Simulating and analyzing order book data:

The queue-reactive model

Weibing Huang1,2 , Charles-Albert Lehalle3 and Mathieu Rosenbaum1

arXiv:1312.0563v2 [q-fin.TR] 3 Sep 2014

LPMA, University Pierre et Marie Curie (Paris 6)

2
Kepler-Cheuvreux
3
Capital Fund Management
September 5, 2014
Abstract

Through the analysis of a dataset of ultra high frequency order book updates, we introduce
a model which accommodates the empirical properties of the full order book together with
the stylized facts of lower frequency financial data. To do so, we split the time interval of
interest into periods in which a well chosen reference price, typically the midprice, remains
constant. Within these periods, we view the limit order book as a Markov queuing system.
Indeed, we assume that the intensities of the order flows only depend on the current state
of the order book. We establish the limiting behavior of this model and estimate its
parameters from market data. Then, in order to design a relevant model for the whole
period of interest, we use a stochastic mechanism that allows to switch from one period of
constant reference price to another. Beyond enabling to reproduce accurately the behavior
of market data, we show that our framework can be very useful for practitioners, notably
as a market simulator or as a tool for the transaction cost analysis of complex trading
algorithms.

Keywords: Limit order book, market microstructure, high frequency data, queuing model,
jump Markov process, ergodic properties, volatility, mechanical volatility, market simulator,
execution probability, transaction costs analysis, market impact.

Introduction

Electronic limit order books (LOB for short), where market participants send their buy and
sell orders via a continuous-time double auction system, are nowadays the dominant mode of
exchange on financial markets. Consequently, understanding the LOB dynamics has become a
fundamental issue. Indeed, a deep knowledge of the LOBs behavior enables policy makers to
design relevant regulations, market makers to provide liquidity at cheaper prices, and investors
to save transaction costs while mounting and unwinding their positions, thus reducing the cost
of capital of listed companies. Furthermore, it can also provide insights on the macroscopic
features of the price which emerges from the LOB.
1

In the seminal work on zero intelligence LOB models of Smith, Farmer, Gillemot, and Krishnamurthy (2003), a mean-field approach is suggested in order to study the properties of the LOB.
In such models, the underlying assumption is that the order flows follow independent Poisson
processes. Although this hypothesis is not really compatible with empirical observations, the
authors show that its simplicity allows for the derivation of many interesting formulas, some of
them being testable on market data. This work has been followed by numerous developments.
For example, in Cont, Stoikov, and Talreja (2010), the probabilities of various order book related events are computed in this framework, whereas stability conditions of the system are
studied in Abergel and Jedidi (2011). We wish to extend this approach in two directions. On
the one hand, we want our model to be more consistent with market data, so that we can give
new insights on the dynamics of the LOB. On the other hand, we aim at providing a useful and
relevant tool for market practitioners, notably in the perspective of transaction costs analysis.
Under the first in first out rule (which we assume in the sequel), a LOB can be considered as
a high-dimensional queuing system, where orders arrive and depart randomly. We consider the
three following types of orders:
Limit orders: insertion of a new order in the LOB (a buy order at a lower price than the
best ask price, or a sell order at a higher price than the best bid price).
Cancellation orders: cancellation of an already existing order in the LOB.
Market orders: consumption of available liquidity (a buy or sell order at the best available
price).
In practice, market participants (or their algorithms) analyze many quantities before sending
a given order at a given level. One of the most important variables in this decision process is
probably the distance between their target price and their reference market price, typically
the midprice. This reference price is linked with the order flows since it is usually determined
by the LOB state. This interconnection makes the design of LOB models quite intricate. To
overcome this difficulty, we split the time interval of interest into periods of constant reference
price, and consider two parts in our modeling. First, we study the LOB as a Markov queuing
system during the time periods when the reference price is constant. Then, we investigate the
dynamics of the reference price. Such a framework is particularly suitable for large tick assets1 ,
for which constant reference price periods are quite long and allow for accurate parameter estimations.
Two kinds of public information are available to market participants at the high frequency
scale: the historical order flows and the current state of the LOB. In this paper, we are mostly
interested in how the state of the LOB impacts market participants decisions. Surprisingly
enough, this question has been rarely considered in the literature. Let us mention as an exception the interesting approach in Gareche, Disdier, Kockelkoren, and Bouchaud (2013), where
the impact of the LOB state on the queue dynamics is analyzed through PDE type arguments.
Within periods of constant reference price, we model the LOB as a continuous-time Markov
jump process, and estimate its infinitesimal generator matrix under various assumptions on the
1
A large tick asset is defined as an asset whose bid-ask spread is almost always equal to one tick, see Dayri
and Rosenbaum (2012). In practice, our framework can be considered relevant for any asset whose average
spread is smaller than 2.5 ticks.

information set used by market participants. From these results, we are able to analyze how
market participants react towards different configurations of the LOB. Furthermore, we provide
the asymptotic distributions of the LOB. The level of realism of our approaches is assessed by
comparing expected features from the models with empirical ones. Thus, all our developments
are illustrated on two specific examples of large tick stocks on Euronext Paris: France Telecom
and Alcatel-Lucent (in appendix).
In the second part of the paper, we extend our framework by allowing reference price moves,
so that our model also accommodates macroscopic properties of the asset (roughly summarized
by the volatility). Modifications of the reference price2 will possibly occur provided one of the
best queues is totally depleted or a new order is inserted within the spread. This model is
called queue-reactive model. In particular, it enables us to bring to light a quantity, the
maximal mechanical volatility, which represents the amount of price volatility generated by
the generic randomness of the order flows. In practice, this parameter is typically smaller than
the empirical volatility estimated from market data. The reason for this is simple: the market
does not evolve like a closed physical system, where the only source of randomness would be the
endogenous interactions between participants. It is also subject to external informations, such
as the news, which increase the volatility of the price. Hence, it will be necessary to introduce
an exogenous component within the queue-reactive model.
Throughout the paper, we illustrate the fact that many useful short term predictions can be
computed in our framework: execution probabilities of passive orders, probability of price increase. . . More importantly, we show that the queue-reactive model turns out to be a very
relevant market simulator, notably in view of the analysis of complex trading tactics, using for
example a mixture of market and limit orders.
The paper is organized as follows. In Section 2, we consider periods when the reference price is
constant. We first present a very general framework for the LOB dynamics and then introduce
three specific models. The first one is a birth and death process in which the queues are
assumed to be independent. In this setting, we are able to fully characterize the asymptotic
behavior of the LOB. The second approach is a queuing system in which the bid and ask sides
are independent, but the first two lines on each side can exhibit correlations. We show that
this model can be seen as a quasi birth and death process (QBD for short) and thus admits a
matrix geometric solution as its invariant distribution. In the last approach, we allow for cross
dependences between bid and ask queues. An application of these models to the computation of
execution probabilities is presented at the end of the same section. In Section 3, we investigate
the dynamics of the reference price. In particular, we build the queue-reactive model which is a
relevant LOB model for the whole time period of interest. We end this section by showing how
our framework can be used for transaction costs and market impact analysis of high frequency
trading strategies. A conclusion and some perspectives are given in Section 4. Some proofs and
further empirical results are gathered in an appendix.

Note that the reference price will not be exactly the midprice, see Section 2.2.

Dynamics of the LOB in a period of constant reference

price

Within time periods when the reference price is constant, we consider three different models for
the LOB. These models can be jointly introduced through the general framework we present
now.

2.1

General Framework

In the general framework, the LOB is seen as a 2K-dimensional vector, where K denotes
the number of available limits on each side3 , see Figure 1. The reference price pref defines
the center of the 2K-dimensional vector, and divides the LOB into two parts: the bid side
[Qi : i = 1, ..., K] and the ask side [Qi : i = 1, ..., K], where Qi 4 represents the limit at the
distance i 0.5 ticks to the right (+i) or to the left (i) of pref . The number of orders at Qi
is denoted by qi . We assume that on the bid (resp. ask) side, market participants send buy
(resp. sell) limit orders, cancel existing buy (resp. sell) orders and send sell (resp. buy) market
orders. We consider a constant order size at each limit. However, the order sizes at the different
limits are allowed to be different. In practice, these sizes can be chosen as the average event
sizes observed at each limit Qi (AESi for short)5 .
The 2K-dimensional process X(t) = (qK (t), ..., q1 (t), q1 (t), ..., qK (t)) is then modeled as a
continuous-time Markov jump process in the countable state space = N2K , with jump size
equal to one. For q = (qK , ..., q1 , q1 , ..., qK ) , and ei = (aK , ..., ai , ..., aK ), where aj = 0
for j 6= i and ai = 1, the components Qq,p of the infinitesimal generator matrix Q of the process
X(t) are assumed to be of the following form:
Qq,q+ei = fi (q)
Qq,qei = gi (q)
X
Qq,q =
Qq,p
q,p6=q

Qq,p = 0, otherwise.
We now give a theoretical result on the ergodicity of the system under two very general assumptions. Let us denote by Pq,p (t) the transition probability from state q to state p in a time
t. Recall that a Markov process in a countable state space is said to be ergodic if there exists
a probability measure that satisfies P = ( is called invariant measure) and for every q
and p:
lim Pq,p (t) = p .

We consider the two following assumptions.

Note that an empty limit can be part of the LOB in our setting.
To simplify our notations, we write i /i as i , and i /i as i .
5
In our framework, AESi is a more suitable choice than ATS (Average Trade Size) that computes only the
average size of market orders, see Section 5.5 in appendix for more details.
4

Figure 1: Limit order book

Assumption 1. (Negative individual drift) There exist a positive integer Cbound and > 0,
such that for all i and all q , if qi > Cbound ,
fi (q) gi (q) < .
Assumption 2. (Bound on the incoming flow) There exists a positive number H such that for
any q ,
X
fi (q) H.
i[K,...,1,1,...,K]

Assumption 1 can be interpreted as follows: the queue size of a limit tends to decrease when
it becomes too large. Assumption 2 ensures no explosion in the system: the order arrival
speed stays bounded for any given state of the LOB. Under these two assumptions, we have
the following ergodicity result for the 2K-dimensional queuing system. The proof is given in
appendix.
Theorem 2.1. Under Assumptions 1 and 2, the 2K-dimensional Markov jump process X(t) is
ergodic.
This theorem is the basis for the asymptotic study of the LOB dynamics in the following
sections.

2.2
2.2.1

Data description and estimation of the reference price

The database

The data used in our empirical studies are collected from Cheuvreuxs6 LOB database, from
January 2010 to March 2012, on Euronext Paris. It records the LOB data (prices, volume and
number of orders) up to the fifth best limit on both sides, whenever the LOB state changes.
Note that we remove market data corresponding to the first and last hour of trading, as these
6
Cheuvreux is a brokerage firm based in Paris, formerly a subsidiary of Credit Agricole Corporate Investment
Bank, and now merged with Kepler Capital Market.

periods have usually specific features because of the opening/closing auction phases. Two large
tick European stocks, France Telecom and Alcatel-Lucent, are studied and they exhibit very
similar behaviors. Some characteristics of these two stocks are given in Table 1. We have chosen
the stock France Telecom as illustration example for all the developments in the sequel. The
results for Alcatel-Lucent can be found in appendix. Although only stocks are considered in
this paper, our method applies also to other financial assets, such as interest rates or index
futures (among which large tick assets are quite numerous, see Dayri and Rosenbaum (2012)).
stock
France Telecom
Alcatel Lucent

average number of
orders per day
159250
129400

average number of
trades per day
7282
8626

average spread size

(in number of ticks)
1.43
1.99

Table 1: Data description

2.2.2

Estimation of the reference price

As mentioned in the introduction, the estimation of a relevant reference price pref is the basis
for defining the limits in the order book. Indeed, pref provides the center point of the LOB and
thus the positions of the 2K limits. In our framework, if we write pi for the price level of the
limit Qi , i = K, ..., 1, 1, ..., K, we must have
pref =

p1 + p1
.
2

When the observed bid-ask spread is equal to one tick, pref is obviously taken as the midprice
(denoted by pmid ) and both Q1 and Q1 are non empty. When it is larger than one tick, several
choices are possible for pref . We build pref from the data the following way: when the spread
is odd (in tick unit), it is still natural to use pmid as the LOB center:
pref = pmid =

(pbestbid + pbestask )
.
2

When it is even, pmid is no longer appropriate since it is now itself a possible position for order
arrivals. In such case, we use either
pmid +

tick size
tick size
or pmid
,
2
2

choosing the one which is the closest to the previous value of pref . Note that more complex
methods could be used for the estimation of pref , see for example Delattre, Robert, and Rosenbaum (2013).

2.3

Model I: Collection of independent queues

We now give a first simple LOB model around a fixed reference price.

2.3.1

Description of the model

In this model, we assume independence between the flows arriving at different limits in the
LOB. Three types of orders are considered: limit orders, cancellations and market orders. We
suppose that the intensities of these point processes at different limits are only functions of the
target queue size (that is the available volume at the considered limit Qi ). Furthermore, at a
given limit, conditional on the LOB state, the arrival processes of the three types of orders are
taken independent. The values of these intensities are denoted by Li (n) (limit orders), C
i (n)
M
(cancellations) and i (n) (market orders) when qi = n. Moreover, the intensity functions at
Qi and Qi are chosen identical, considering the symmetry property of the LOB. We then have
M
M
C
Li (n) = Li (n), C
i (n) = i (n), i (n) = i (n), and
fi (q) = Li (qi )
M
gi (q) = C
i (qi ) + i (qi ).
In this model, market orders sent to Qi consume directly the volume available at Qi . Therefore,
we can have a market order at the second limit while the first limit is not empty. However,
for large tick assets, this assumption is reasonable as their market order flow is almost fully
concentrated on first limits (Q1 ) and the estimated intensities of this flow at (Qi ), i 6= 1 are
very small. Under these assumptions, the LOB becomes a collection of 2K independent queues,
each of them being a birth and death process.
2.3.2

Empirical study: Collection of independent queues

In Model I, the intensities of the different queues can be estimated separately. The value of K is
set to 3, as our numerical experiments show that for the considered stocks, both the dynamics
and empirical distributions at Qi , i = 4, 5 are quite similar to that at Q3 . This value of K
will also apply to other experiments in the paper.
The estimation method goes as follows. We define an event as any modification of the
queue size. For queue Qi , we record the waiting time ti () (in number of seconds) between
the event and the preceding event at Qi , the type of the event Ti () and the queue size qi ()
before the event. The queue size is then approximated by the smallest integer that is larger
than or equal to the volume available at the queue, divided by the stocks average event size
AESi at the corresponding queue. We set the type of the event the following way:
Ti () E + for limit order insertion at Qi ,
Ti () E for limit order cancellation at Qi ,
Ti () E t for market order at Qi .
When the reference price changes, we restart the recording process. Once we have collected
M
(ti (), Ti (), qi ()) from historical data, it is easy to estimate Li (n), C
i (n) and i (n) by the
maximum likelihood method:

Limit order cancellation intensity, Model I

0.9

2.5
2
1.5
1
0.5

0.7
0.6
0.5
0.4
0.3
0.2

First limit
Second limit
Third limit

0.1

0
10
20
30
40
50
Queue Size (per average event size)

0
10
20
30
40
50
Queue Size (per average event size)

Intensity (num per second)

0.8
Intensity (num per second)

Intensity (num per second)

Limit order insertion intensity, Model I

3.5
First limit
Second limit
3
Third limit

Market order arrival intensity, Model I

0.3
First limit
Second limit
0.25
Third limit
0.2
0.15
0.1
0.05
0
-0.05

0
10
20
30
40
50
Queue Size (per average event size)

Figure 2: Intensities at Qi , i = 1, 2, 3, France Telecom

1
mean(ti ()|qi () = n)
#{Ti () E + , qi () = n}
L (n) =

(n)
i
i
#{qi () = n}

C (n) =
i (n) #{Ti () E , qi () = n}

i
#{qi () = n}
t
M (n) =
i (n) #{Ti () E , qi () = n} ,

i
#{qi () = n}
i (n) =

where mean denotes the empirical mean and #A the cardinality of the set A.
In Figure 2, we present the estimated intensities. Data at Qi and Qi are aggregated together
(simply by combining the two collected samples) and confidence intervals (dotted lines) are
computed using central limit approximations detailed in appendix. We now comment the
obtained graphs.
Behaviors under the independence assumption
Limit order insertion:
Q1 : The intensity of the limit order insertion process is approximately a constant function
of the queue size, with a significantly smaller value at 0. Note that inserting a limit
order in an empty queue creates a new best limit and the market participant placing
this order is the only one standing at this price level. Such action is often risky.
Indeed, when the spread is different from one tick, one is quite uncertain about the
position of the so-called efficient or fair price, see for example Delattre, Robert,
and Rosenbaum (2013) for discussions on this notion. This smaller value can also be
due to temporary realizations of the structural relation between the bid-ask spread
and the volatility: if the spread is large because the inventory risk of market makers
is high, the probability that anyone inserts a limit order in the spread is likely to
be low, see among others Madhavan, Richardson, and Roomans (1997), Avellaneda
8

and Stoikov (2008), Wyart, Bouchaud, Kockelkoren, Potters, and Vettorazzo (2008)
and Dayri and Rosenbaum (2012) for more details about market making and the
relation between spread, volatility and inventory risk.
Q2 : The intensity is now approximately a decreasing function of the queue size. This
interesting result probably reveals a quite common strategy used in practice: posting
orders at the second limit when the corresponding queue size is small to seize priority.
More details on this strategy are given in Section 2.4.2.
Q3 : The intensity function shows similar properties to that at the second limit.
Limit order cancellation:
Q1 : The rate of order cancellation is an increasing concave function for q1 between 0 and
25, and becomes flat/slightly decreasing for larger values. This result is in contrast
to the classical way to model this flow, where one often considers a linearly increasing
cancellation rate, see for example Cont, Stoikov, and Talreja (2010). On this first in
first out market, the priority value, that is the advantage of a limit order compared
with another limit order standing at the rear of the same queue, can be one of the
reasons for this behavior. Indeed, the priority value is an increasing function of the
queue size and orders having a high priority value are less likely to be canceled.
Q2 : The rate of order cancellation attains more rapidly its asymptotic value, which is
lower than for Q1 . Compared to the first limit case, market participants at the
second limit have even stronger intention not to cancel their orders when the queue
size increases. This is probably due to the fact that these orders are less exposed
to short term market trends than those posted at Q1 (since they are covered by
the volume standing at Q1 and their price level is farther away from the reference
price).
Q3 : The priority value is smaller at the third limit since it takes longer time for Q3 to
become the best quote if it does. The rate of order cancellation increases almost
linearly for queue sizes larger than 3 AES3 . We also find a quite large cancellation
rate when the queue size is equal to one, which shows that market participants cancel
their orders more quickly when they find themselves alone in the queue.
Market orders:
Q1 : The rate decreases exponentially with the available volume at Q1 . This phenomena
is easily explained by market participants rushing for liquidity when liquidity is
rare, and waiting for better price when liquidity is abundant.
Q2 : In practice, market orders can arrive at Q2 only if Q1 = 0 (that is when Q2 is
the best offer queue). The shape of the intensity is very similar to the one obtained
in the case of Q1 . The values are of course much smaller.
Q3 : In some rare cases, one can still find some market orders arriving at Q3 (market
orders occurring when the spread is large). The intensity function remains exponentially decreasing.

2.3.3

Asymptotic behavior under Model I

The invariant distribution of the LOB can be computed explicitly in Model I. We denote by i
the stationary distribution of the limit Qi , and define the arrival/departure ratio vector i by
i (n) =

Li (n)
.
M
(C
i (n + 1) + i (n + 1))

Then the following result for the invariant distribution is easily obtained, see for example Gross
and Harris (1998):

i (n) = i (0)
i (0) =

1+

n
Y

i (j 1)

j=1
Y
n
X

i (j 1)

1

n=1 j=1

Hence the long term behavior of the LOB is completely determined by . This implies that two
assets can have very different flow dynamics, but still the same invariant distribution provided
that their arrival/departure ratios are the same.
We now compare the asymptotic results of the model with the empirical distributions observed
at Q1 , Q2 and Q3 . To compute these empirical laws, we use a sampling frequency of 30
seconds (every 30 seconds, we look at the LOB and record its state)7 . The results are gathered
in Figure 3, as well as the invariant distributions from a Poisson model (constant limit/market
order arrival rate, linear cancellation rate, parameters estimated from the same dataset).
One can see that the invariant distributions approximate very well the empirical distributions
of the LOB. This shows that in order to explain the shape of the LOB, such mean-field type
approach, where the LOB profile arises from interactions between the average behaviors of
market participants, can be very relevant.

2.4

Model II: Dependent case

We now present some extensions of Model I. We assume here that buy/sell market orders
consume volume at the best quote limits, defined as the first non empty ask/bid queue. Thus,
we consider a buy market order process with intensity M
buy and a sell market order process with
M
intensity sell . The limit order, cancellation, and market order arrival processes are assumed to
be independent conditional on the LOB state. So we can write fi (q) and gi (q) in the following
form:
fi (q) = Li (q)
M
gi (q) = C
i (q) + buy (q)1bestask(q)=i , if i > 0
M
gi (q) = C
i (q) + sell (q)1bestbid(q)=i , if i < 0.
7
Other sampling frequencies have also been tested and the estimated distributions are found to be very
similar. These sampled data will also be used to estimate the joint distributions of the LOB limits in Model IIa
and IIb .

10

First limit

Second limit

0.2
Empirical estimation
Model I
Poisson model

0.18
0.16

0.2
Empirical estimation
Model I
Poisson model

0.18
0.16

0.18

0.1
0.08

0.14
Distribution

0.12

0.12
0.1
0.08

0.12
0.1
0.08

0.06

0.06

0.06

0.04

0.04

0.04

0.02

0.02

0.02

10
20
30
40
Queue Size (in AES)

50

Empirical estimation
Model I
Poisson model

0.16

0.14
Distribution

0.14
Distribution

Third limit

0.2

10
20
30
40
Queue Size (in AES)

50

10
20
30
40
Queue Size (in AES)

50

Figure 3: Model I, invariant distributions of q1 , q2 , q3 , France Telecom

As for Model I, we consider some bid-ask symmetry, that is, for q = [q3 , q2 , q1 , q1 , q2 , q3 ], q 0 =
0
M
M
0
M
[q3 , q2 , q1 , q1 , q2 , q3 ] and i = 1, 2, 3, Li (q) = Li (q 0 ), M
i (q) = i (q ) and buy (q) = sell (q ) .
2.4.1

Model IIa : Two sets of dependent queues

Institutional traders and brokers tend to place most of their limit orders at best limits, while
many market makers, arbitragers and other high frequency traders stand also in queues beyond
these best limits. This suggests for example that the dynamics at Q2 may not only depend on
q2 , but also on whether or not Q1 is empty. We thus propose to use the following intensity
functions for the queue Q2 : in this model, L2 and C
2 are functions of q2 and 1q1 >0 .
Intensities at Qi , i 6= 2 remain functions of qi only. For large tick assets, the probability that
Qi , i 3 is the best limit is negligible. It is thus reasonable to also assume that market orders
are only sent to Q1 and Q2 . This enables us to keep the independence property between Q3
and (Q1 , Q2 ). When q1 > 0, the market order intensity M
buy/sell is assumed to be a function
of q1 ; when q1 = 0, it is a function of q2 only.
2.4.2

Model IIa : Empirical study

In this empirical study, our goal is to understand how market participants make trading decisions at Q2 in two different situations: q1 = 0 and q1 > 0. Since we are now studying
a two-dimensional problem, the data recording process is slightly different. In particular, for
(Q1 , Q2 ), it goes as follows: we record the waiting times ti () between events that happen at
Q1 or Q2 , the type of event T () and the two queue sizes (q1 (), q2 ()) before the event. The
M
maximum likelihood method is again used to estimate the intensity functions Li , C
i , i for
i = 1, 2. For i = 1 and i = 3, as the dynamics at Qi only depend on the queue size at Qi ,
M
the estimated values of L1 , C
1 and 1 are very close to those obtained in Model I and are not
shown here. The estimated intensity functions at Q2 are given in Figure 4. Some comments
are in order:
Limit order insertion: Both curves are decreasing functions of the queue size. In the
first case (q1 = 0), the limit order insertion intensity reaches very rapidly its asymptotic
value. The relatively high value observed for q2 = 0 is probably due to the fact that
for large tick assets, market makers rarely allow for spreads larger than 3 ticks. In the
11

Second limit, limit order cancellation

2
q1 == 0
1.8
q1 > 0

q1 == 0
q1 > 0

1.6
1.4
1.2
1
0.8
0.6
0.4

Intensity (num per second)

Intensity (num per second)

1.8

1.6
1.4
1.2
1
0.8
0.6
0.4

0.2

0.2

0
10
20
30
40
Queue Size (per average event size)

Second limit, market order insertion

0.06
q1 == 0
q1 > 0
0.05
Intensity (num per second)

Second limit, limit order insertion

2

0.04

0.03

0.02

0.01

0
10
20
30
40
Queue Size (per average event size)

0
10
20
30
40
Queue Size (per average event size)

Figure 4: Intensities at Q2 as functions of 1q1 >0 and q2 , France Telecom

second case (q1 > 0), the intensity continues to go down to a much lower value. This
is likely to be related to the arbitrage strategy introduced in Section 2.3.2: post passive
orders at a non-best limit when its size is small, wait for this limit to eventually become
the best limit and then gain the profit from having the priority value. For example, when
the considered limit becomes the best one, one can decide to stay in the queue if its size
is large enough to cover the risk of short term market trend, or to cancel the orders if the
queue size is too small.
Limit order cancellation: The cancellation rate is higher when q1 = 0. This can be
related to the concentration of the trading activity at best limits. When q1 > 0, the
cancellation rate is quite large when q2 = 1, as it is the case at Q3 (see Section 2.3.2).
Market orders: No market order can arrive at Q2 when there are still limit orders at
Q1 (cross limits large market orders that consume several limits are treated as several
market orders that arrive sequentially at those limits within a very short time period).
The market order arrival rate when Q2 is the best limit is not very different from that at
Q1 , but shows a rather unexpected increasing trend when the queue size becomes larger
than 5 AES2 .
2.4.3

Model IIa : Asymptotic behavior

Model IIa belongs to a special class of Markov processes, called quasi birth and death processes
(QBD). Their asymptotic behavior can be studied by the matrix geometric method. Definitions of QBD processes and explanations about the matrix geometric method can be found in
appendix. In Figure 5, we show the theoretical joint distribution of (q1 , q2 ) for the stock France
Telecom and compare it with the joint distribution estimated from empirical data. Here also,
we see that the theoretical results provide a very satisfying approximation.
12

Figure 5: Model IIa : joint distribution of q1 , q2 , France Telecom

2.4.4

Model IIb : Modeling bid-ask dependences

We now study the interactions between the bid queues and the ask queues. Let Q0 , Q , Q,
+
Q be four marks which represent in the following ranges of values for the queue sizes. Let m
and l be two integers. We define the function Sm,l (x):
Sm,l (x)
Sm,l (x)
Sm,l (x)
Sm,l (x)

=
=
=
=

Q0 if x = 0
Q if 0 < x m
if m < x l
Q
Q+ if x > l.

This function associates to a queue size x four possible ranges: empty: x = 0, small: x (0, m],
usual: x (m, l] and large: x (l, +). We set m as the 33% lower quantile and l as the 33%
upper quantile of q1 (conditional on positive values). In this model, market participants at
Q1 adjust their behavior not only according to the target queue size, but also to the size of the
opposite queue. The rates L1 and C
1 are therefore modeled as functions of q1 and Sm,l (q1 ).
a
As in Model II , we suppose that market orders consume volume at the best limits and are
only sent to Q1 and Q2 . When q1 > 0, the market order intensity M
buy/sell is assumed to
be a function of q1 and Sm,l (q1 ). Regime switching at Q2 is kept in this model: L2 , C
2
are assumed to be functions of 1q1 >0 and q2 , and when q1 = 0, the market order intensity
M
buy/sell is modeled as a function of q2 .
Under these assumptions, the 2K-dimensional problem is reduced to the study of the 4dimensional continuous-time Markov jump process (Q2 , Q1 , Q1 , Q2 ). One important feature
of this model is that the queues Q2 have no influence on the dynamics at Q1 . Therefore, we
only need to study the 3-dimensional process (Q1 , Q1 , Q2 ) (or even the 2-dimensional process
(Q1 , Q1 ) if one is only interested in the dynamics at Q1 . Remark also that other choices for
the specification of the intensity functions at Q1 are possible. For example, one can consider
q1
, or simply as functions
them as functions of the first level bid/ask imbalance, defined as qq11 +q
1
of the spread size.
13

1
0.8
0.6
0.4

First limit, limit order cancellation

1.4
1.2
Intensity (num per second)

Intensity (num per second)

1.2

q(-1) = 0
0<q(-1)<=4
4 < q(-1) <= 9
q(-1) > 9

q(-1) = 0
0<q(-1)<=4
4 < q(-1) <= 9
q(-1) > 9

1
0.8
0.6
0.4

First limit, market order insertion

0.3

0.25
Intensity (num per second)

First limit, limit order insertion

1.4

0.2

0.15

0.1

0.2

0.2

0.05

0
10
20
30
40
Queue Size (per average event size)

0
10
20
30
40
Queue Size (per average event size)

q(-1) = 0
0<q(-1)<=4
4 < q(-1) <= 9
q(-1) > 9

0
10
20
30
40
Queue Size (per average event size)

Figure 6: Intensities at Q1 as functions of Sm,l (q1 ) and q1 , France Telecom

2.4.5

Model IIb : Empirical study

We focus here on the estimation of the intensity functions at Q1 . We consider the deparM
ture flow intensities C
1 (q1 , Sm,l (q1 )) and buy/sell (q1 , Sm,l (q1 )), and the arrival flow intensities L1 (q1 , Sm,l (q1 )). Using again the symmetry property of the LOB, we take L1 (x, y) =
C
M
M
L1 (x, y), C
1 (x, y) = 1 (x, y) and sell (x, y) = buy (x, y). We record the waiting times t()
between events that happen at Q1 or Q1 , the types of event T () and the two queue sizes
(q1 (), q1 ()) before the event. Then we estimate these intensity functions using the maximum likelihood method. The results are shown in Figure 6 (m = 4 AES1 , l = 9 AES1 )8 . Some
remarks are in order:
Limit order insertion: The limit order insertion rate is a decreasing function of the opposite
queue size. In particular, we see that when the opposite queue is empty (pink curve),
it is significantly larger. Indeed, in that case, the efficient price is likely to be closer
to the opposite side. Therefore limit orders at the non empty first limit are likely to be
profitable.
Limit order cancellation: The cancellation rates for different ranges of Q1 are similar in
their forms but have different asymptotic values. This rate is not surprisingly a decreasing
function of the liquidity level on the opposite side. Indeed, when this level becomes low,
many market participants cancel their limit orders and send market orders since the
market is likely to move in an unfavorable direction.
Market orders: We see that when the liquidity available on the opposite side is abundant,
more market orders are sent. Indeed, in that case, transactions at the target queue are
relatively cheap as its price level is temporarily closer to the efficient price. In the special
situation q1 = 0, the price level at Q1 can seem relatively attractive since it is much
closer to the reference price than the opposite best price, which is in that case 2 ticks
away from it. This explains why the market order intensity is larger when the opposite
queue is empty than when its size is small.
8
Note that the computation of the confidence intervals becomes more intricate for this model and the results
presented are slightly approximate ones, see details in appendix.

14

Figure 7: Model IIb : joint distribution of q1 , q1 , France Telecom

Figure 8: Model III: joint distribution of q1 , q1 , France Telecom

2.4.6

Model IIb : Asymptotic behavior

Monte-Carlo simulations are used to obtain the theoretical invariant distribution of the LOB in
Model IIb . The theoretical and empirical joint distributions of Q1 and Q1 are shown in Figure
7. The difference between the two graphs comes from the relatively high probabilities of states
of the form (x, y) with x and y both small in empirical data, which are somehow replaced by
states of the form (x, 0) or (0, y) in the model. Indeed, in practice, a situation where one of
the first queue is empty is not likely to remain long since it often leads to a reference price
change. This effect is not taken into account in Model IIb where the reference price is constant,
but will be investigated in Model III in Section 3.1. We anticipate here by giving in Figure 8
the joint distribution obtained when suitable moves of the reference price are added within the
framework of Model IIb (following the approach of Model III in Section 3.1). We now find that
the simulated density becomes very close to the empirical one.

15

Model II(a)
1

Initial Bid 1 Size

Initial Bid 1 Size

Model I
30
20
0.5
10
5

10
15
20
Initial Ask 1 Size

25

30

30

20
0.5
10
5

20
0.5
10
5

10
15
20
Initial Ask 1 Size

25

30

Poisson model

30

Initial Bid 1 Size

Initial Bid 1 Size

Model II(b)

10
15
20
Initial Ask 1 Size

25

30

30

20
0.5
10
5

10
15
20
Initial Ask 1 Size

25

30

Figure 9: Execution probability of a buying order placed at Q1 at t = 0, France Telecom

2.5

Example of application: Probability of execution

The preceding models can be used to compute short term predictions about several important
LOB related quantities. One relevant example is the probability of executing an order before
the midprice moves. Suppose that at time t = 0, both Q1 and Q1 are not empty. Then a
trader (called A) submits a buy limit order at Q1 of size n0 and waits in the queue until
either the order is executed or the opposite queue Q1 is totally depleted. The probability of
execution can be computed in all of the three preceding models, using Monte-Carlo simulations.
There are two types of orders at Q1 : orders placed before t = 0, thus having higher priority
compared with the order of trader A, and orders placed after t = 0, having lower priority.
When a market order arrives at Q1 , the limit order with the highest priority is executed.
Hence trader As order starts being executed only when all orders placed at Q1 before t = 0
have been either canceled or executed. When a cancellation event happens at Q1 , the precise
order being canceled is not clearly defined in our models. So, we need to make two additional
assumptions for the cancellation process.
Assumption 3. When a cancellation event occurs at Q1 , orders at Q1 have the same probability of being canceled (except for the limit order submitted by trader A, which is never canceled).
q1 n0
Assumption 4. The cancellation intensity at Q1 is supposed to be equal to C
1 (q1 ) q1
instead of C
1 (q1 ), since the order placed by trader A is never canceled.

Orders with lower priority are actually more likely to be canceled, see Gareche, Disdier, Kockelkoren, and Bouchaud (2013). However, in order to investigate precisely this feature, we would
need more detailed market data keeping records of the identifiers of the submitted and canceled
orders. As a result, execution probabilities might be slightly overestimated using Assumptions
3 and 4. Simulation results (for n0 = 1) are shown in Figure 9, together with the predictions
associated to a Poisson model that assumes a linearly increasing cancellation rate. We see that
our three models give fairly similar execution probabilities, while the Poisson model clearly
overestimates them.

16

The queue-reactive model: a time consistent model

with stochastic LOB and dynamic reference price

We now wish to obtain a model which is relevant on the whole period of interest and provides
useful applications.

3.1
3.1.1

Model III: The queue-reactive model

Building the model

Let denote the tick value. We assume here that pref changes with some probability when
some event modifies the midprice pmid . More precisely, when pmid increases/decreases9 , pref
increases/decreases by with probability , provided q1 = 0 at that moment. Hence changes
of pref are possibly triggered by one of the three following events:
The insertion of a buy limit order within the bid-ask spread while Q1 is empty at the
moment of this insertion, or the insertion of a sell limit order within the bid-ask spread
while Q1 is empty at the moment of this insertion.
A cancellation of the last limit order at one of the best offer queues.
A market order that consumes the last limit order at one of the best offer queues.
When pref changes, the value of qi switches immediately to the value of one of its neighbors (right
if pref increases, left if it decreases). Thus, q1 becomes zero when pref decreases/increases.
Recall that we keep records of the LOB up to the third limit. Consequently, the value for
q3 when pref increases/decreases is drawn from its invariant measure. Note that the queue
switching process must be handled very carefully: the average event sizes are not the same for
different queues. So, when qi becomes qj , its new value should be re-normalized by the ratio
between the two average event sizes at Qi and Qj .
To possibly incorporate external information, we moreover assume that with probability reinit ,
the LOB state is redrawn from its invariant distribution around the new reference price when
pref changes. The parameter reinit can be understood as the percentage of price changes
due to exogenous information. In this case, we consider that market participants readjust
very quickly their order flows around the new reference price, as if a new state of the LOB
was drawn from its invariant distribution. A similar approach has been used in Cont and
De Larrard (2013) in a model for best bid and best ask queues, in which reinit is set to 1.
Under these assumptions, the market dynamics is now modeled by a (2K + 1)-dimensional

= N2K N, where
Markov process: X(t)
:= (X(t), pref (t)), in the countable state space
X(t) = (qK (t), ..., q1 (t), q1 (t), ..., qK (t)) represents the available volumes at different limits.
In the sequel, Model I is used to describe the LOB dynamics during periods when pref is
constant (very similar results are obtained in simulations using Model IIa or IIb ). The pref
change probability and the LOB reinitialization probability reinit are calibrated using the 10

9
Note that in this model, pref does not necessarily match its estimated value using the method introduced
in Section 2.2. However, for large tick assets, the difference is negligible.

17

Figure 10: 10 min volatility and mean reversion ratio, France Telecom
minutes standard deviation of the returns of pmid (the volatility) and the mean reversion ratio
introduced in Robert and Rosenbaum (2011), defined by
=

Nc
,
2Na

where Nc is the number of continuations of the estimated pref on the interval of interest (that
is the number of consecutive moves in the same direction) and Na is the number of alternations
(that is the number of consecutive moves in opposite directions)10 . Indeed, the microstructure
of large tick assets is well summarized by the parameter , see Robert and Rosenbaum (2011)
and Dayri and Rosenbaum (2012) and the volatility is of course one of the most important
low frequency statistics. In Figure 10, we show the surfaces of the 10 min volatility and for
different values of and reinit .

3.1.2

About the maximal mechanical volatility

Let us comment now the particular case where we take reinit = 0. In such situation, Model
III becomes a purely order book driven model since the price fluctuations are completely
generated by the LOB dynamics. Our simulations show that under this setting, the maximal
attainable volatility level (when = 1), which we call maximal mechanical volatility, is much
lower than the empirical volatility (5 bps compared with 14 bps for the stock France Telecom).
This suggests that endogenous LOB dynamics alone may not be enough for reproducing the
market volatility. A closer look at these results shows that the model approximates actually
quite well the average frequency of price changes, and that the small value of the mechanical
volatility is mainly due to the strong mean reverting behavior of the price in this purely order
book driven model. This is because of the often reversed bid-ask imbalance immediately after
a change of pref . In Figure 10, we can see that the mean reversion ratio is equal to 0.08 when
= 1, reinit = 0, which is much smaller than the empirical ratio 0.39.

10

Note that here we compute the mean reversion ratio of pref while the transaction price is usually considered.

18

3.2

Example of application: Order placement analysis

We now show how the queue-reactive model can be used in the context of optimal trading. In
the general framework of optimal execution, the trading horizon is split into small slices (typically 5-10 minutes) and an algorithm of execution determines the volume to be executed in each
slice. This problem, often called order scheduling problem, has been widely studied in the literature, see Bertsimas, Lo, and Hummel (1999); Almgren and Chriss (2000); Bouchard, Dang,
and Lehalle (2011) for representative examples on this topic. In practice, another optimization
issue, the order placement problem, arises naturally once the order scheduling problem has
been solved: how should the algorithm place orders to execute the target volume?
This second optimization problem can be seen as the microstructural version of the first one.
However, it is much more difficult to solve. Indeed, the price dynamics can no longer be approximated by a Brownian motion at these (ultra) high frequency scales. Moreover, the queue
priority plays an important role as well as other microstructural features of the asset such as the
tick size, the state of the LOB and the trading speed. This in particular implies that execution
strategies based on limit or market orders can lead to very different outcomes. Some papers
investigate the consequences of using different types of order, see Harris and Hasbrouck (1996),
while others aim at finding the best position to place limit orders, see Laruelle, Lehalle, and
Pag`es (2013). However, in practice, order placement tactics are usually more complex than
the ones considered in the academic literature. For example, traders can hide their trading
intentions by splitting furthermore the target volume within each slice. Also, they may start
passively, sending limit orders, and then switch to market orders when some market conditions
change or a stopping time criteria is met. Very few quantitative tools are available for the analysis of sophisticated tactics and one often needs to rely on so-called market replayers, in which
the number of simulations is limited by that of the available trading days in the historical data.
Moreover, the market impact, that is the average price drift due to our own trading between
the beginning of the execution and a later time, is often neglected. In contrast, our framework
is unlimited in number of simulations and is both relevant and easy to use in order to study
market impact profiles and execution costs of complex placement tactics.
We write ntotal for the total quantity to execute and M for the number of slices. An order
scheduling
strategy gives the target quantity to be executed in each slice, denoted by ni (ni 0
P
n
and M
i=1 i = ntotal ). An order placement tactic can be seen as a predefined procedure of
order management, ensuring the execution of the target quantity within the slice. Here, as
illustration examples, we present two simple tactics, denoted by T1 and T2. In the i-th slice,
both tactics post a limit order of size ni at the best offer queue at the beginning of the period,
and send a market order with all the remaining quantity to complete the execution of the target
volume at the end time of the slice. In between:
T1 (Fire and forget): When pmid changes, cancel the limit order and send a market
order at the opposite side with all the remaining volume if any.
T2 (Pegging to the best): When the best offer price changes or our order is the only
remaining order at the best offer limit, cancel the order and repost all the remaining
volume at the newly revealed best offer queue.
Since an order placement tactic is often specifically designed for a given order scheduling strategy, comparisons between two tactics should take into account the associated scheduling strat19

egy, together with the target benchmark11 . Other parameters can also have influence when
comparing two tactics, such as the total quantity to execute ntotal and the number of slices M .
To simplify our analysis, we simulate a buy order of size ntotal = 60 AES1 , with M = 20 and
the duration of each slice is fixed to 10 minutes (a total trading period of 3h20). We focus on
two benchmarks: the VWAP on the total period (volume weighted average transaction price)
and the arrival price S0 (the midprice when the execution algorithm starts). Moreover, two
types of order scheduling strategies, denoted by S1 and S2, are considered to partly reflect the
diversity of optimal trading schemes:
S1: A linear scheduling (ni = ntotal /M ), used for the VWAP benchmark.
S2: An exponential scheduling ni = ntotal (e(i1)/4 ei/4 ), used for the benchmark S0 .
Finally, note that Assumptions 3 and 4 are in force for the order cancellation processes.
3.2.1

Tactic performance analysis

The performance of an execution algorithm is often measured by its slippage, defined (for a
buy order) by
Pbenchmark Pexec
.
Slippage =
Pbenchmark
To understand the effects of the order placement tactic on the executions slippage, we define
the theoretical scheduling slippage by:

theo
Pexec

M
X

ni VWAPi

i=1

Slippagetheo

theo
Pbenchmark Pexec
=
,
Pbenchmark

where VWAPi denotes the volume weighted average transaction price of the ith slice. Indeed,
VWAPi is often considered as a simple proxy for the execution price in the slice when one
focuses on the scheduling algorithm. Hence, Slippagetheo essentially measures the quality of the
scheduling strategy and neglects the randomness in execution prices due to the order placement tactic. Note that here a market impact component is included in the computation of the
theoretical scheduling slippage. This is because the value of VWAPi in each slice is obviously
impacted by our execution.
We launched 2000 simulations for each couple of (S1/S2, T1/T2). The intensity functions
estimated for the stock France Telecom are used in these simulations, as well as the two parameters = 0.7 and reinit = 0.85 calibrated in Section 3.1. Furthermore, we use a standard
kernel smoothing method when estimating the probability density functions of Slippagetheo and
Slippage. The results are shown in Figure 11.

11
In execution services, the client often wants the execution algorithm to target some specific price (the
arrival price, the average market price during a predefined period,...). The quality of the execution is then
assessed on the basis of the difference between the realized execution price and this target benchmark price.

20

Linear scheduling, VWAP

0.9

0.7
0.6

0.02

Density Function

0.8

Density Function

Exponential scheduling, Arrival Price

0.03
s1+t1,simulated slippage
s1+t1,theoretical slippage
s1+t2,simulated slippage 0.025
s1+t2,theoretical slippage

0.5
0.4
0.3

s2+t1,simulated slippage
s2+t1,theoretical slippage
s2+t2,simulated slippage
s2+t2,theoretical slippage

0.015

0.01

0.2
0.005
0.1
0

-6

-4

-2

0
2
Slippage (bp)

0
-80

-60

-40

-20
0
Slippage (bp)

20

40

60

Figure 11: Simulation results for the tactics

Figure 11 suggests that the slippage distributions of the same scheduling strategy using two
different tactics can be very different: T2 (Pegging to the best) performs better than T1
(Fire and forget) when being coupled with a linear scheduling strategy with VWAP benchmark, while T1 slightly outperforms T2 when an exponential scheduling strategy with arrival
price benchmark is considered. In our setting, the limit orders change the queue sizes and
therefore modify the behaviors of the order flows. Consequently they generate market impact.
By constantly following the best offer queue until the total volume is filled, T2 achieves on
average a higher passive execution rate (defined as the volume passively executed12 divided by
the total executed volume). Thus, in each slice, it often obtains a better price than that of a
more market orders based tactic. However, at the same time, it creates a larger impact than
T1 since the order stays longer in the queues. This explains why the theoretical scheduling
slippage of T2 is worse than that of T1 for an execution with arrival price benchmark using
an exponential scheduling strategy.
3.2.2

Market impact profiles

We now study the market impact profiles of these two tactics. Recall that an order placement
tactic has two parameters: the slice duration T and the quantity to execute n. In the following experiments, T is set to 10 minutes, and the value of n varies from 1 to 60 AES1 . We
denote by MIi (t, n) the market impact at time t of Tactic i with target quantity n, defined by:
0
], with St the midprice at time t. We launched 2000 simulations for each
MIi (t, n) = E[ StSS
0
value of n, t in the ranges 1-60 AES1 and 1-600 seconds. Impact profiles are given in Figure 12.
In agreement with the celebrated square-root law, see Gatheral (2010); Toth, Lemperiere,
Deremble, De Lataillade, Kockelkoren, and Bouchaud (2011a); Farmer, Gerig, Lillo, and Waelbroeck (2013), the market impact curves are concave both in time and volume. One can also
see that the impact of T1 is quite instantaneous and depends essentially on the target quantity
n, while the impact of T2 is a progressive process, depending both on the target quantity n
and the time t. Remark that T2 seems suitable when dealing with small orders since its market
impact is small and it has a higher passive execution rate than T1. If one needs to trade larger
12
A buy execution is said to be passive if it occurs at the bid side of the LOB, aggressive if it occurs at the
ask side of the LOB.

21

Market impact profile Tactic 1

Market impact profile Tactic 2

25

n=1
n = 10
n = 20
n = 30
n = 40
n = 50
n = 60

20

n=1
n = 10
n = 20
n = 30
n = 40
n = 50
n = 60

100

200
300
400
Time (second)

500

Impact (bp)

Impact (bp)

600

15

10

100

200
300
400
Time (second)

500

600

Figure 12: Market impact profiles

orders, T1 becomes probably more relevant since the cost of market impact is likely to outweigh
the benefit from passive execution of T2. Finally, note that in our Markovian framework, no
significant price relaxation (that is the fact that on average, after the completion of the execution of a buy order, the price may drop to a lower level than the one reached at the end of the
execution) can be observed.

Conclusion and perspectives

In this work, we have modeled market participants intelligence through their average behaviors
towards various states of the LOB. This enabled us to analyze the different order flows and to
design a suitable market simulator for practitioners, allowing notably to investigate the transaction costs of complex trading strategies. To our knowledge, our model is the first one where
such pre-trade cost analysis is possible in a simple and efficient way.
Another important public information, the historical order flow, is not considered in this approach. Market order flows have been shown to be autocorrelated in several empirical studies,
see for example Toth, Palit, Lillo, and Farmer (2011b). Thus, adding such feature in our framework would probably be relevant. Another possible direction for future research would be to
explain the shape of the estimated intensity functions in a more sophisticated way. For example,
it would be interesting to design some agent based model where these repetitive patterns of the
LOB dynamics would be reproduced, providing an even better understanding of the nature of
these intensity curves.

22

5
5.1

Appendix
Proof of Theorem 2.1

Proof. For some z > 1, set

K
X

V (q) =

z |qj Cbound |+ .

j=K,j6=0

For any q , we have:

X
QV (q) =
Qq,p [V (p) V (q)]
p6=q

K
X

[fi (q)(z |qi +1Cbound |+ z |qi Cbound |+ ) + gi (q)(z |qi 1Cbound |+ z |qi Cbound |+ )]

i=K,i6=0

K
X

1
[fi (q)1qi Cbound z |qi Cbound |+ (z 1) + gi (q)1qi Cbound +1 z |qi Cbound |+ ( 1)]
z
i=K,i6=0

= (z 1)

K
X

[fi (q)1qi Cbound

i=K,i6=0

= (z 1)

fi (q) + (z 1)

i:qi =Cbound

gi (q)1qi Cbound +1 |qi Cbound |+

]z
z
[fi (q)

i:qi >Cbound

gi (q) qi Cbound
]z
.
z

(1)

Under Assumption 1 and 2, we can find a z sufficiently close to 1 such that, if qi > Cbound ,
fi (q)

gi (q)
< z 1 (r + H(z 1)) = r0 < 0.
z

So, from Equation (1), we have

QV (q) (z 1)H (z 1)

r0 z qi Cbound

i:qi >Cbound

(z 1)r0

z |qi Cbound |+ + (z 1)H + 2(z 1)r0 K

i
0

(z 1)r V (q) + (z 1)[H + 2r0 Kz].

Thus X(t) is V-uniformly ergodic. Then using Theorem 4.2 in Meyn and Tweedie (1993), X(t)
is Harris positive recurrent and has a finite invariant measure. Furthermore, by Theorem 3.6.2
in Norris (1998), the process X(t) converges to its equilibrium and is therefore ergodic.

5.2

Computation of confidence intervals

When the queues are independent, by the central limit theorem, we have, with asymptotic
+ ,q ()=n}
i
probability 95% (we note pi L (n) = #{T ()E
):
#{qi ()=n}

i (n) + p 1.96i (n) ]

i (n) p 1.96i (n) ,
i (n) [
#{qi () = n}
#{qi () = n}
q
q
L
L
L
1.96 pi (n)(1 pi (n)) L
1.96 pi L (n)(1 pi L (n))
i (n)
L
p
p
[pi (n)
, pi (n) +
].
i (n)
#{qi () = n}
#{qi () = n}
23

So, at least with probability 90%:

q
1.96
pi L (n)(1 pi L (n))
L
L

p
i (n) [(i (n) p
)(pi (n)
)
#{qi () = n}
#{qi () = n}
q

pi L (n)(1 pi L (n))
1.96
1.96i (n)
L

p
(i (n) + p
)(pi (n) +
)].
#{qi () = n}
#{qi () = n}
i (n)
1.96

M
Similar results can be computed for C
i and i . The method used to compute confidence
intervals of Model IIa is quite similar. Confidence intervals are more difficult to compute in
Model IIb , and we use approximations by neglecting the possible intersections between the two
sets: {q1 () = n, Sm,l (q1 ()) s)} and {q1 () = n, Sm,l (q1 ()) s)}.

5.3

Quasi birth and death process

Definition 5.1. (Quasi birth and death process, from Latouche and Ramaswami (1999)): A
quasi birth and death (QBD) process is a bivariate Markov process with countable state space
S = {(i, j) : i 0, j = 0, 1, ..., m} where the first element i is called the level of the process, and
the second element j is called the phase of the process. The parameter m can be either finite or
infinite. The process is restricted in level jumps only to its nearest neighbors, meaning that the
probability of jumping from level i directly to level l, l i + 2 or l i 2 is equal to zero.
We can easily see that the Markov process (q1 , q2 ) in Model IIa is indeed a QBD process with
countable phases. Its infinitesimal generator matrix is of the following form:
(0)

(0)
A1 A0
0
0 ...
(1)

(1)
(1)
A
A1 A0
0 ...
Q= 2
,
(2)
(2)
0 A(2)
A1 A0 ...
2
...
...
...
... ...
(`)

(`)

where the matrix A0 encodes transitions from level q1 = ` to level q1 = ` + 1, matrix A2

(`)
encodes transitions from level q1 = ` to level q1 = ` 1, and matrix A1 encodes transitions
(`)
within level q1 = `. More specifically, the element (i, j) of A0 is the transition rate from state
(`)
(q1 = `, q2 = i) to state (q1 = ` + 1, q2 = j), the element (i, j) of A2 is the transition rate from
(`)
state (q1 = `, q2 = i) to state (q1 = ` 1, q2 = j), and the element (i, j) of A1 is the transition
rate from state (q1 = `, q2 = i) to state (q1 = `, q2 = j).
(`)

We write the intensity functions at Q2 when q1 = 0 with a . For matrix Ai , i = 0, 1, 2, we

have:
(k)

= L1 (k)I,

(k)

M
= (C
1 (k) + buy (k))I,

A0

A2

L (0)
L (0)
L1 (0)

0
2
2
L
C (1) +
M (1) L (0)
L (1)
C (1)
M (1)

2 (1)
2
1
2
2
buy
buy
=
C
M
L
L

C (2)
M (2)
0
2 (2) + buy (2)
1 (0) 2 (2)
2
buy
...
...
...
24

(0)

A1

...
...
,
...
...

and for k 1:

M
L
L
C
L2 (0)
0
...
1 (k) buy (k) 1 (k) 2 (0)
(k)
M
L
L
C
L
.
C
C
A1 =
2 (1)
1 (k) buy (k) 1 (k) 2 (1) 2 (1) 2 (1) ...
...
...
...
...
We define i,j = P[q1 = i, q2 = j] the stationary distribution of this QBD process, and:
n = [n,0 , n,1 , ...]
= [0 , 1 , ...].
We shall have:
Q = 0
1 = 1.
The dynamics of the two queues system (q1 , q2 ) is level dependent, meaning that its transition kernel depends on the value of q1 . This makes the computation or approximation of its
asymptotic behavior quite difficult. Thus we consider an additional assumption in order to turn
(q1 , q2 ) into a so-called level independent QBD process. This is particularly interesting since it
enables us to easily express the invariant measure in a matrix geometric form and to compute
(i)
(i)
it numerically. The level independence property is defined by the fact that for i 1, A0 , A1
(i)
(0)
(i)
and A2 do not depend on i and A0 = A0 , see Latouche and Ramaswami (1999). Under the
following assumption, this property is satisfied by (q1 , q2 ) in Model IIa .
Assumption 5. (Independent Poisson flows at first limits) There are two positive constants
1 and 1 , with 1 < 1 , such that for k 1:
M
C
1 (k) + buy (k) = 1

L1 (k) = 1
L1 (0) = 1 .
In practice, 1 and 1 are taken as the average values of the estimated intensity functions
at first limits. Under this assumption, a quite simple numerical computation of the invariant
distribution is possible. Generally speaking, QBD processes with finite phase (meaning that
the value set of the second dimension, in our case q2 , is finite) can be easily treated, see for
example Latouche and Ramaswami (1999). In the infinite case, truncation methods must be
applied to obtain approximate results. Thanks to the special structure of the generator in our
model, one simple truncation method, called first column augmentation by block, can be
applied. Details of this truncation method can be found in Bean and Latouche (2010). The
mathlab toolbox SMCSolver, see Bini, Meini, Steffe, and Van Houdt (2006), is used to compute
the invariant measure.

5.4

Alcatel-Lucent

Results for the stock Alcatel-Lucent are presented in the following figures (Figure 13 to Figure
18).
25

Limit order cancellation intensity, Model I

0.7

1
0.8
0.6

0.5
0.4
0.3

First limit
Second limit
Third limit

0.2

Intensity (num per second)

1.2

Market order arrival intensity, Model I

0.4

0.3
0.25
0.2
0.15
0.1

0.4

0.1

0.05

0.2

0
10
20
30
40
50
Queue Size (per average event size)

First limit
Second limit
Third limit

0.35

0.6
Intensity (num per second)

Intensity (num per second)

Limit order insertion intensity, Model I

1.6
First limit
Second limit
1.4
Third limit

0
10
20
30
40
50
Queue Size (per average event size)

0
10
20
30
40
50
Queue Size (per average event size)

Figure 13: Intensities at Q1,2,3 , Alcatel Lucent

q1 == 0
q1 > 0

0.8

Intensity (num per second)

Intensity (num per second)

0.9

0.7
0.6
0.5
0.4
0.3
0.2

Second limit, limit order cancellation

1
q1 == 0
0.9
q1 > 0
0.8
0.7
0.6
0.5
0.4
0.3
0.2

0.1

0.1

0
10
20
30
40
Queue Size (per average event size)

Second limit, market order insertion

0.15
q1 == 0
q1 > 0
Intensity (num per second)

Second limit, limit order insertion

1

0.1

0.05

0
10
20
30
40
Queue Size (per average event size)

0
10
20
30
40
Queue Size (per average event size)

Figure 14: Intensities at Q2 as functions of 1q1 >0 and q2 , Alcatel Lucent

First limit, limit order insertion

First limit, limit order cancellation

0.8

0.5
0.4
0.3
0.2

0.5
0.4
0.3
q(-1) = 0
0<q(-1)<=2
2 < q(-1) <= 5
q(-1) > 5

0.2

0.1

0.1

0
10
20
30
40
Queue Size (per average event size)

0.25

0.6

0
10
20
30
40
Queue Size (per average event size)

Intensity (num per second)

0.6

0.7
Intensity (num per second)

Intensity (num per second)

0.7

First limit, market order insertion

0.3

0.8
q(-1) = 0
0<q(-1)<=2
2 < q(-1) <= 5
q(-1) > 5

q(-1) = 0
0<q(-1)<=2
2 < q(-1) <= 5
q(-1) > 5

0.2

0.15

0.1

0.05

0
10
20
30
40
Queue Size (per average event size)

Figure 15: Intensities at Q1 as functions of Sm,l (q1 ) and q1 , Alcatel Lucent

26

First limit

Second limit

0.2
Empirical estimation
Model I
Poisson Model

0.18
0.16

0.2
Empirical estimation
Model I
Poisson Model

0.18
0.16

0.16

0.1
0.08

0.14
Distribution

0.12

0.12
0.1
0.08

0.12
0.1
0.08

0.06

0.06

0.06

0.04

0.04

0.04

0.02

0.02

0.02

10
20
30
40
Queue size (in AES)

50

Empirical estimation
Model I
Poisson Model

0.18

0.14
Distribution

0.14
Distribution

Third limit

0.2

10
20
30
40
Queue size (in AES)

50

10
20
30
40
Queue size (in AES)

Figure 16: Queue distribution, Alcatel Lucent

Figure 17: Model IIa : joint distribution of q1 , q2 , Alcatel Lucent

Figure 18: Model IIb : joint distribution of q1 , q1 , Alcatel Lucent

27

50

5.5

AES

AESi is defined as the average size of all events (including limit order insertion, cancellation
and trades) at Qi , while ATS computes only the average size of all trade events. In Table 5.5
we show the estimated values of AES at different distances to pref and the estimated value of
ATS, for the stocks France Telecom and Alcatel-Lucent.
stock
France Telecom
Alcatel Lucent

ATS
637
2340

AES1
836
3033

AES2
1068
3451

AES3
1069
3528

Table 2: AES and ATS (in number of stocks)

References
Abergel, F. and Jedidi, A. (2011), A mathematical approach to order book modelling, in
Econophysics of Order-driven Markets, Springer, pp. 93107.
Almgren, R. and Chriss, N. (2000), Optimal execution of portfolio transactions, Journal of
Risk, 3, 539.
Avellaneda, M. and Stoikov, S. (2008), High-frequency trading in a limit order book, Quantitative Finance, 8, 217224.
Bean, N. and Latouche, G. (2010), Approximations to quasi-birth-and-death processes with
infinite blocks, Advances in Applied Probability, 42, 11021125.
Bertsimas, D., Lo, A. W., and Hummel, P. (1999), Optimal control of execution costs for
portfolios, Computing in Science and Engg., 1, 4053.
Bini, D., Meini, B., Steffe, S., and Van Houdt, B. (2006), Structured Markov chains solver:
algorithms, in Proceeding from the 2006 workshop on Tools for solving structured Markov
chains, ACM, p. 13.
Bouchard, B., Dang, N.-M., and Lehalle, C.-A. (2011), Optimal control of trading algorithms:
a general impulse control approach, SIAM Journal on Financial Mathematics, 2, 404438.
Cont, R. and De Larrard, A. (2013), Price dynamics in a Markovian limit order market,
SIAM Journal on Financial Mathematics, 4, 125.
Cont, R., Stoikov, S., and Talreja, R. (2010), A stochastic model for order book dynamics,
Operations research, 58, 549563.
Dayri, K. and Rosenbaum, M. (2012), Large tick assets: implicit spread and optimal tick size,
Preprint arXiv:1207.6325.
Delattre, S., Robert, C. Y., and Rosenbaum, M. (2013), Estimating the efficient price from the
order flow: a Brownian Cox process approach, Stochastic Processes and their Applications,
123, 26032619.
28

Farmer, J. D., Gerig, A., Lillo, F., and Waelbroeck, H. (2013), How efficiency shapes market
impact, Quantitative Finance, 13, 17431758.
Gareche, A., Disdier, G., Kockelkoren, J., and Bouchaud, J.-P. (2013), A Fokker-Planck description for the queue dynamics of large tick stocks, Preprint arXiv:1304.6819.
Gatheral, J. (2010), No-dynamic-arbitrage and market impact, Quantitative Finance, 10,
749759.
Gross, D. and Harris, C. M. (1998), Fundamentals of queuing theory, Wiley New York.
Harris, L. and Hasbrouck, J. (1996), Market vs. limit orders: the SuperDOT evidence on order
submission strategy, Journal of Financial and Quantitative analysis, 31, 213231.
Laruelle, S., Lehalle, C.-A., and Pag`es, G. (2013), Optimal posting price of limit orders:
learning by trading, Mathematics and Financial Economics, 7, 359403.
Latouche, G. and Ramaswami, V. (1999), Introduction to matrix analytic methods in stochastic
modeling, vol. 5, ASA-SIAM.
Madhavan, A., Richardson, M., and Roomans, M. (1997), Why do security prices change? A
transaction-level analysis of NYSE stocks, Review of Financial Studies, 10, 10351064.
Meyn, S. P. and Tweedie, R. L. (1993), Stability of Markovian processes III: Foster-Lyapunov
criteria for continuous-time processes, Advances in Probability, 518548.
Norris, J. R. (1998), Markov chains, no. 2008, Cambridge University Press.
Robert, C. Y. and Rosenbaum, M. (2011), A new approach for the dynamics of ultra-highfrequency data: the model with uncertainty zones, Journal of Financial Econometrics, 9,
344366.
Smith, E., Farmer, D. J., Gillemot, L., and Krishnamurthy, S. (2003), Statistical theory of the
continuous double auction, Quantitative Finance, 3, 481514.
Toth, B., Lemperiere, Y., Deremble, C., De Lataillade, J., Kockelkoren, J., and Bouchaud, J.-P.
(2011a), Anomalous price impact and the critical nature of liquidity in financial markets,
Physical Review X, 1, 021006.
Toth, B., Palit, I., Lillo, F., and Farmer, J. D. (2011b), Why is order flow so persistent?
arXiv preprint arXiv:1108.1632.
Wyart, M., Bouchaud, J.-P., Kockelkoren, J., Potters, M., and Vettorazzo, M. (2008), Relation between bidask spread, impact and volatility in order-driven markets, Quantitative
Finance, 8, 4157.

29