Rama Cont 2

Limit order markets
A limit order book model with heterogeneous order flow

High-frequency dynamics of the limit order book
Price dynamics in Limit Order Markets:

from multi-scale stochastic models to free-boundary problems
Rama Cont
Dept of Mathematics
Imperial College London
Imperial-ETH Workshop on Mathematical Finance, 2015
Rama CONT Price dynamics in Limit Order Markets:

Limit order markets
References :
Rama Cont, Sasha Stoikov and Rishi Talreja (2010) A stochastic
model for order book dynamics, Operations Research, Volume 58,
No. 3, 549-563.
Rama CONT (2011) Statistical modeling of high frequency data:
facts, models and challenges, IEEE Signal Processing, Vol 28,
No 5, 16–25.
Rama Cont and Adrien de Larrard (2013) Price dynamics in a
Markovian limit order market, SIAM Journal on Financial
Mathematics, Vol 4, 1–25.
Rama Cont and Adrien de Larrard (2011) Order book dynamics in
liquid markets: limit theorems and diffusion approximations,
http://ssrn.com/abstract=1757861.
Rama Cont and Adrien de Larrard (2012) Price dynamics in limit
order markets: linking volatility with order flow, Working Paper.
Rama Cont (2014) High frequency dynamics of limit order markets:
multi-scale modeling and asymptotic analysis.
Limit order markets
Outline
1 Limit order markets

2 Stochastic models of the limit order book
3 PDE models of the price formation: Lasry & Lions (2007)
4 The separation of time scales
5 A multiscale model of trading in limit order markets
6 Hydrodynamic limit: a two -phase moving boundary problem

Limit order markets
A limit order book model with heterogeneous order flow Time scales
At the core of liquidity: the limit order book
Figure: A limit order book.

Limit order markets
Limit orders
A limit order is an order to buy (sell) a certain quantity at a given price.
Limit orders queue according to time priority until they are executed
against a market order.
Figure: A limit buy order: Buy 2 at 69200.

Limit order markets
A market order
A market order is an order to buy (sell) a certain quantity at the best
available price. Market orders are executed immediately against available
limit orders at the best price.
Figure: A market sell order of 10.

Limit order markets
A cancellation
Figure: Cancellation of 3 sell orders at 69900.

Limit order markets
Electronic Limit order markets
The advent of electronic trading has transformed markets and led to

a new market landscape dominated by algorithms which can submit
and cancel orders at very high speeds.
This has enabled the emergency of High Frequency Trading (HFT),
a new category of trading strategies operating at millisecond
frequency.
At the same time, there exists a population of market participants
submitting orders at lower frequencies.
The delicate balance between these heterogeneous order flows was
seen as the root of the recent Flash Crash(es).

Limit order markets
Flash Crash
Figure: The Flash Crash of May 2010 in the US equity markets.

Limit order markets
Electronic Limit order markets
Many questions of interest to regulators and market participants:

What do we understand about market dynamics in such an
environment?
How does order flow interact with price dynamics?
How does high-frequency market activity affect market dynamics at
lower frequencies?
How does the co-existence of heterogeneous order flows operating at
different frequencies affect market depth/liquidity and price
dynamics?
Our objective: develop a quantitative modeling framework capable of
providing some analytical insight into these complex questions.

Limit order markets
Stochastic models of order book dynamics
Traditional market microstructure theory models the strategic

interaction between a small number of agents
(informed/non-informed trader, market maker) in a 1 or 2 period
game theoretical setting, with an emphasis on information
asymmetry and adverse selection.
These models provide conceptual insights into market design and
analysis but are not amenable to quantitative analysis or a realistic
comparison with data: need for quantitative modeling.
The recent years have witnessed the emergence of stochastic models
for order book dynamics, which aim at incorporating the information
in the order flow in view of
1 estimation of intraday risk (volatility, loss distribution)
2 short-term (< second) prediction of order flow and price movements
for trading strategies
3 optimal order execution
These applications requires analytical tractability and computability.
Limit order markets
Limit order books as queueing systems
A limit order book may also be viewed as a system of queues subject to

order book events modeled as a multidimensional point process. A
variety of stochastic models for dynamics of order book events and/or
trade durations at high frequency: Poisson processes for each order type,
Self exciting and mutually exciting Hawkes processes ( Cont, Jafteson &
Vinkovskaya 2010, Bacry et al 2010), Autoregressive Conditional Duration
(ACD) model (Engle & Russell 1997, Engle & Lunde 2003, ..), ...
Most of these models are high-dimensional and applications may require
heavy simulation/ numerics.
In general: price is not Markovian, increments neither independent nor
stationary and depend on the state of the order book.
Common approach: model separately order flow dynamics and price
dynamics through ad-hoc price impact relations/assumptions.

Limit order markets
Example: a Markovian limit order book
C. , Stoikov, Talreja (Operations Research, 2010) [CST 2010]

State of limit order book X (t) = (Xi (t)): Xi (t) = volume of limit order
(< 0 for sell, > 0 for buy) at price level i.
Bid / ask price:
pb (t) = sup{i = 1..N, Xi (t) > 0} ≤ pa (t) = inf{i = 1..N, Xi (t) < 0}
Arrival of market orders, limit orders and cancelations at different
price levels i = 1..N described by a (spatial) Poisson point process
with intensity depending on distance from best quote.
All orders have same size.
→ limit order book X (t) described by a continuous-time Markovian
birth-death process ⇒ analytical formulas for
distribution of durations between price changes,
distribution of time to execution of limit orders,
probability of price increase conditional on state of the order book.

Limit order markets
The limit order book as a measure-valued process
The state of limit order book may be viewed as a signed measure µ on R:

µ(B) = vol of limit buy orders with prices in B - vol. of limit sell orders
with prices in B
The buy/sell side of the book correspond to the Hahn-Jordan
decomposition of the measure µ:
µ = µ+ − µ− a(µ) = inf (supp(µ− )) ≥ b(µ) = sup (supp(µ+ )) ,
supp(µ+ ) ⊂ (−∞, b(µ)] supp(µ− ) ⊂ [a(µ), ∞)

We denote L the set of signed measures whose Hahn-Jordan
decomposition is of the form above.
Thus, the limit order book may be viewed in terms of a pair of Radon
measures (µ+ , µ−) ∈ M(R)2 .
In the above example, this leads to a measure-valued Markov process
with values in M(R)2 .

Limit order markets
PDE models of price formation
Lasry & Lions (2007) proposed a PDE model for the dynamics of the
density of buy/sell orders: this model assumes µ± t (dx) = ρ± (t, x)dx and
postulate that the density ρ is the solution of the following free boundary
problem:
∂ρ+ σ 2 ∂ 2 ρ+ ∂ρ−
= + 2
− (t, St )δSt −a for x < St (1)
∂t 2 ∂x ∂x
∂ρ− σ 2 ∂ 2 ρ− ∂ρ+
= − − (t, St )δSt +a for x > St (2)
∂t 2 ∂x 2 ∂x
− +
ρ (t, x) = 0 for x > St , ρ (t, x) = 0 for x ≥ St (3)
Interpretation: after trading takes place at price St , buyers become sellers

at price St + a and sellers become buyers at price St − a where a > 0 is a
’transaction cost’.

Limit order markets
PDE models of price formation
∂ρ+ σ 2 ∂ 2 ρ+ ∂ρ−
= + − (t, St )δSt −a for x < St (4)
∂t 2 ∂x 2 ∂x
2
∂ρ− σ ∂ 2 ρ− ∂ρ+
= − 2
− (t, St )δSt +a for x > St (5)
∂t 2 ∂x ∂x
ρ− (t, x) = 0 for x > St , ρ+ (t, x) = 0 for x ≥ St (6)
Caffarelli, Markowich & Pietsch (2013), Caffarelli, Markowich & Wolfram

(2011)
there exists a unique smooth solution
price dynamics is continuous: S ∈ C ([0, ∞), R)
−
if µ+
0 (−∞, S0 ) = M+ 6= µ0 (S0 , ∞) = M− then
t→∞ √ M+
St ∼ t erf −1 ( )
M−

Limit order markets
Relation between modeling approaches
What is the relation between the discrete, stochastic models

describing high-frequency dynamics of limit order books and the
PDE-based price formation models?
Can the latter be derived as an appropriate scaling limit of the
former and if so, under what assumptions?
How are the parameters of the PDE models related to the
parameters of the point processes describing order flow at high
frequency?
Tool: asymptotic analysis of the fluid limit for stochastic limit order book
models

Limit order markets
A hierarchy of time scales
Regime Time scale Issues

Ultra-high frequency (UHF) ∼ 10−3 − 1 s Microstructure,
Latency
High Frequency (HF) ∼ 10 − 102 s Optimal execution
Low Frequency (minutes - hours) ∼ 103 − 104 s Trading strategies,
Hedging
The separation between these time scales opens the door to the use of
asymptotic methods for connecting dynamics at different time scales.
Idea: start from a description of the limit order book at the finest scale
and derive probabilistic limit theorems for computing quantities at larger
time scales.
Analogies with ’hydrodynamic description’ of interacting particle systems.

Limit order markets
Moving across time scales: fluid and diffusion limits
Idea: study limit of rescaled limit order book as

tick size → 0
frequency of order arrivals → ∞
order size → 0
All these quantities are usually parameterized / scaled as a power of a
large parameter n → ∞, which one can think of as number of market
participants or frequency of orders.
The limit order book having a natural representation as a (pair of)
measures, vague convergence in D([0,∞), M(R)2 ) is a natural notion of
convergence to be considered.
Various combination of scaling assumptions are possible, which may lead
to very different limits.

Limit order markets
Moving across time scales: fluid and diffusion limits

Various combination of scaling assumptions are possible for the same
process, which lead to very different limits.
When scaling assumptions are such that variance vanishes asymptotically,
the limit process is deterministic and often described by a PDE or ODE:
this is the functional equivalent of a Law of Large Numbers, known as
the ’fluid’ ( or ’hydrodynamic’ limit).
Ex: Nin Poisson process with intensity λin .
n
N1 − N2n

n→∞
λin ∼ nλi , t ≥ 0 ⇒ ((λ1 − λ2 )t, t ≥ 0)
n
Other scaling assumptions for the same process may lead to a random
limit (”diffusion limit”). Example:
√ N1n − N2n n→∞
λin ∼ nλ, λ1n − λ2n = σ 2 n, √ ⇒ σW
n

Limit order markets
’Heavy traffic’ asymptotics for limit order books

Kruk (2003): fluid limit for a simple auction process
Cont & Larrard (2013): diffusion limit of a reduced-form Poisson
order book model → diffusion dynamics for bid/ask queue sizes,
price jumps at each exit time of queue from positive orthant
Cont & Larrard (2012): diffusion limit of a reduced-form order book
with general point process dynamics → diffusion limit for price,
expression of price volatility in terms of order intensities
Maglaras & Moallemi (2013): fluid limit for a modified (CST 2010)
model → (Average) order book profile and price described by ODE
Dai et al (2013): fluid limit for (CST 2010) model → constant price,
linear ODE with constant coefficients for limit order book
Horst & Paulsen (2013) : fluid limit for a model with IID Poisson
arrivals → ODE for price, 1st order PDE for limit order book
Lakner et al (2014): (yet another) heavy traffic limit of a one-sided
(CST 2010) model → strictly increasing price process, degenerate
(single price level) or flat (block) limit for order book.
Limit order markets
’Heavy traffic’ asymptotics for limit order books
Which scaling limit is the relevant one for real markets?

Constant, deterministic or strictly increasing price dynamics do not
seem realistic, neither at high frequency nor at daily frequency.
We would like to derive the price process from order book dynamics
rather than specify it exogenously.
Moreover, given the heterogeneity of the order flow (co-existence of
high and low frequency traders) it is unlikely that a single time
scale/ frequency will give the right asymptotics.
Finally: all these models are queueing models where limit orders
arrive randomly and wait for execution. Is that ALL that is going on
in the order flow?
How to choose the right scaling assumptions? Hint: examine the
DATA...

Limit order markets
Net order flow at the bid and ask levels displays a diffusion-like behavior
over a time scale of seconds or minutes.
Figure: Intraday dynamics of net order flow at bid and ask: Citigroup, June 26,
2008.
Limit order markets
A limit order book model with heterogeneous order flow Decomposition of the order flow into components
Decomposition of the order flow into components
(Joint with A Kirilenko, A Kukanov, E Vinkovskaya)

Study of detailed database of order flow in one of the most liquid
eletronic markets: S&P e-mini futures market (CME).
Electronic limit order market with around 10,000 participants (trade
accounts).
Data: all messages exchanged between market participants and
Globex: creation/modification/cancelation of new orders, execution
confirmations
Trader IDs included in data: we can trace order flow of a given
account.
We compute for each trader ID, a range of statistics to describe the
characteristics of its sequence of orders/cancels.
These statistics are then used to classify trader IDs in more or less
homogeneous groups with similar characteristic of order flow.

Limit order markets
Decomposition of the order flow into components
Market order flow is a superposition of heterogeneous order flows

operating at widely different frequencies.
The vast majority of accounts are ”low frequency traders” who
submit infrequent, small orders, cancel very few of them, trade
directionally, and accumulate inventory. These are the main
contribution to the volume of the order book at deeper levels .
A very small number of HFTs account for around 50% of volume of
orders and trades. Their order flow is concentrated close to the
bid/ask with the vast majority of orders being placed at the best or
second-best levels. Cross section distribution of order arrivals and
sizes appear to be random.
HFT order flow is NOT simply an accelerated version of the order
flow of other participants: in particular, HFTs do not accumulate
inventory, contribute zero net volume to the book on average and
shift orders across different levels close to the best bid/ask.

Limit order markets
A multi-component order flow model
Motivated by these observations, we model the order flow as a

superposition of two distinct components.
The 1st component is a large population of ”low frequency traders”
who submit infrequent, small orders (order size → 0) at all price
levels, cancel very few of them. Their order flow is modeled as a
Poisson point process as in (CST 2010). The heavy traffic regime
arises here due to their large number n → ∞, but the volume of
orders at each level remains finite .
If the price submitted by such a trader is better than the best
available price, it is executed as a market order, at the best quote of
the opposite side: this leads to an intensity of market orders at the
best bid/ask levels which is ALSO of order n.
Order arrival intensities can be allowed to depend on distance to
best bid/ask and more generally, on the state of the order book.

Limit order markets
A multi-component order flow model
2nd component: HFT order flow

The 2nd component (”HFT order flow”) is a constantly balanced
order flow occuring at high frequency (∼ n): each buy order is
followed by a sell order after a very short time (∼ 1/n) and vice
versa. Thus, the net result is that this components shifts orders in
the order book from one level to a neighboring one.
A very small number of HFTs account for around 50% of volume of
orders and trades. Their order flow is concentrated close to the
bid/ask with the vast majority of orders being placed at the best or
second-best levels. Cross section distribution of order arrivals and
sizes appear to be random.
At the best bid/ask, we thus have submission/deletion
√ of orders at
rates ∼ n: to account for the fact that best/bid ask n

Limit order markets
A multi-component model of order flow
Mathematical model: a multi-scale Markov model for the limit order

book
State space
L = {η : Z → Z, ∃p ∈ Z, η1x<p ≥ 0, η1x>p ≤ 0}
For η ∈ L define
Bid price : b(η) = sup{x ∈ Z, η(x) > 0}
Ask price : a(η) = inf{x ∈ Z, η(x) < 0}
We will now describe the evolution of the order book in L through
elementary ’order book events’ and their occurrence rates.

Limit order markets
A multi-component model of order flow
i Arrival of a new limit order at price x ∈ N at rate λ+ (x, η):
(i) x ≤ b(η) η 7→ η + 1x at rate λb+ (x − b(η), η)
x ≥ a(η) η 7→ η − 1x at rate λa+ (x − a(η), η)

ii Cancellation of a limit order without replacement, at rate λ− :
(ii) x ≤ b(η) η 7→ η − 1x at rate λb− (x − b(η), η)
x ≥ a(η) η 7→ η + 1x at rate λa− (x − a(η), η)

iii Cancellation of an order and its replacement by an order closer to
the bid/ask:
(iii) x < b(η) : η 7→ η + 1x+1 − 1x at rate r b (x − b(η), η)
x > a(η) : η 7→ η − 1x−1 + 1x at rate r a (x − a(η), η)

Limit order markets
iv Execution of HFT orders: limit orders at the best bid/ask prices may
get executed against incoming market orders of the opposite sign.
If a market order is executed against a limit order posted by a
high-frequency trader, the trader posts a new limit order on the
opposite side of the book. If a limit buy order is executed at b(η),
the traders posts a limit sell order at a slightly higher price b(η) + ξ
where ξ is modeled as a positive random variable with distribution
g . a limit sell order is executed at a(η), HFTs posts a limit sell order
at a slightly lower price a(η) − ξ. Denoting by q the proportion of
limit orders posted by high-frequency traders, this gives
(iv ) η 7→ η − 1b(η) + 1a(η)+x at rate qµb g (x)
η 7→ η − 1a(η) + 1b(η)−x at rate qµa g (x)
b a
where µ ,µ the rate of arrival of market orders
v Execution of market order against non-HFT limit orders
(v ) η 7→ η − 1b(η) at rate (1 − q)µb
η 7→ η − 1a(η) at rate (1 − q)µa
Limit order markets
The limit order book as a measure-valued Markov process
(ηt , t ≥ 0) is a L-valued Markov process with infinitesimal generator
Af (η) = x≤b(η) λb+ (x, η)[f (η + 1x ) − f (η)] + λb− (x, η)[f (η − 1x ) − f (η)]
P
b
P
+ x<b(η) r (x, η)[f (η − 1x + 1x+1 ) − f (η)]
a
P
+ x>a(η) r (x, η)[f (η + 1x − 1x−1 ) − f (η)]
+ x≥a(η) λa+ (x, η)[f (η − 1x ) − f (η)] + λa− (x, η)[f (η + 1x ) − f (η)]
P
+ (1 − q)µb [f (η − 1b(η) ) − f (η)]

+ (1 − q)µa [f (η + 1a(η) ) − f (η)]
qµb x≥a(η) g (x − a(η)) [f (η − 1b(η) − 1x ) − f (η)]
P
+
qµa x≤b(η) g (b(η) − x) [f (η + 1a(η) − 1x ) − f (η)].
P
+
for any cylindrical function f : L 7→ R.

Limit order markets
High-frequency traders submit orders very frequently, cancel a high

percentage of their orders before execution (up to 95%) and maintain a
low inventory (i.e. they do not accumulate a large number of buy or sell
orders). This is only possible if high frequency traders primarily submit or
cancel orders through procedures (ii) and (iv). Thus, in a market where
order flow is dominated by HFTs, we expect
r b , r a |λa+ − λb− |, |λb+ − λb− | and (1 − q) 1 (7)
We translate this into a scaling regime where:

√ √
r b , r a ∼ N, |λa+ − λb− |, |λb+ − λb− | ∼ N and (1 − q) ∼ 1/ N (8)

Limit order markets

Assumptions on parameters and scaling of initial condition ηN0
1 Assumption 1: the distribution g has compact support.
Interpretation: HFTs post orders near the bid/ask.
2 Assumption 2: centering on initial price ∀N ≥ 1, b(ηN0 ) = S0 .
3 Assumption 3: There exists ρ0,+ , ρ0,− ∈ C (R, R+ ) ∩ L1 (R) with
supp(ρ0,+ ) ⊂ (−∞, 0], supp(ρ0,− ) ⊂ [0, ∞)
and ∀f ∈ CK0 (R− ), ∀g ∈ CK0 (R+ ), ∀ > 0
Z ∞ !
1 X x
N
lim P f ( )η0 (x) − f (u)ρ0,− (u)du ≥ = 0.

N→∞ N N 0
x∈N
Z ∞ !
1 X x
N
lim P g ( )η0 (x) − g (u)ρ0,− (u)du ≥ = 0.

N→∞ N N 0
x∈N

Limit order markets
A fluid limit for the multicomponent limit order book

Assumption (F): market orders, cancels and limit orders balance each
other at scale N: N(µa (x) − λa (x)) → 0, N(µb (x) − λb (x)) → 0.
Theorem: Under the scaling assumptions 1, 2, 3, (F) the
measure-valued process
1 + x x
ηtN = ηt ( ), ηt− ( ) (9)
N N N
converges weakly as N → ∞, in D([0, ∞), M(R)2 ) equipped with the
−
Skorokhod topology, to a measure (µ+ t , µt ) whose density
− +
(ρ (t, x), ρ (t, x), t ≥ 0) is a weak solution of
∂ρ+ σ 2 ∂ 2 ρ+ ∂ρ+
= + 2
+ b+ (t, x) + λρ+ (t, x) for x < St (10)
∂t 2 ∂x ∂x
∂ρ− σ 2 ∂ 2 ρ− ∂ρ−
= − + b − + λρ− (t, x) for x > St (11)
∂t 2 ∂x 2 ∂x
ρ− (t, x) = 0 for x > St , ρ+ (t, x) = 0 for x ≥ St (12)
−
+
1 ∂ρ ∂ρ
Ṡ(t) = (t, St −) − (t, St +) (13)
θ ∂x ∂x
| {z }
Limit order markets
Fluid limit: two phase Stefan problem
In particular we have: ∀t ≥ 0, ∀f ∈ CK0 (R), ∀ > 0

Z ∞ !
1 X x N
lim P | f ( )ηt (x) − f (u)ρ(t, u)du| ≥ = 0.
N→∞ N N 0
x∈N

1 N
lim P | b(ηt ) − St | ≥ = 0.
N→∞ N

Limit order markets
Fluid limit: two phase Stefan problem
This is a ’moving boundary’ problem, known in physics as the Stefan

problem, which captures the average evolution of the limit order
book profile.
The evolution of the price is driven by order flow imbalance as in the
Kyle model.
This is consistent with empirical studies (C., Kukanov, Stoikov
2013) which show evidence for linear impact of small orders on price.
Different from Lasry & Lions (2007) price formation model.
This result provides a micro-foundation for these models and relates
their parameters to arrival rates and variances of order flows.
Price moves are generated purely by the market marker/HFT order
flow: the other agents are pure liquidity providers but their flow is
not directional and perfectly equilibrates at 1/N scale in the fluid
limit.

Limit order markets
Fluid limit: Lasry-Lions (2007) model
Recall that HFTs, once their orders are executed, place a new order of
opposite sign at a distance S(t) + X where X is a random variable with
distribution g . Here we have assumed
g has compact support
We can recover the Lasry-Lions (2007) models if instead we assume
gN (.) = g (./N) weakly converges to δa

This assumption is less natural if the tick/price unit is scaled to zero as
1/N since it would imply a ’macroscropic’ transaction cost (fixed cost)
rather than a proportional cost.
More generally if gN ⇒ G we obtain a variant of the Lasry-Lions (2007)
models with a ’smoothed’ integral source term delocalized over
St ± supp(G ).

Limit order markets
Fluid limit: Lasry-Lions (2007) model

Assume that gN (.) = g (./N) → δa where a > 0. Then under the scaling
assumptions 2, 3 the measure-valued process
1 + x x
ηtN = ηt ( ), ηt− ( ) (14)
N N N
converges weakly as N → ∞, in D([0, ∞), M(R)2 ) equipped with the
−
Skorokhod topology, to a measure (µ+ t , µt ) whose density
− +
(ρ (t, x), ρ (t, x), t ≥ 0) is a weak solution of
∂ρ+ σ 2 ∂ 2 ρ+ ∂ρ−
= + 2
− (t, St )δSt −a for x < St (15)
∂t 2 ∂x ∂x
∂ρ− σ 2 ∂ 2 ρ− ∂ρ+
= − − (t, St )δSt +a for x > St (16)
∂t 2 ∂x 2 ∂x
− +
ρ (t, x) = 0 for x > St , ρ (t, x) = 0 for x ≥ St (17)
with initial condition ρ(0, .) = limN η0N .

Limit order markets
Homogeneous vs inhomogeneous scaling
In this result, as in the previous work on fluid limits of limit order book
models (and all the literature on hydrodynamic limits of particle systems)
we use scaling assumptions that are uniform in space (price variable).
However, it is empirically observed that the intensity of order submissions
and cancellations is an order of magnitude higher at the best bid/ask
price levels.
This can be modeled by assuming a different scaling behavior of
intensities of events at the best price level:
N(µa (x) − λa (x)) → 0, √ N(µb (x) − λb (x)) → 0 for √ x 6= b(η), a(η)
while (µa − λa (0)) ∼ 1/ N, (µb (x) − λb (x)) ∼ 1/ N
at the best bid/ask.
This allows us to take into account temporary (random) imbalances at
the best bid/ask, which is a realistic feature of intraday dynamics of
supply and demand

Limit order markets
A stochastic PDE model for evolution of the limit order

book
Intuitively, if the order flow at the best price levels is much higher than
other price levels, there exists a scaling regime in which fluctuatins vanish
away from the interface but not at the interface.
Then one expects to obtain a stochastic version of the price dynamics:
∂ρ+ σ 2 ∂ 2 ρ+ ∂ρ+
= + + b + (t, x) + λρ+ (t, x) for x < St (18)
∂t 2 ∂x 2 ∂x
∂ρ− σ 2 ∂ 2 ρ− ∂ρ−
= − 2
+ b− + λρ− (t, x) for x > St (19)
∂t 2 ∂x ∂x
ρ+ (t, x) = 0 for x ≥ St , ρ− (t, x) = 0 for x > St (20)
−
+
1 ∂ρ ∂ρ
dSt = (t, St −) − (t, St +) dt + σdWt (21)
θ ∂x ∂x
| {z }
Order flow imbalance

Limit order markets
−
Assume we have a measure-valued process ρt = (ρ+
t , ρt ) with values in
2
M(R) and a process S verifying
∂ρ+ ∂ρ−
= Lρ+ (t, x) for x < St = Lρ− (t, x) for x > St (22)
∂t ∂t
ρ+ (t, x) = 0 for x ≥ St , ρ− (t, x) = 0 for x ≤ St (23)
−
+
1 ∂ρ ∂ρ
θ ∂x ∂x
Then the process ρ is characterized by the property that, for any test
function ϕ ∈ C0∞ ([0, T ] × R),
+
< ρ+
t , ϕ >=< ρ0 , ϕ > +
t
σ2 ∂ρ−
Z
∂ϕ
du < ρ+
u ,( + L∗ ϕ) > + ϕ(u, Su ) (u, Su −)
0 ∂t 2 ∂x
−
< ρ−
t , ϕ >=< ρ0 , ϕ > +
Z t
σ2 ∂ρ−

+ ∂ϕ ∗
du < ρu , ( + L ϕ) > + ϕ(u, Su ) (u, Su −)
0 ∂t 2 ∂x

Limit order markets
Definition: A weak solution on [0, τ ) of

∂ρ±
= Lρ± (t, x), ρ+ (t, x) = 0 for x ≥ St , ρ− (t, x) = 0 for x ≤ St (25)
∂t
1 ∂ρ+ ∂ρ−

θ ∂x ∂x
−
is a pair (ρ, S) where S is a semimartingale and ρ = (ρ+
t , ρt , t ∈ [0, τ ) a
2
measure-valued process in M(R) such that
∀t ≤ τ, ρ+ ([St , ∞) ) = 0, ρ− ((−∞, St ] ) = 0,
∀t ≤ τ, the following limits exist P − a.s.:
ρ+ [St − , St ) ∂ρ+ ρ+ (St , St + ] ∂ρ−
lim = (t, St −) lim = (t, St +)
↓0 ∂x ↓0 ∂x
For any test function ϕ ∈ C0∞ ([0, T ] × R),
Z t
∂ϕ σ2 ∂ρ−
< ρ+t , ϕ >=< ρ+
0 , ϕ > + du < ρ+ u ,( + L∗ ϕ) > + ϕ(u, Su ) (u, Su )
0 ∂t 2 ∂x
Z t
− ∂ϕ σ2 ∂ρ+
< ρ−t , ϕ >=< ρ0 , ϕ > + du < ρ+ u ,( + L∗ ϕ) > + ϕ(u, Su ) (u, Su )
0 ∂t 2 ∂x
Rt + Rama CONT−

Price dynamics in Limit Order Markets:
Limit order markets

√
Assumption
√ (D): imbalance √ of order N at the best bid/ask :
N(µa − λa (0)) → σ 2 , N(µb − λb (0)) → σ 2
Theorem: Under Assumptions 1, 2, 3, (D) ηtN converges weakly as
−
N → ∞ to a measure-valued process (µ+ t , µt ) whose density
− +
(ρ (t, x), ρ (t, x), t ≥ 0) is a weak solution of the stochastic partial
differential equation
∂ρ+ σ 2 ∂ 2 ρ+ ∂ρ+
= + 2
+ b+ (t, x) + λρ+ (t, x) for x < St (27)
∂t 2 ∂x ∂x
− 2 2 − −
∂ρ σ ∂ ρ ∂ρ
= − 2
+ b− + λρ− (t, x) for x > St (28)
∂t 2 ∂x ∂x
∂ρ−
+
1 ∂ρ
θ ∂x ∂x
| {z }
with initial condition ρ(0, .) = limN η0N .

Limit order markets
A stochastic 2-phase Stefan problem

Proposition: the ”stochastic 2-phase Stefan problem”
∂ρ+ σ 2 ∂ 2 ρ+ ∂ρ+
= + 2
+ b+ (t, x) + λρ+ (t, x) for x < St (31)
∂t 2 ∂x ∂x
∂ρ− σ 2 ∂ 2 ρ− ∂ρ−
= − + b − + λρ− (t, x) for x > St (32)
∂t 2 ∂x 2 ∂x
∂ρ−
+
1 ∂ρ
θ ∂x ∂x
| {z }
admits a weak solution (ρ, S) where

ρ is a random field on [0, τ ] × R with continuous sample paths
S is a semimartingale with decomposition (34)
τ = {t > 0, sup(|∆∂x ρ(t, St )|) < ∞}
Limit order markets
Diffusion limit: a stochastic two-phase Stefan problem
Two–phase moving-boundary problem with ”random forcing” at the

boundary: stochastic version of the two-phase Stefan problem.
Similar (but not identical) stochastic moving boundary problem studied
recently by Mueller et al (2013), Keller-Ressel & Müller (2014).
Price follows a ”diffusion in a random environment” defined by the limit
order books:
its drift is generated by price impact of the HFT/market maker order flow
at the bid/ask
its volatility is driven by stochastic imbalance between the market and
limit orders at the best-bid ask.
Perpetual competition between
the random drift term (action of ’market marker’ ) which stabilizes the
price and
the white noise term which prevents the price from settling down.

Limit order markets
References :
Rama Cont, Sasha Stoikov and Rishi Talreja (2010) A stochastic model
for order book dynamics, Operations Research, Volume 58, No. 3,
549-563.
Rama CONT (2011) Statistical modeling of high frequency data: facts,
models and challenges, IEEE Signal Processing, Vol 28, No 5,
16–25.
Rama Cont and Adrien de Larrard (2013) Price dynamics in a Markovian
limit order market, SIAM Journal on Financial Mathematics, Vol 4, 1–25.
Rama Cont and Adrien de Larrard (2011) Order book dynamics in liquid
markets: limit theorems and diffusion approximations,
http://ssrn.com/abstract=1757861.
Rama Cont and Adrien de Larrard (2012) Price dynamics in limit order
markets: linking volatility with order flow, Working Paper.
Rama Cont (2014) High frequency dynamics of limit order markets: an
asymptotic analysis.

Rama Cont 2

Uploaded by

Document Informationclick to expand document information

Copyright:

Available Formats

Rama Cont 2

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Rama Cont 2

Uploaded by

Copyright:

Available Formats

Limit order markets

A limit order book model with heterogeneous order flow

Price dynamics in Limit Order Markets:

Imperial-ETH Workshop on Mathematical Finance, 2015

Rama CONT Price dynamics in Limit Order Markets:

1 Limit order markets

Rama CONT Price dynamics in Limit Order Markets:

At the core of liquidity: the limit order book

Figure: A limit order book.

Rama CONT Price dynamics in Limit Order Markets:

Figure: A limit buy order: Buy 2 at 69200.

Rama CONT Price dynamics in Limit Order Markets:

Figure: A market sell order of 10.

Rama CONT Price dynamics in Limit Order Markets:

Figure: Cancellation of 3 sell orders at 69900.

Rama CONT Price dynamics in Limit Order Markets:

Electronic Limit order markets

The advent of electronic trading has transformed markets and led to

Rama CONT Price dynamics in Limit Order Markets:

Figure: The Flash Crash of May 2010 in the US equity markets.

Rama CONT Price dynamics in Limit Order Markets:

Electronic Limit order markets

Many questions of interest to regulators and market participants:

Rama CONT Price dynamics in Limit Order Markets:

Stochastic models of order book dynamics

Traditional market microstructure theory models the strategic

Limit order books as queueing systems

A limit order book may also be viewed as a system of queues subject to

Rama CONT Price dynamics in Limit Order Markets:

Example: a Markovian limit order book

C. , Stoikov, Talreja (Operations Research, 2010) [CST 2010]

Rama CONT Price dynamics in Limit Order Markets:

The limit order book as a measure-valued process

The state of limit order book may be viewed as a signed measure µ on R:

µ = µ+ − µ− a(µ) = inf (supp(µ− )) ≥ b(µ) = sup (supp(µ+ )) ,

supp(µ+ ) ⊂ (−∞, b(µ)] supp(µ− ) ⊂ [a(µ), ∞)

Rama CONT Price dynamics in Limit Order Markets:

PDE models of price formation

Interpretation: after trading takes place at price St , buyers become sellers

Rama CONT Price dynamics in Limit Order Markets:

PDE models of price formation

Caffarelli, Markowich & Pietsch (2013), Caffarelli, Markowich & Wolfram

Rama CONT Price dynamics in Limit Order Markets:

Relation between modeling approaches

What is the relation between the discrete, stochastic models

Rama CONT Price dynamics in Limit Order Markets:

A hierarchy of time scales

Regime Time scale Issues

Rama CONT Price dynamics in Limit Order Markets:

Moving across time scales: fluid and diffusion limits

Idea: study limit of rescaled limit order book as

Rama CONT Price dynamics in Limit Order Markets:

Moving across time scales: fluid and diffusion limits

Rama CONT Price dynamics in Limit Order Markets:

’Heavy traffic’ asymptotics for limit order books

’Heavy traffic’ asymptotics for limit order books

Which scaling limit is the relevant one for real markets?

Rama CONT Price dynamics in Limit Order Markets:

r b , r a |λa+ − λb− |, |λb+ − λb− | and (1 − q) 1 (7)

and ∀f ∈ CK0 (R− ), ∀g ∈ CK0 (R+ ), ∀ > 0

In particular we have: ∀t ≥ 0, ∀f ∈ CK0 (R), ∀ > 0