explanatory Power (R2) for 42 U.S. Tickers for
Five-Minute Time Intervals

on average, move the mid-price by 1.8e-5 × 1000 = Dark trade imbalance coefficients β3 are much
$0.018, while a 1,000-share limit order cancellation on smaller than β1 and β2. This suggests that trades in dark
the offer side would, on average, move the mid-price pools have less market impact than lit trades (which
by 9.6e-6 × 1000 = $0.0096—which is almost half of remove the posted liquidity in exchanges) and agrees
the trade impact. with the conventional notion that information leakage

SUMMER 2015 THE JOURNAL OF T RADING

EXHIBIT 4
Coefficients of Quote Imbalance (QTE_IM), Lit Trade Imbalance (L_TRD_IM) and Dark Trade Imbalance
(D_TRD_IM) on Stock Price Movements in Various Time Intervals for DNR, CB, IBM, PPG, NFLX and ICPT

BEHIND STOCK P RICE MOVEMENT: SUPPLY AND DEMAND IN M ARKET M ICROSTRUCTURE AND M ARKET I NFLUENCE SUMMER 2015
on the order can be reduced by trading in EXHIBIT 5
dark pools. Market movement coefficients The R2 of Each Model Factor for 42 U.S. Tickers in Five-Minute
β4 show how stock prices are correlated Time Intervals
with market prices. All the tickers but one
show positive coefficients, which means
their prices move together with the market
to some degree.
The model was also fitted at time
intervals from 30 seconds to 1 hour. Results
of six tickers are shown in Exhibit 4. As the
time interval gets longer, β1 and β2 gener-
ally decrease; for example, on ticker CB,
β1 and β2 decrease from 5.6e-6 $/shares
and 8.9E-6 $/shares in time intervals of 30
seconds to 3.5e-6 $/shares and 3.9e-6 $/
shares in time intervals of 1 hour, respec-
tively. This suggests that the short-term
price movement is more sensitive to the
liquidity supply–demand on (quote and
trade) imbalance in lit order books than the
long-term price movement. For most time
intervals and tickers, β2 is larger than β1.
The difference between them is larger for
shorter time intervals and for tickers with
wider spreads. On the other side, β3 tends
to be larger for time intervals of 2 minutes
to 30 minutes than for very short or very
long time intervals. β3 is smaller than β1 and
β2 for DNR, CB and IBM—whose spreads
are less than $0.5—while β3 becomes larger
than β2 for PPG, NFLX and ICPT—whose
spreads are wider than $0.5. The different
behavior between narrow-spread tickers
and wide-spread tickers will be discussed
in detail later.

EXPLANATORY POWER

The explanatory power of the model,

that is, percentage of stock price movement
variance that can be explained by the model,
is measured by the adjusted R-squared. The
attribution of explanatory power to each
factor can be readily deduced by performing variance VOD. On average, quote imbalance explains most of
analysis on the linear model. The measured model’s the stock price movement, or 38.5% of the price change.
explanatory power for 42 U.S. tickers in five-minute Next are two significant factors: lit trade imbalance and
time intervals is summarized in Exhibit 5. The total market return, which explain 8.5% and 5.9% of price
R-squared ranges from 3.3% for NATH to 78.2% for movement, respectively. The dark trade imbalance has

SUMMER 2015 THE JOURNAL OF T RADING

EXHIBIT 6
Explanatory Powers of Quote Imbalance (QTE_IM), Lit Trade Imbalance (L_TRD_IM), Dark Trade Imbalance
(D_TRD_IM) and Market Return (RET_MKT) on Stock Price Movements in Various Time Intervals for DNR,
CB, IBM, PPG, NFLX and ICPT

BEHIND STOCK P RICE MOVEMENT: SUPPLY AND DEMAND IN M ARKET M ICROSTRUCTURE AND M ARKET I NFLUENCE SUMMER 2015
the lowest explanatory power, with average R-squared a similar magnitude of trade. Whereas, when spread is
of 0.3%, which is consistent with the observed smaller more than one tick wide, trades offer more value in the
fitted values of β3. process of price discovery.
In different time intervals, the explanatory power The explanatory powers of the four factors are
of each factor as well as the whole mode change dras- measured for 42 U.S. tickers with spreads ranging from
tically. Exhibit 6 plots the explanatory powers of fac- $0.01 to $1.49; results are plotted against bid–ask spread
tors on stock price movements in various time intervals and are shown in Exhibit 7. As the bid–ask spread widens,
for DNR, CB, IBM, PPG, NFLX and ICPT. The the explanatory power of quote imbalance decreases
results show that as the time interval gets longer, the dramatically—while the explanatory power of trade
explanatory power of quote imbalance decreases for all imbalance and market movement increases moderately.
tickers except for DNR, and the explanatory power of A possible explanation is that on wider-spread stocks
lit trade imbalance, dark trade imbalance and market liquidity takers pay higher transaction costs, which
return becomes higher. This suggests that 1) informa- must then be compensated for by the greater amount
tion about the market sentiment or news is digested and of information traders have at the time of trade. Con-
ref lected in the individual stock price at a slower rate sequently, the trade would be interpreted by the public
than quote update and trade information. Market return market as containing more private information. Another
exerts greater inf luence on stock price on a long time possible reason is that to avoid the higher spread cost,
horizon. 2) In very short intervals, price movement is more trading activities occur within the spread through
dominantly inf luenced by quote add or quote cancel. hidden order type in lit exchanges or dark pools, and
And, 3) trades in lit/dark venues have more explana- those activities are captured by trade imbalance instead
tory power at longer time intervals. For example for of quote imbalance, thus quote imbalance explains less
NFLX, quote imbalance, lit trade imbalance and market of price variation. For example, ICPT has an average
move account for 39.1%, 15.9% and 4.1% of stock price spread of $1.04 and 49.5% of trades occur in dark pools,
move in 30-second intervals, respectively, while they while CB has an average spread of $0.02 and 22.1% of
account for 12.6%, 23.7% and 11.4% in 1-hour intervals, trades occur in dark pools.
respectively.
SUMMARY
BID–ASK SPREAD
In this article, we develop a multivariate linear
Bid–ask spread, one of most important funda- model to explain short-term stock price movement
mental properties of a stock, is the dominant part of from the perspectives of both micro-level order book
immediate transaction cost. Taking liquidity on the supply–demand dynamics (quote imbalance, lit trade
opposite side will cost a spread versus passive order imbalance, dark trade imbalance) and macro-level
posting on the same side. Exhibit 4 and Exhibit 6 show market inf luence (market movement). This empirical
that coefficients and explanatory powers vary with dif- and intuitive model has explanatory power of up to
ferent tickers. Average spreads of DNR, CB, IBM, PPG, around 80% for time intervals of 30 seconds to 1 hour
NFLX and ICPT are $0.010, $0.022, $0.048, $0.102, in the U.S. equity market. The inf luence of each factor
$0.315 and $1.045, respectively. The results show that varies with the type of stock as well as the length of
as the bid–ask spread widens, both the ratio of trade the time interval. The model also reveals the various
imbalance coefficient to quote imbalance coefficient and interesting mechanisms of moving stock price by
and the ratio of trade imbalance R-squared to quote classifying stocks into quote-driven stocks and trade-
imbalance R-squared become larger. This means that driven stocks. The bid–ask spread has been shown to be
narrow-spread stocks are more quote-driven in nature, the determining factor for classification. Further work
while wide-spread stocks are more trade-driven. When can be done to exploit this model for a wider range of
the spread is 1 tick or $0.01, an update on best bid and applications such as price explanation or prediction for
ask quantities has a direct impact on price pressure with other asset classes.

SUMMER 2015 THE JOURNAL OF T RADING

EXHIBIT 7
Relationships between Bid–Ask Spread and R2 Values of Quote Imbalance (QTE_IM), Lit Trade Imbalance
(L_TRD_IM), Dark Trade Imbalance (D_TRD_IM) and Market Return (RET_MKT) in Time Intervals of
300 Seconds

BEHIND STOCK P RICE MOVEMENT: SUPPLY AND DEMAND IN M ARKET M ICROSTRUCTURE AND M ARKET I NFLUENCE SUMMER 2015
ENDNOTE Engle, R., and R. Ferstenberg. “Execution Risk.” The Journal
of Portfolio Management, 33 (2007), pp. 34-44.
The authors thank Kapil Phadnis and Amber Anand for
their insightful discussions and comments on this work. Evans, M.D.D., and R. Lyons. “Order Flow and Exchange
Rate Dynamics.” NBER Working paper #7317, 1999.
REFERENCES
Farmer, J.D., L. Gillemot, F. Lillo, S. Mike, and A. Sen.
Alfonsi, A., A. Schied, and A. Slynko. “Order Book Resil- “What Really Causes Large Price Changes?” Quantitative
ience, Price Manipulation, and the Positive Portfolio Finance, 4 (2004), pp. 383-397.
Problem.” SIAM Journal on Financial Mathematics, 3 (2012),
pp. 511-533. Hopman, C. “Do Supply and Demand Drive Stock Prices?”
Quantitative Finance, 7 (2007), pp. 37-53.
Almgren, R., and N. Chriss. “Optimal Execution of Portfolio
Transactions.” Journal of Risk, 3 (2000), pp. 5-39. Kritzman, M., S. Myrgren, and S. Page. “Optimal Execution
for Portfolio Transitions.” The Journal of Portfolio Management,
Bouchard, J.-P., M. Mezard, and M. Potters. “Statistical Prop- 33 (2007), pp. 33-39.
erties of Stock Order Books: Empirical Results and Models.”
Quantitative Finance, 2 (2002), pp. 251-256. Liu, J., and K. Phadnis. “Optimal Trading Algorithm Selec-
tion and Utilization: Traders’ Consensus versus Reality.” The
Chordia, T., R. Roll, and A. Subrahmanyam. “Order Imbal- Journal of Trading, 8 (2013), pp. 9-19.
ance, Liquidity, and Market Returns.” Journal of Financial Eco-
nomics, 65 (2002), pp. 111-130. Obizhaeva, A.A., and J. Wang. “Optimal Trading Strategy
and Supply/Demand Dynamics.” Journal of Financial Markets,
Cont, R., A. Kukanov, and S. Stoikov. “The Price Impact 16 (2013), pp. 1-32.
of Order Book Events.” Journal of Financial Econometrics, 12
(2014), pp. 47-88.
To order reprints of this article, please contact Dewey Palmieri
Cont, R., S. Stoikov, and R. Talreja. “A Stochastic Model at dpalmieri@ iijournals.com or 212-224-3675.
for Order Book Dynamics.” Operations Research, 58 (2010),
pp. 549-563.

SUMMER 2015 THE JOURNAL OF T RADING