You are on page 1of 56

Algorithmic Trading

and the
Market Price of Liquidity Risk
Q UANTITATIVE F INANCE
S UMMER C AMP 2016

Christopher Ting
http://mysmu.edu.sg/faculty/christophert/

Lee Kong Chian School of Business
Singapore Management University

August 11, 2016

Christopher Ting 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 August 11, 2016 2 / 56

Introduction Client-Broker-Market Interaction

In the Past: High-Touch Brokers

Broker Broker
ne

ne
ord
pho

pho

con
er to
tele

tele

firm
the

atio
floo

n
Client Client
r

Market Market

Christopher Ting August 11, 2016 3 / 56

Introduction Definition of AT

Algorithmic Trading (AT)

Definition
The use of programs (algorithms) and computers (automation) to
generate and submit orders in electronic markets.

Origin: U.S. institutional investors in late 1990s needed the tools
to deal with major changes:
electronic markets: electronic communications networks
alternative trading systems, dark pools
decimalization (reduction) of tick size
reduction of commissions and exchange fees

Today, brokers compete actively for the commission pool
associated with algorithmic trading around the globe.

Christopher Ting August 11, 2016 4 / 56

Introduction Definition of AT

Characteristics of Algorithmic Trading (AT)
Non-discretionary
Purely procedural or mechanical
Emotionless

Algorithmic
Trading rules
Artificial intelligence, machine learning, cyborg

Data driven
Tick-by-tick trades and quotes, news, fundamentals etc.
Live, real-time feeds absolutely necessary

Automatic
Automation: major investment in high-tech IT infrastructure (ultra
high-frequency, low latency, direct market access, co-location)
If the algorithm is not speed-sensitive, automation is not absolutely
necessary.
Christopher Ting August 11, 2016 5 / 56

2016 6 / 56 . Introduction Electronic Markets Low-Touch Electronic Market Broker Broker t et rne ern nte int algo i con ne/ ne/ firm exe pho pho cuti atio tele tele on n Client Client Market Market Christopher Ting August 11.

2016 7 / 56 . Introduction Electronic Markets Electronic Market with API Gateway ay Broker Broker tew t rne Ga e I int AP algo con / u line thr firm exe sed ers at cuti ord lea ion on Client o Client Alg Market Market Christopher Ting August 11.

2016 8 / 56 . Introduction Electronic Markets Direct Market Access (DMA) and Co-loc e Broker ons No touch que p res Co-location: the trading ry a and algo sits in a server next n d re to and connected to the ry que spo exchange’s matching Client nse engine algo Low latency ord ers Market Christopher Ting August 11.

2016 9 / 56 . Introduction An Example of DMA and Co-Loc Arrownet Source: JPX Christopher Ting Source: JPX August 11.

Christopher Ting August 11. Introduction Market Fragmentation Market Fragmentation: Case Study of Japan Exchanges JPX Group First Section Second Section Jasdaq Mothers Nagoya Stock Exchange Sapporo Securities Exchange Fukuoka Stock Exchange Proprietary trading system (PTS) Chi-X Japan SBI Japannext Exchanges versus PTS: More than one venue to transact the same stock. 2016 10 / 56 .

2016 11 / 56 . Introduction Market Trend Adoption of AT in Asia Christopher Ting August 11.

2016 12 / 56 . Algorithmic Trading Building Blocks of AT Three Building Blocks of AT Order Routing Strategies Order Placement Strategies Algorithmic Trading Strategies Christopher Ting August 11.

2016 13 / 56 . Buy Side Location arbitrage Triangular arbitrage Statistical arbitrage Momentum Mean reversion Market making Christopher Ting August 11. Algorithmic Trading Alpha Generation and Brokerage (Agency) Execution Two Major Categories of AT Strategies Alpha Generation.

2016 14 / 56 . Sell Side Services rendered by sell side to and for the buy side Time-weighted average price (TWAP) Volume-weighted average price (VWAP) Percentage of volume (POV) The Almgren-Chriss algorithm Liquidity-seeking Christopher Ting August 11. Algorithmic Trading Alpha Generation and Brokerage (Agency) Execution Two Major Categories of AT Strategies (cont’d) Brokerage Execution.

To prevent slippage. a large (parent) order must be sliced into many smaller child orders. larger than what the market can absorb without impacting the price. institutional clients need to trade large amounts of stocks. What is the algorithm to bring about an optimal solution? Providing trade execution services for institutional clients is a BIG business these days. Algorithmic Trading Institutional Brokerage Execution Institutional Brokerage Execution Since the 2008 financial crisis. Nevertheless. 2016 15 / 56 . Christopher Ting August 11. “alpha” generation is strongly curtailed for major investment banks.

Predict intra-day trading volume and price range for each interval by some analytics. Trade at a constant percentage participation rate. In each time-interval. VWAP: Divide the trading session into 5. But market’s trading volume typically exhibits a U-shape pattern from opening to closing. Algorithmic Trading Institutional Brokerage Execution First-Generation AT for Brokerage Execution TWAP: Equal amount of shares or contracts in each time interval. execute an amount proportional to the predicted volume for that interval. POV: Target a defined fraction of the actual volume for each time interval. Christopher Ting August 11.to 15-minute interval. The intention is to keep the trading activity in line with total volume. 2016 16 / 56 .

Total volume traded is n X v= vi . i = 1. n be the prices and volumes. i=1 With wi = vi /v being the weight. . 2016 17 / 56 . Algorithmic Trading Institutional Brokerage Execution Volume Weighted Average Price Let pi . . n n Pn vi i=1 pi vi Dollar Volume X X Pvwap := wi pi = pi = = v v Volume i=1 i=1 VWAP is the break-even price. . by definition. The VWAP price is. vi . Christopher Ting August 11. respectively for which n child orders are filled. 2.

Algorithmic Trading VWAP Algorithmic Trading Objective Function and Approximation Objective of the algorithm is to minimize the expected error (see [Konishi. 2016 18 / 56 . 2002]):  2  min E VWAP − vwap v(t) Definition: Percentage of remaining volume M(T) − M(t) v(T) − v(t) X(t) := . x(t) := M(T) v(T) Change of objective to Z T 2 !  min E X(t) − x(t) dP(t) x(t) 0 Christopher Ting August 11.

2002] shows that an optimal schedule is 1 x∗ (t) = E X(t) ±  . and the dynamics of the market volume M(t)! Christopher Ting August 11. you need to forecast the total volume M(T) of the market. which is a small amount if the number of shares to trade is large. Algorithmic Trading VWAP Algorithmic Trading A Solution [Konishi. which It is crucial to forecast the is the proxy for E X(t) . 2016 19 / 56 . 2v(T)  The optimal schedule is to overshoot E X(t) by ±1/(2v(T)). Therefore.  mean X(t) at each time period t.

Christopher Ting August 11. . So x0 = X. 2000]’s mathematical modelling Consider liquidating X shares. . t = 0. as  → 0. T.  Accordingly. Let the number of shares yet to be liquidated be xt . Optimal Execution Framework of [Almgren and Chriss. and xs ≤ xt if s > t. 2000] Second-Generation AT for Brokerage Execution [Almgren and Chriss. . Hence xt+ − xt −→ dxt . t). 2016 20 / 56 . 1. xT = 0. − xt+ − xt is the number of stocks sold in the time interval (t + . .

2000] Almgren-Chriss’ Assumptions The price St is a drift-less (arithmetic) Brownian motion Bt : dSt = σ dBt with constant (intra-day) volatility σ.e. Optimal Execution Framework of [Almgren and Chriss.. Christopher Ting August 11. The price at which a sale takes place is not St but Pt given by Pt = St + η vt . dt i. where the rate of trading is dxt vt ≡ ẋt := . xt is a (deterministically) differentiable function of t. 2016 21 / 56 .

Optimal Execution Main Idea of [Almgren and Chriss. 2000] is to minimize the mean and variance with λ being the Lagrange multiplier: Z T  Z T  A := E Pt vt dt + λV xt dSt 0 0 Z T = K + λσ 2 x2t dt 0 Z "  T 2 # dxt = η + λσ 2 x2t dt 0 dt Christopher Ting August 11. 2016 22 / 56 . 2000] Trading Risk Trading risk is captured by Z T  Z T 2 V := V xt dSt = σ x2t dt. 0 0 Main idea of [Almgren and Chriss.

2 2 2 L = η ẋt − − λσ xt . dt ∂ ẋt ∂xt For the [Almgren and Chriss. i. it is 2η ẍt − 2λσ 2 xt = 0.e.e. Optimal Execution Analytical Solution for Optimal Execution Analytical Solution The Euler-Lagrange equation (principle of least action) is d ∂L ∂L − = 0. 2016 23 / 56 . 2000] Lagrangian. ẍt = κ2 xt .. i.. where λσ 2 κ2 = η The solution that satisfies the boundary conditions is  sinh κ(T − t) x?t = X sinh(κT) Christopher Ting August 11.

Optimal Execution Analytical Solution for Optimal Execution Optimal Schedule for Different Urgency κ Christopher Ting August 11. 2016 24 / 56 .

and price certainty (risk) subject to client’s specification of urgency. 2016 25 / 56 . invisibility (stealth). Christopher Ting August 11. both dark and lit Inclusion of order placement and order routing algorithms Urgency specification Low: dark. Optimal Execution Third-Generation AT Third-Generation: Liquidity-Seeking AT Multi-period optimization of trade scheduling Access to all available liquidity. price improvement (passiveness) trading cost (implementation shortfall). passive Medium: combination of dark and opportunistic participation High urgency: high participation rate Optimization of probability of fill (execution certainty).

P. Optimal Execution BEAT Systems BEAT Systems and Business KCG Catch UBS J. 2016 26 / 56 . Morgan NEO Aqua Brokerage Credit Suisse Execution Deutsche Bank AES Algorithmic SuperX Trading GS Nomura SONAR SmartSeek ITG Smart Limit Christopher Ting August 11.

intense competition. Optimal Execution Dark Liquidity Lit and Unlit Orders In the past. resulting in dark liquidity. and new regulatory requirements. Most brokers support iceberg orders for lit exchanges. 2016 27 / 56 . Is there a smart way to ping for dark liquidity? Christopher Ting August 11. “upstairs” markets and crossing networks allowed institutional block trades to execute. Dark pools are the results of technological advancement.

VWAP. 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. 2000] Liquidity-Seeking Christopher Ting August 11. 2016 28 / 56 . POV (Participation) Implementation Shortfall (IS): Inspired by and generalization of [Almgren and Chriss.

2011] Sample Period: Q4 2009 to Q4 2010 Q4 Christopher Ting August 11. 2016 29 / 56 . Optimal Execution Back to the Real World Preferred BEAT of 20 Buy-Side Desks [Domowitz and Yegerman.

Optimal Execution Back to the Real World Order Placement Strategies [Domowitz and Yegerman. 2016 30 / 56 . 2011] Sample Period: Q4 2009 to Q4 2010 Q4 Christopher Ting August 11.

5% of MDV 2 0. Christopher Ting August 11. Almgren-Chriss’ Implementation Shortfall (IS) strategy was most popular.5% to 1% of MDV 3 1% to 5% (and rarely above) of MDV During the “quiet” period (Q4 2009 to Q4 2010 Q4). 81% of orders were executed by IS VWAP is still dominantly used (58% of orders) by some trading desks. 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. 2016 31 / 56 .

2011] Sample Period: Q4 2009 to Q4 2010 Q4 Close Dark IS Liquidity-Seeking Scheduled Christopher Ting August 11. 2016 32 / 56 . Optimal Execution Back to the Real World Trading Strategy Costs By Demand For Liquidity [Domowitz and Yegerman.

2012]) Permeation of BEAT systems to FX. 2016 33 / 56 . and fixed income securities Asia-Pacific markets still have a lot of room to grow the BEAT business. futures and options. Optimal Execution Summary Summary BEAT is a big business and R&D of algorithmic trading systems will continue. Generalization to account for stochastic liquidity and volatility (see [Almgren. Christopher Ting August 11.

1984] of bid-ask spread C from price change ∆Pt := Pt − Pt−1 : C Pt = St + Qt + εt . ∆Pt−1 . or 0 (trade @ the midpoint). 2016 34 / 56 . 2 where Qt is either +1 (trade @ the ask price). Classical Market Microstructure Roll’s Model of Bid-Ask Spread Unobservable Price versus Transaction Price Earlier in Slide 21. q   C = 2 −Cov ∆Pt . or −1 (trade @ the bid price). Christopher Ting August 11. we have seen Pt = St + η vt A model by [Roll.

2016 35 / 56 . 2014] Tick-by-tick change of price ∆Pti regressed on change in trade sign ∆Qti and on trade duration ∆ti : C ∆Pti = c0 + ∆Qti + b ∆ti + ti 40 2 OLS Estimate Roll’s Estimate 35 Flash Crash 30 Missing Roll’s Estimate 25 SGD per Contract 20 15 10 5 0 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 13−05−08 Christopher Ting August 11. Classical Market Microstructure Reformulation of Roll’s Model OLS Approach [Ting.

. Classical Market Microstructure Signed Volume Vt is Important! Structural Models: A Synthesis Unobservable Price Transaction Price Price Change Dynamics St = Pt = ∆Pt = C C Roll RW. i. 2016 36 / 56 . St−1 + t St + xt ∆Qt + t 2 2 αC C C C Huang & Stoll RW + Qt−1 Midquote + Qt ∆Qt + (α + β) Qt−1 + t 2 2 2 2  Madhavan et al. RW + θ Qt − ρQt−1 St + φQt (θ + φ)∆Qt + θ(1 − ρ)Qt−1 + t Glosten & Harris RW + a0 Qt + a1 Vt St + b0 Qt + b1 Vt (a0 + b0 )∆Qt + a0 Qt−1 +(a1 + b1 )∆Vt + a1 Vt−1 + t   Tee & Ting (2015) RW + θ1 Qt − ρ1 Qt−1 St + φ1 Qt + φ2 vt θ1 + φ1 ∆Qt  + θ2 Vt − ρ2 Vt−1 + θ1 (1 − ρ1 )Qt−1 + (θ2 + φ2 )∆Vt + θ2 (1 − ρ2 )Vt−1 + t Christopher Ting August 11.e.

0 0.5 1.00 −0.5 1.03 0.15 0.015 v 0.10 0.4 2.5 0.4 0.20 0.5 0.0 1.002 0.004 0.8 0.2 0.8 dx 1.0 2.00 −0.0 0.8 1.0 1.6 Contract = NO ro 4.0 gh 3.6 0.025 0.0 Contract = NK 2.02 0.02 −0.4 0.010 0.25 0.04 0.030 0.3 x 0.2 0.5 0.7 0.07 0.01 −0.00 0.05 0. 2016 37 / 56 .08 0. Classical Market Microstructure Signed Volume Vt is Important! 3.01 0.5 2.03 0.00 0.0 0.000 dv 0.000 2 2 3 3 3 3 4 4 4 4 2 2 3 3 3 3 4 4 4 4 2 2 3 3 3 3 4 4 4 4 201 201 201 201 201 201 201 201 201 201 201 201 201 201 201 201 201 201 201 201 201 201 201 201 201 201 201 201 201 201 Jul Oct Jan Apr Jul Oct Jan Apr Jul Oct Jul Oct Jan Apr Jul Oct Jan Apr Jul Oct Jul Oct Jan Apr Jul Oct Jan Apr Jul Oct Christopher Ting August 11.1 −0.05 0.02 0.4 1.0 1.02 −0.5 2.5 4.04 0.6 0.006 0.6 2.02 0.10 0.8 0.002 0.01 −0.01 0.2 1.05 0.02 0.04 0.04 0.005 −0.1 0.05 0.00 0.01 −0.5 2.35 0.02 0.06 0.0 1.0 0.004 −0.05 0.0 Contract = TP 5.03 0.06 0.020 0.07 0.0 2.06 0.2 −0.30 0.06 0.006 0.2 mh 3.

05 0.00 11 12 21 15 13 17 10 18 19 16 14 22 1d 23 20 9 24 1d 1+ 2+ Christopher Ting August 11.v TP NK 0.dv.15 0.10 0.20 T T+1 0.25 NO 0. Classical Market Microstructure Signed Volume Vt is Important! Relative Joint Price Impact by Hour 0.30 RJPI: dx → dx.x. 2016 38 / 56 .

the transaction price P depends on the order flow buy order X+ sell order X− [Evans and Lyons. (e. What is the market price of liquidity risk for demanding immediate execution? Christopher Ting August 11. 1 minute. Classical Market Microstructure Why Order Flow? Order Flow and Price Change According to [Kyle. duration of 100 trades. a positive (negative) order flow Xt indicates more (less) buyers than sellers. 2016 39 / 56 . etc).g. 2002] show that order flow X := X+ − X− is a superior candidate for modelling the variation in foreign exchange rate compared to arguments along the lines of macroeconomics. Over a period of time t. 1985].

and η are assumed to be constants. η is the volatility of order flow. The parameters m. (1) where wt is the standard Brownian motion. Model of Order Flow A Postulate Order Flow Dynamics Joint Work with Kijima Stochastic differential equation for the order flow Xt is assumed to be a stationary (mean reverting) process: dXt = c(m − Xt )dt + η dwt . c. Christopher Ting August 11. c is the speed of mean-reversion. m is the long-run average. 2016 40 / 56 .

among others. 2016 41 / 56 . the “true price” St is not directly observable.. and Price Impact For a certain function f (X). 2008]. Christopher Ting August 11. (2) This formalism is similar to [Çetin et al. 2004] and [Huang and Ting. Assumption 1: True Price. Let φ(X) be the price impact when the order flow is Xt that relates the trade price Pt to the true price St . Trade Price. φ(X) = ef (X) ⇐⇒ f (X) = log φ(X) and Pt = St φ(Xt ) ⇐⇒ St = Pt e−f (Xt ) . Model of Order Flow True Price versus Market Transaction Price Price Impact In financial economics.

and where wt and zt are standard Brownian motions with correlation ρ dwt dzt = ρ dt. Assumption 3: Trade Price Dynamics Pt be the time t trade price and dPt = µ(Xt )dt + σ(Xt )dzt . 2016 42 / 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. Christopher Ting August 11. (3) Pt where µ(x) and σ(x) are some smooth functions of order flow x. The trade price Pt as well as the order flow Xt are both observable in the market.

e t = dwt + λw dw t dt. respectively. Christopher Ting August 11. 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 risks associated with zt and wt . True price process becomes dSt   = µS (Xt ) − σ(Xt )λz (Xt ) + η f 0 (Xt )λw (Xt ) dt St +σ(Xt )dezt − η f 0 (Xt )dw et (4) under Q. Let Q be a probability measure that makes the processes ezt and w et standard Brownian motions. Define dezt = dzt + λzt dt. 2016 43 / 56 .

2016 44 / 56 . Model of Order Flow Two RIsks and Their Market Prices The Market Price of Liquidity Risk For some positive function ξ(x) > 0.  (6) 2 Christopher Ting August 11. (5) Let g(x) = f 0 (x). the market price of liquidity risk is written as λw (x) = ξ(x) − η f 0 (x). The ODE then becomes 1  c(m − x) + ρησ(x) − ηξ(x) g(x) + η 2 g0 (x) + g2 (x) = 0.

e. λw (x) = −η f 0 (x). For the market price of risk. i.. Christopher Ting August 11.. a constant. ξ(x) = 0. the Sharpe ratio: µ−r λz (x) = . 2016 45 / 56 . A Special Case Seeking an Explicit Price Impact Function f (x) A Special Case Let c = 0 and σ(x) = σ. suppose it is a constant of x.e. i. σ Moreover. for the market price of liquidity risk.

A Special Case Seeking an Explicit Price Impact Function f (x) A Special Case (cont’d) Let µ(x) = µ − r. (9) η 1−A Christopher Ting August 11. under the same Q. 2016 46 / 56 . (8) St The ODE (7) admits a closed-form solution: ! b 1 − A exp − βx/η f (x) = x + log . Then 1 h i r + ρση f 0 (x) + η 2 f 00 (x) + (f 0 (x))2 = 0 (7) 2 Consequently. the true price process becomes dSt = r dt + σ dzt + λw (x)dwt .

4 -0. A Special Case Concave Function of x Numerical Example of f (x) r = 0.6 -0.8 -500 -400 -300 -200 -100 0 100 200 300 400 500 x Christopher Ting August 11. σ = 0.4 -0. η = 10.2 A = 0. and ρ = 0.2 0 A = 0.16 0.6 A = 0.4 0.017.2 f(x) -0. 2016 47 / 56 .

Christopher Ting August 11. A Special Case Important Concept Market Price of Liquidity Risk In this special case − βη x ! 0 βAe λw (x) = −η f (x) = − b + (10) − βη x 1 − Ae If b ≈ 0. Why? (Market) order flow x demands liquidity and therefore must pay the price for enjoying the liquidity created by liquidity providers (e. then the nonlinear term is dominant. Market price of liquidity is NEGATIVE.g. 2016 48 / 56 . traders who submit limit orders to trade).

20 trades. 10 minutes Business clock: 10 trades. 100 trades Christopher Ting August 11. no trade. . i.e. and β are larger. Then the “Implied bid-ask spread” to be paid by liquidity consumers can be defined as   0 βA LIX := 2η f (0) = 2 b + . 3 minutes. 2 minutes.. . . “Term structure” of LIX: Time interval to compute the order flow Wall clock: 1 minute. intuitive indeed. A. 2016 49 / 56 . Liquidity Risk Index: LIX Liquidity Risk Index: LIX Liquidity Risk Index: LIX Suppose x = 0. .. 30 trades.. . 1−A LIX is higher if b. .

Liquidity Risk Index: LIX Liquidity Risk Index: LIX On-Shore and Off-Shore Nikkei 225 Index Futures On-Shore JPX (> 16. multiplier U500 NI and NX are mutually fungible. CME (> 50 ms from SGX) Regular Niikei (NI) at SGX: tick size 5 points. order routing. 2016 50 / 56 . etc Christopher Ting August 11. multiplier U500 Regular Nikkei (NX) at CME: tick size 5 points. multiplier U100 Off-Shore SGX. Quanto Nikkei (NH) at CME: tick size 5 points. LIX.5 ms from SGX) Big Nikkei (NK) at Osaka: tick size 10 points. multiplier $5 Excellent test bed for studying market fragmentation. multiplier U1.000 Mini Nikkei (NO) at Osaka: tick size 5 points. order placement.

65 3.72 -2.35 956.31 -6. 2016 51 / 56 .91 0.0013 1.12 0.63 0.0005 0.96 10. A.35 191.433 2.61 0.52 0.986 -894 2. A.10 0.3750 3.15 4.17 20.19 NK 174.52 0.64 -2.27 NO 1.87 2.69 -2.32 155.80 3.31 NI 201.30 -6.36 0. Liquidity Risk Index: LIX Liquidity Risk Index: LIX EU Referendum Vote Counting −→ Brexit June 24.93 6.375 -53. 2016 Ticker TV TOF σ η ρ b0 b A β SSE R2 LIX 10−3 bps 10−3 10−3 10−6 % bps NH 54.36 -1.8567 1. Christopher Ting August 11.492 2.25 53.72 -2.45 21.30 7.85 NO5 306. Estimation of b0 .49 0.51 NX 29.79 30.89 19.29 0.698 2.79 Volatility of log return is σ.44 0.567 -7. b.530.98 0.625 2. and β.03 24.510 -2.45 21.38 -2.39 15.3 3.70 0.51 18.075 -10.26 -2.523 2.79 30.7223 2.16 4.18 4. and β by constrained nonlinear least squares LIX is fairly invariant with different combinations of b.33 0.796 -4.20 0.0302 19.21 158.

Liquidity Risk Index: LIX Liquidity Risk Index: LIX Time Series of LIX (Nov 5. 2014 to Dec 10. 2015) 18 NI NX NO 16 14 12 10 NK NH bps 8 6 4 2 0 2 015 20 15 2 015 20 15 201 5 2 015 Jan Mar May Jul Sep Nov Christopher Ting August 11. 2016 52 / 56 .

49 2.11 2.41 1.4% volatility Christopher Ting August 11.89 2.59 11.22 1.49 2.27 90th 3.06 2.98 3. 2015) NH NX NI NK NO mean 2.13 2.15 9.31 3.33 50th 1.42 7.04 6.63 std dev 1.73 5.13 correlation with 28.3% 78.15 1.6% 34.58 1. Liquidity Risk Index: LIX Liquidity Risk Index: LIX Descriptive Statistics for LIX (Nov 5.56 5.93 25th 1.38 1.94 3.51 3.3% 51.84 95th 3.09 1.73 2.99 75th 2.14 1.76 99th 6.99 4.80 11.86 1.94 6.73 10th 1.72 1.03 1.78 1.87 2.15 4.25 2.02 1st 0.2% 43.47 2.50 1.67 1. 2014 to Dec 10.07 1.97 4.60 6.98 2.67 0. 2016 53 / 56 .39 1.45 5th 0.49 4.

Market depth: larger β implies lower depth Urgency: larger A implies traders are more anxious when they trade Christopher Ting August 11. 2016 54 / 56 . Market price of risk is negative for liquidity consumers. LIX is a useful tool to reflect not only the market price of liquidity risk but also possibly volatility. Conclusions LIX is Useful Conclusions Price impact function is a concave function of order flow.

(2011). A. Jarrow. P. A functional approach to the price impact of stock trades and the implied true price. R. 110:170–180. and Chriss.. ITG Analytics Incubator. and Yegerman. (2004). (2012). D. U.. 2016 55 / 56 . Journal of Risk. I Huang. SIAM Journal of Financial Mathematics. Journal of Empirical Finance. R. Christopher Ting August 11. 8(3):311–341. References Thank You for Your Attention! References I Almgren. Finance and Stochastics. 3:163–181. I Evans. Liquidity risk and arbitrage pricing theory. Order flow and exchange rate dynamics. R. Optimal execution of portfolio transactions. R. (2000). Algorithmic trading usage patterns and their costs. K. and Protter. N. (2002). D. Optimal trading with stochastic liquidity and volatility. C. I Çetin. and Ting. I Domowitz. I Almgren. 3(2):5–39. R. H. Journal of Political Economy. (2008). 15:1–16. and Lyons. I. M.

(2014). A common measure of liquidity costs for futures and stock exchanges. In Lee. C. H. Trading. R. (2002). References Thank You for Your Attention! References (cont’d) I Konishi.. I Kyle. A. and Gregoriou. Massachusetts. I Roll. pages 201–213. Optimal slice of a {VWAP} trade. 53:1315–1335. Journal of Finance. Elsevier Academic Press. D. I Ting. Christopher Ting August 11. A simple implicit measure of the effective bid-ask spread in an efficient market. 5(2):197–221. 2016 56 / 56 . Journal of Financial Markets. G. (1984). editors. S. Continuous auctions and insider trading. USA. Handbook of Asian Finance: REITs. volume 2. N. and Fund Performance. 39:1127–1139. (1985). Econometrica.