algorithmic trading and the market price of …algorithmic trading strategies order placement...
TRANSCRIPT
Algorithmic Tradingand the
Market Price of Liquidity RiskQUANTITATIVE FINANCE
SUMMER CAMP 2016
Christopher Tinghttp://mysmu.edu.sg/faculty/christophert/
Lee Kong Chian School of BusinessSingapore Management University
August 11, 2016
August 11, 2016
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 1 / 56
Flow of Talk
Table of Contents
1 Introduction
2 Algorithmic Trading
3 Optimal Execution
4 Classical Market Microstructure
5 Model of Order Flow
6 A Special Case
7 Liquidity Risk Index: LIX
8 Conclusions
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 2 / 56
Introduction Client-Broker-Market Interaction
In the Past: High-Touch Brokers
Client
Broker
Market
tele
phon
e order tothe
floor Client
Broker
Markette
leph
one
confirmation
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 3 / 56
Introduction Definition of AT
Algorithmic Trading (AT)
DefinitionThe use of programs (algorithms) and computers (automation) togenerate and submit orders in electronic markets.
Origin: U.S. institutional investors in late 1990s needed the toolsto deal with major changes:
electronic markets: electronic communications networksalternative trading systems, dark poolsdecimalization (reduction) of tick sizereduction of commissions and exchange fees
Today, brokers compete actively for the commission poolassociated with algorithmic trading around the globe.
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 4 / 56
Introduction Definition of AT
Characteristics of Algorithmic Trading (AT)
Non-discretionaryPurely procedural or mechanicalEmotionless
AlgorithmicTrading rulesArtificial intelligence, machine learning, cyborg
Data drivenTick-by-tick trades and quotes, news, fundamentals etc.Live, real-time feeds absolutely necessary
AutomaticAutomation: major investment in high-tech IT infrastructure (ultrahigh-frequency, low latency, direct market access, co-location)If the algorithm is not speed-sensitive, automation is not absolutelynecessary.
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 5 / 56
Introduction Electronic Markets
Low-Touch Electronic Market
Client
Broker
Market
tele
phon
e/in
tern
et
algoexecution
Client
Broker
Markette
leph
one/
inte
rnet
confirmation
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 6 / 56
Introduction Electronic Markets
Electronic Market with API Gateway
Client
Broker
Market
Alg
oor
ders
thru
API
Gat
eway
algoexecution Client
Broker
Marketle
ased
line/
inte
rnet
confirmation
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 7 / 56
Introduction Electronic Markets
Direct Market Access (DMA) and Co-loc
Client
Broker
Market
quer
yan
dre
spon
se
queryand
responsealgo orders
No touchCo-location: the tradingalgo sits in a server nextto and connected to theexchange’s matchingengineLow latency
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 8 / 56
Introduction An Example of DMA and Co-Loc
ArrownetSource: JPX
Source: JPXChristopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 9 / 56
Introduction Market Fragmentation
Market Fragmentation: Case Study of Japan
ExchangesJPX Group
First SectionSecond SectionJasdaqMothers
Nagoya Stock ExchangeSapporo Securities ExchangeFukuoka Stock Exchange
Proprietary trading system (PTS)Chi-X JapanSBI Japannext
Exchanges versus PTS: More than one venue to transact the samestock.
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 10 / 56
Introduction Market Trend
Adoption of AT in Asia
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 11 / 56
Algorithmic Trading Building Blocks of AT
Three Building Blocks of AT
Algorithmic Trading Strategies
Order Placement Strategies
Order Routing Strategies
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 12 / 56
Algorithmic Trading Alpha Generation and Brokerage (Agency) Execution
Two Major Categories of AT Strategies
Alpha Generation, Buy Side
Location arbitrageTriangular arbitrageStatistical arbitrage
MomentumMean reversion
Market making
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 13 / 56
Algorithmic Trading Alpha Generation and Brokerage (Agency) Execution
Two Major Categories of AT Strategies (cont’d)
Brokerage Execution, Sell Side
Services rendered by sell side to and for the buy sideTime-weighted average price (TWAP)Volume-weighted average price (VWAP)Percentage of volume (POV)The Almgren-Chriss algorithmLiquidity-seeking
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 14 / 56
Algorithmic Trading Institutional Brokerage Execution
Institutional Brokerage Execution
Since the 2008 financial crisis, “alpha” generation is stronglycurtailed for major investment banks.
Nevertheless, institutional clients need to trade large amounts ofstocks, larger than what the market can absorb without impactingthe price.
To prevent slippage, a large (parent) order must be sliced intomany smaller child orders.
What is the algorithm to bring about an optimal solution?
Providing trade execution services for institutional clients is a BIGbusiness these days.
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 15 / 56
Algorithmic Trading Institutional Brokerage Execution
First-Generation AT for Brokerage Execution
TWAP: Equal amount of shares or contracts in each time interval.
But market’s trading volume typically exhibits a U-shape patternfrom opening to closing.
VWAP: Divide the trading session into 5- to 15-minute interval.Predict intra-day trading volume and price range for each intervalby some analytics. In each time-interval, execute an amountproportional to the predicted volume for that interval.
POV: Target a defined fraction of the actual volume for each timeinterval. The intention is to keep the trading activity in line withtotal volume. Trade at a constant percentage participation rate.
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 16 / 56
Algorithmic Trading Institutional Brokerage Execution
Volume Weighted Average Price
Let pi, vi, i = 1, 2, . . .n be the prices and volumes, respectively forwhich n child orders are filled.Total volume traded is
v =
n∑i=1
vi.
With wi = vi/v being the weight, The VWAP price is, by definition,
Pvwap :=
n∑i=1
wipi =
n∑i=1
vi
vpi =
∑ni=1 pivi
v=
Dollar VolumeVolume
VWAP is the break-even price.
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 17 / 56
Algorithmic Trading VWAP Algorithmic Trading
Objective Function and Approximation
Objective of the algorithm is to minimize the expected error (see[Konishi, 2002]):
minv(t)
E((
VWAP − vwap)2)
Definition: Percentage of remaining volume
X(t) :=M(T)−M(t)
M(T); x(t) :=
v(T)− v(t)v(T)
Change of objective to
minx(t)
E
((∫ T
0
(X(t)− x(t)
)dP(t)
)2)
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 18 / 56
Algorithmic Trading VWAP Algorithmic Trading
A Solution
[Konishi, 2002] shows that an optimal schedule is
x∗(t) = E(X(t)
)± 1
2v(T).
The optimal schedule is to overshoot E(X(t)
)by ±1/(2v(T)),
which is a small amount if the number of shares to trade is large.
It is crucial to forecast the mean X(t) at each time period t, whichis the proxy for E
(X(t)
).
Therefore, you need to forecast the total volume M(T) of themarket, and the dynamics of the market volume M(t)!
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 19 / 56
Optimal Execution Framework of [Almgren and Chriss, 2000]
Second-Generation AT for Brokerage Execution
[Almgren and Chriss, 2000]’s mathematical modelling
Consider liquidating X shares. Let the number of shares yet to beliquidated be xt, t = 0, 1, . . . ,T. So
x0 = X, xT = 0,
andxs ≤ xt if s > t.
Accordingly, −(xt+ε − xt
)is the number of stocks sold in the time
interval (t + ε, t). Hence
xt+ε − xt −→ dxt, as ε→ 0.
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 20 / 56
Optimal Execution Framework of [Almgren and Chriss, 2000]
Almgren-Chriss’ Assumptions
The price St is a drift-less (arithmetic) Brownian motion Bt:
dSt = σ dBt
with constant (intra-day) volatility σ.
The price at which a sale takes place is not St but Pt given by
Pt = St + η vt,
where the rate of trading is
vt ≡ xt :=dxt
dt,
i.e., xt is a (deterministically) differentiable function of t.
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 21 / 56
Optimal Execution Main Idea of [Almgren and Chriss, 2000]
Trading Risk
Trading risk is captured by
V := V(∫ T
0xt dSt
)= σ2
∫ T
0x2
t dt.
Main idea of [Almgren and Chriss, 2000] is to minimize the meanand variance with λ being the Lagrange multiplier:
A := E(∫ T
0Pt vt dt
)+ λV
(∫ T
0xt dSt
)= K + λσ2
∫ T
0x2
t dt
=
∫ T
0
[η
(dxt
dt
)2
+ λσ2x2t
]dt
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 22 / 56
Optimal Execution Analytical Solution for Optimal Execution
Analytical Solution
The Euler-Lagrange equation (principle of least action) is
ddt∂L∂xt− ∂L∂xt
= 0.
For the [Almgren and Chriss, 2000] Lagrangian, i.e.,L = ηx2
t −(− λσ2x2
t), it is
2ηxt − 2λσ2xt = 0,
i.e., xt = κ2xt, where
κ2 =λσ2
η
The solution that satisfies the boundary conditions is
x?t = Xsinh
(κ(T − t)
)sinh(κT)
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 23 / 56
Optimal Execution Analytical Solution for Optimal Execution
Optimal Schedule for Different Urgency κ
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 24 / 56
Optimal Execution Third-Generation AT
Third-Generation: Liquidity-Seeking AT
Multi-period optimization of trade scheduling
Access to all available liquidity, both dark and lit
Inclusion of order placement and order routing algorithms
Urgency specificationLow: dark, passiveMedium: combination of dark and opportunistic participationHigh urgency: high participation rate
Optimization of probability of fill (execution certainty),invisibility (stealth), price improvement (passiveness) trading cost(implementation shortfall), and price certainty (risk) subject toclient’s specification of urgency.
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 25 / 56
Optimal Execution BEAT Systems
BEAT Systems and Business
BrokerageExecution
AlgorithmicTrading
J.P. MorganAqua
KCGCatch
UBSNEO
Credit SuisseAES
GSSONAR
ITGSmart Limit
NomuraSmartSeek
Deutsche BankSuperX
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 26 / 56
Optimal Execution Dark Liquidity
Lit and Unlit Orders
In the past, “upstairs” markets and crossing networks allowedinstitutional block trades to execute.
Dark pools are the results of technological advancement, intensecompetition, and new regulatory requirements.
Most brokers support iceberg orders for lit exchanges, resulting indark liquidity.
Is there a smart way to ping for dark liquidity?
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 27 / 56
Optimal Execution Back to the Real World
ITG’s Classification of Algo Trading Strategies
Close: Trading at the close
Dark: Dark pool and unlit orders
Scheduled: TWAP, VWAP, POV (Participation)
Implementation Shortfall (IS): Inspired by and generalization of[Almgren and Chriss, 2000]
Liquidity-Seeking
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 28 / 56
Optimal Execution Back to the Real World
Preferred BEAT of 20 Buy-Side Desks
[Domowitz and Yegerman, 2011] Sample Period: Q4 2009 to Q4 2010 Q4
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 29 / 56
Optimal Execution Back to the Real World
Order Placement Strategies[Domowitz and Yegerman, 2011] Sample Period: Q4 2009 to Q4 2010 Q4
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 30 / 56
Optimal Execution Back to the Real World
Order Placement Strategies (cont’d)
Demand for liquidity: percentage of median daily volume (MDV)for the parent order
1 Less than (LT) 0.5% of MDV2 0.5% to 1% of MDV3 1% to 5% (and rarely above) of MDV
During the “quiet” period (Q4 2009 to Q4 2010 Q4),Almgren-Chriss’ Implementation Shortfall (IS) strategy was mostpopular.
81% of orders were executed by IS
VWAP is still dominantly used (58% of orders) by some tradingdesks.
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 31 / 56
Optimal Execution Back to the Real World
Trading Strategy Costs By Demand For Liquidity
[Domowitz and Yegerman, 2011] Sample Period: Q4 2009 to Q4 2010 Q4
Close Dark IS Liquidity-Seeking Scheduled
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 32 / 56
Optimal Execution Summary
Summary
BEAT is a big business and R&D of algorithmic trading systemswill continue.
Generalization to account for stochastic liquidity and volatility(see [Almgren, 2012])
Permeation of BEAT systems to FX, futures and options, and fixedincome securities
Asia-Pacific markets still have a lot of room to grow the BEATbusiness.
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 33 / 56
Classical Market Microstructure Roll’s Model of Bid-Ask Spread
Unobservable Price versus Transaction Price
Earlier in Slide 21, we have seen
Pt = St + η vt
A model by [Roll, 1984] of bid-ask spread C from price change∆Pt := Pt − Pt−1:
Pt = St +C2
Qt + εt,
where Qt is either +1 (trade @ the ask price), or −1 (trade @ thebid price), or 0 (trade @ the midpoint).
C = 2√−Cov
[∆Pt , ∆Pt−1
].
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 34 / 56
Classical Market Microstructure Reformulation of Roll’s Model
OLS Approach [Ting, 2014]
Tick-by-tick change of price ∆Pti regressed on change in tradesign ∆Qti and on trade duration ∆ti:
∆Pti = c0 +C2∆Qti + b∆ti + εti
12−10−01 12−10−19 12−11−12 12−12−04 12−12−26 13−01−17 13−02−07 13−03−04 13−03−25 13−04−16 13−05−080
5
10
15
20
25
30
35
40
SG
D p
er C
ontr
act
Flash Crash
OLS EstimateRoll’s Estimate
Missing Roll’s Estimate
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 35 / 56
Classical Market Microstructure Signed Volume Vt is Important!
Structural Models: A Synthesis
Unobservable Price Transaction Price Price Change DynamicsSt = Pt = ∆Pt =
Roll RW, i.e., St−1 + εt St +C2
xtC2∆Qt + εt
Huang & Stoll RW +αC2
Qt−1 Midquote +C2
QtC2∆Qt + (α+ β)
C2
Qt−1 + εt
Madhavan et al. RW + θ(Qt − ρQt−1
)St + φQt (θ + φ)∆Qt
+ θ(1− ρ)Qt−1 + εt
Glosten & Harris RW + a0Qt + a1Vt St + b0Qt + b1Vt (a0 + b0)∆Qt + a0Qt−1
+(a1 + b1)∆Vt + a1Vt−1 + εt
Tee & Ting (2015) RW + θ1(Qt − ρ1Qt−1
)St + φ1 Qt + φ2 vt
(θ1 + φ1
)∆Qt
+ θ2(Vt − ρ2Vt−1
)+ θ1(1− ρ1)Qt−1
+ (θ2 + φ2)∆Vt
+ θ2(1− ρ2)Vt−1 + εt
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 36 / 56
Classical Market Microstructure Signed Volume Vt is Important!
0.5
1.0
1.5
2.0
2.5
3.0dx
Contract = TP
romhgh
−0.2
0.0
0.2
0.4
0.6
0.8
1.0
x
−0.02−0.01
0.000.010.020.030.040.050.060.07
dv
Jul 2012
Oct 2012
Jan 2013
Apr 2013
Jul 2013
Oct 2013
Jan 2014
Apr 2014
Jul 2014
Oct 2014−0.02
0.00
0.02
0.04
0.06
0.08
0.10
v
0.51.01.52.02.53.03.54.04.55.0 Contract = NK
−0.10.00.10.20.30.40.50.60.70.8
−0.02−0.01
0.000.010.020.030.040.050.06
Jul 2012
Oct 2012
Jan 2013
Apr 2013
Jul 2013
Oct 2013
Jan 2014
Apr 2014
Jul 2014
Oct 20140.00
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.81.01.21.41.61.82.02.22.42.6 Contract = NO
−0.050.000.050.100.150.200.250.300.35
−0.006
−0.004
−0.002
0.000
0.002
0.004
0.006
Jul 2012
Oct 2012
Jan 2013
Apr 2013
Jul 2013
Oct 2013
Jan 2014
Apr 2014
Jul 2014
Oct 20140.000
0.005
0.010
0.015
0.020
0.025
0.030
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 37 / 56
Classical Market Microstructure Signed Volume Vt is Important!
Relative Joint Price Impact by Hour
9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 241+
1d2+
1d
0.00
0.05
0.10
0.15
0.20
0.25
0.30
T T+1
RJPI: dx → dx,x,dv,v
TPNKNO
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 38 / 56
Classical Market Microstructure Why Order Flow?
Order Flow and Price Change
According to [Kyle, 1985], the transaction price P depends on theorder flow
buy order X+
sell order X−
[Evans and Lyons, 2002] show that order flow X := X+ − X− is asuperior candidate for modelling the variation in foreignexchange rate compared to arguments along the lines ofmacroeconomics.
Over a period of time t, (e.g. 1 minute, duration of 100 trades, etc),a positive (negative) order flow Xt indicates more (less) buyersthan sellers.
What is the market price of liquidity risk for demandingimmediate execution?
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 39 / 56
Model of Order Flow A Postulate
Order Flow Dynamics
Joint Work with Kijima
Stochastic differential equation for the order flow Xt is assumed tobe a stationary (mean reverting) process:
dXt = c(m− Xt)dt + η dwt, (1)
wherewt is the standard Brownian motion.m is the long-run average.c is the speed of mean-reversion.η is the volatility of order flow.
The parameters m, c, and η are assumed to be constants.
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 40 / 56
Model of Order Flow True Price versus Market Transaction Price
Price Impact
In financial economics, the “true price” St is not directlyobservable.
Let φ(X) be the price impact when the order flow is Xt that relatesthe trade price Pt to the true price St.
Assumption 1: True Price, Trade Price, and Price ImpactFor a certain function f (X),
φ(X) = ef (X) ⇐⇒ f (X) = logφ(X)
andPt = St φ(Xt) ⇐⇒ St = Pt e−f (Xt). (2)
This formalism is similar to [Çetin et al., 2004] and[Huang and Ting, 2008], among others.
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 41 / 56
Model of Order Flow True Price versus Market Transaction Price
Price Impact (cont’d)
Assumption 2: Zero Order Flow Condition
φ(0) = 1 ⇐⇒ f (0) = 0.
Assumption 3: Trade Price DynamicsPt be the time t trade price and
dPt
Pt= µ(Xt)dt + σ(Xt)dzt, (3)
where µ(x) and σ(x) are some smooth functions of order flow x,and where wt and zt are standard Brownian motions withcorrelation ρ
dwtdzt = ρdt.
The trade price Pt as well as the order flow Xt are both observablein the market.
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 42 / 56
Model of Order Flow Two RIsks and Their Market Prices
Market Prices of Risks
Let λzt = λz(Xt) and λw
t = λw(Xt) denote the market prices of risksassociated with zt and wt, respectively.
Definedzt = dzt + λz
t dt, dwt = dwt + λwt dt.
Let Q be a probability measure that makes the processes zt and wtstandard Brownian motions.
True price process becomes
dSt
St=(µS(Xt)− σ(Xt)λz(Xt) + η f ′(Xt)λw(Xt)
)dt
+σ(Xt)dzt − η f ′(Xt)dwt (4)
under Q.
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 43 / 56
Model of Order Flow Two RIsks and Their Market Prices
The Market Price of Liquidity Risk
For some positive function ξ(x) > 0, the market price of liquidityrisk is written as
λw(x) = ξ(x)− η f ′(x). (5)
Let g(x) = f ′(x). The ODE then becomes
{c(m− x) + ρησ(x)− ηξ(x)
}g(x) +
12η2{g′(x) + g2(x)
}= 0. (6)
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 44 / 56
A Special Case Seeking an Explicit Price Impact Function f(x)
A Special Case
Let c = 0 and σ(x) = σ, a constant.
For the market price of risk, suppose it is a constant of x, i.e., theSharpe ratio:
λz(x) =µ− rσ
.
Moreover, for the market price of liquidity risk,
λw(x) = −η f ′(x).
i.e., ξ(x) = 0.
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 45 / 56
A Special Case Seeking an Explicit Price Impact Function f(x)
A Special Case (cont’d)
Let µ(x) = µ− r. Then
r + ρση f ′(x) +12η2[f ′′(x) + (f ′(x))2
]= 0 (7)
Consequently, under the same Q, the true price process becomes
dSt
St= r dt + σ dzt + λw(x)dwt, (8)
The ODE (7) admits a closed-form solution:
f (x) =bη
x + log
(1− A exp
(− βx/η
)1− A
). (9)
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 46 / 56
A Special Case Concave Function of x
Numerical Example of f (x)r = 0, σ = 0.017, η = 10, and ρ = 0.16
x
-500 -400 -300 -200 -100 0 100 200 300 400 500
f(x)
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
A = 0.6
A = 0.2
A = 0.4
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 47 / 56
A Special Case Important Concept
Market Price of Liquidity Risk
In this special case
λw(x) = −η f ′(x) = −
(b +
βAe−βη
x
1− Ae−βη
x
)(10)
If b ≈ 0, then the nonlinear term is dominant.
Market price of liquidity is NEGATIVE. Why?
(Market) order flow x demands liquidity and therefore must paythe price for enjoying the liquidity created by liquidity providers(e.g. traders who submit limit orders to trade).
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 48 / 56
Liquidity Risk Index: LIX Liquidity Risk Index: LIX
Liquidity Risk Index: LIX
Suppose x = 0, i.e., no trade. Then the “Implied bid-ask spread”to be paid by liquidity consumers can be defined as
LIX := 2η f ′(0) = 2(
b +βA
1− A
).
LIX is higher if b, A, and β are larger, intuitive indeed.
“Term structure” of LIX: Time interval to compute the order flowWall clock: 1 minute, 2 minutes, 3 minutes, . . ., 10 minutesBusiness clock: 10 trades, 20 trades, 30 trades, . . ., 100 trades
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 49 / 56
Liquidity Risk Index: LIX Liquidity Risk Index: LIX
On-Shore and Off-Shore Nikkei 225 Index Futures
On-Shore JPX (> 16.5 ms from SGX)
Big Nikkei (NK) at Osaka: tick size 10 points, multiplier U1,000
Mini Nikkei (NO) at Osaka: tick size 5 points, multiplier U100
Off-Shore SGX, CME (> 50 ms from SGX)
Regular Niikei (NI) at SGX: tick size 5 points, multiplier U500
Regular Nikkei (NX) at CME: tick size 5 points, multiplier U500
NI and NX are mutually fungible.
Quanto Nikkei (NH) at CME: tick size 5 points, multiplier $5
Excellent test bed for studying market fragmentation, orderplacement, order routing, LIX, etc
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 50 / 56
Liquidity Risk Index: LIX Liquidity Risk Index: LIX
EU Referendum Vote Counting −→ Brexit
June 24, 2016Ticker TV TOF σ η ρ b0 b A β SSE R2 LIX
10−3 bps 10−3 10−3 10−6 % bps
NH 54,986 -894 2.25 53.20 0.26 -2.31 -6.70 0.8567 1.18 4.87 2.93 6.51NX 29,510 -2,433 2.17 20.61 0.38 -2.12 0.52 0.0013 1.15 4.30 7.96 10.31NI 201,796 -4,625 2.21 158.98 0.64 -2.30 -6.52 0.7223 2.80 3.89 19.39 15.19
NK 174,567 -7,523 2.32 155.63 0.69 -2.33 0.91 0.0005 0.16 4.03 24.51 18.27NO 1,530,375 -53,492 2.35 956.44 0.72 -2.36 -1.10 0.3750 3.65 3.79 30.45 21.85NO5 306,075 -10,698 2.35 191.29 0.72 -2.36 0.49 0.0302 19.3 3.79 30.45 21.79
Volatility of log return is σ.Estimation of b0, b,A, and β by constrained nonlinear least squaresLIX is fairly invariant with different combinations of b,A, and β.
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 51 / 56
Liquidity Risk Index: LIX Liquidity Risk Index: LIX
Time Series of LIX (Nov 5, 2014 to Dec 10, 2015)
Jan 20
15
Mar 2015
May 2
015
Jul 20
15
Sep 2
015
Nov 2
015
0
2
4
6
8
10
12
14
16
18
bps
NI NO
NK
NX
NHNH
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 52 / 56
Liquidity Risk Index: LIX Liquidity Risk Index: LIX
Descriptive Statistics for LIX (Nov 5, 2014 to Dec 10, 2015)
NH NX NI NK NO
mean 2.11 2.47 2.98 3.51 3.63std dev 1.14 1.41 1.73 2.06 2.02
1st 0.67 0.72 1.15 1.09 1.455th 0.86 1.03 1.39 1.50 1.73
10th 1.07 1.22 1.58 1.78 1.9325th 1.38 1.67 1.89 2.25 2.3350th 1.87 2.13 2.49 2.98 2.9975th 2.49 2.94 3.49 4.15 4.2790th 3.31 3.97 4.73 5.56 5.8495th 3.99 4.94 6.04 6.60 6.7699th 6.42 7.15 9.59 11.80 11.13
correlationwith 28.6% 34.2% 43.3% 78.3% 51.4%
volatilityChristopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 53 / 56
Conclusions LIX is Useful
Conclusions
Price impact function is a concave function of order flow.
Market price of risk is negative for liquidity consumers.
LIX is a useful tool to reflect not only the market price ofliquidity risk but also possibly volatility.
Market depth: larger β implies lower depth
Urgency: larger A implies traders are more anxious when theytrade
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 54 / 56
References Thank You for Your Attention!
References
I Almgren, R. (2012).Optimal trading with stochastic liquidity and volatility.SIAM Journal of Financial Mathematics, 3:163–181.
I Almgren, R. and Chriss, N. (2000).Optimal execution of portfolio transactions.Journal of Risk, 3(2):5–39.
I Çetin, U., Jarrow, R. A., and Protter, P. (2004).Liquidity risk and arbitrage pricing theory.Finance and Stochastics, 8(3):311–341.
I Domowitz, I. and Yegerman, H. (2011).Algorithmic trading usage patterns and their costs.ITG Analytics Incubator.
I Evans, M. D. D. and Lyons, R. K. (2002).Order flow and exchange rate dynamics.Journal of Political Economy, 110:170–180.
I Huang, R. and Ting, C. (2008).A functional approach to the price impact of stock trades and the implied true price.Journal of Empirical Finance, 15:1–16.
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 55 / 56
References Thank You for Your Attention!
References (cont’d)
I Konishi, H. (2002).Optimal slice of a {VWAP} trade.Journal of Financial Markets, 5(2):197–221.
I Kyle, A. S. (1985).Continuous auctions and insider trading.Econometrica, 53:1315–1335.
I Roll, R. (1984).A simple implicit measure of the effective bid-ask spread in an efficient market.Journal of Finance, 39:1127–1139.
I Ting, C. (2014).A common measure of liquidity costs for futures and stock exchanges.In Lee, D. and Gregoriou, G. N., editors, Handbook of Asian Finance: REITs, Trading, and FundPerformance, volume 2, pages 201–213. Elsevier Academic Press, Massachusetts, USA.
Christopher Ting (Lee Kong Chian School of Business Singapore Management University [1em] August 11, 2016 [6em] )August 11, 2016 56 / 56