Low-Frequency Traders in a High-Frequency world:

A Survival Guide

Marcos López de Prado

Lawrence Berkeley National Laboratory
Computational Research Division

Electronic
Electronic copycopy availableat:
available at: https://ssrn.com/abstract=2150876
http://ssrn.com/abstract=2150876
 Key Points
• Multiple empirical studies have shown that Order Flow
Imbalance has explanatory power over the trading range.
• The PIN Theory (Easley et al. [1996]) reveals the
Microstructure mechanism by which
– Market Makers adjust their trading range to avoid being
adversely selected by Informed Traders.
– Informed Traders reveal their future trading intentions
when they alter the Order Flow.
– Consequently, Market Makers’ trading range is a function
of the Order Flow imbalance.
• VPIN is a High Frequency estimate of PIN, which can be
used to detect the presence of Informed Traders.
2
Electronic
Electronic copycopy availableat:
available at: https://ssrn.com/abstract=2150876
http://ssrn.com/abstract=2150876
 SECTION I
The great divide

Electronic
Electronic copycopy availableat:
available at: https://ssrn.com/abstract=2150876
http://ssrn.com/abstract=2150876
 Is speed the real issue?

• Faster traders are nothing new:

– Nathan Rothschild is said to have used racing pigeons to
trade in advance on the news of Napoleon’s defeat at
Waterloo.
– Beginning in 1850s, only a limited number of investors had
access to telegraphy.
– The telephone (1875), radio (1915), and more recently
screen trading (1986) offered speed advantages to some
participants over others.
– Leinweber [2009] relates many instances in which
technological breakthroughs have been used to most
investors’ disadvantage. So … what is new this time?
4
Electronic copy available at: https://ssrn.com/abstract=2150876
 A change in paradigm

• High Frequency Trading (HFT) is not Low Frequency

Trading (LFT) on steroids.
• HFT have been mischaracterized as ‘cheetah-traders’.
• Rather than speed, the true great divide is a “change
in the trading paradigm”.
• HFT are strategic traders. In some instances, they:
– act upon the information revealed by LFT’s actions.
– engage in sequential games.
– behave like predators.
• Speed is an advantage, but there is more to it…
5
Electronic copy available at: https://ssrn.com/abstract=2150876
 What is the new paradigm? (1/3)

• Time is a measuring system used to sequence

observations.
• Since the dawn of time, humans have based their
time measurements in chronology: Years, months,
days, hours, minutes, seconds, and since recently
milliseconds, microseconds ...
• This is a rather arbitrary time system, due to the key
role played by the Sun in agricultural societies.

6
Electronic copy available at: https://ssrn.com/abstract=2150876
 What is the new paradigm? (2/3)
• Machines operate on an internal clock that is not
chrono based, but event based: The cycle.
• A machine will complete a cycle at various chrono
rates, depending on the amount of information
involved in a particular instruction.
• As it happens, HFT relies on machines, thus
measuring time in terms of events.
• Thinking in volume-time is challenging for us
humans. But for a ‘silicon trader’, it is the natural way
to process information and engage in sequential,
strategic trading.
7
Electronic copy available at: https://ssrn.com/abstract=2150876
 What is the new paradigm? (3/3)

• The new paradigm is “event-based time”. The

simplest example is dividing the session in equal
volume buckets. This transformation removes most
intra-session seasonal effects.
• For example, HF market makers may target to turn
their portfolio every fixed number of contracts
traded (volume bucket), regardless of the chrono
time.
• In fact, working in volume time presents significant
statistical advantages.
8
Electronic copy available at: https://ssrn.com/abstract=2150876
 Volume time vs. Chrono time
Stats (50) Chrono time Volume time Stats (100) Chrono time Volume time
Mean 0.0000 0.0000 Mean 0.0000 0.0000
StDev 1.0000 1.0000 StDev 1.0000 1.0000
Skew -0.0788 -0.2451 Skew -0.1606 -0.4808
Kurt 31.7060 15.8957 Kurt 44.6755 23.8651
Min -21.8589 -20.6117 Min -28.3796 -29.2058
Max 19.3092 13.8079 Max 24.6700 15.5882
L-B* 34.4551 22.7802 L-B* 115.3207 36.1189
White* 0.0971 0.0548 White* 0.0873 0.0370
J-B* 34.3359 6.9392 J-B* 72.3729 18.1782
0.25

0.2
Sampling by Volume
time allows for a
0.15
partial recovery of
Normality, IID
0.1

0.05

0
-5 -4 -3 -2 -1 0 1 2 3 4 5

Time clock Volume clock Normal Dist (same bins as Time clock)

9
Electronic copy available at: https://ssrn.com/abstract=2150876
 SECTION II
High Frequency and Adverse Selection

Electronic copy available at: https://ssrn.com/abstract=2150876

 Little known species you should be aware of
• Predatory algorithms are a special kind of informed
traders. Rather than possessing exogenous information
yet to be incorporated in the market price, they know
that their endogenous actions are likely to trigger a
microstructure mechanism, with foreseeable outcome.
Examples include:
– Quote stuffing: Overwhelming an exchange with messages, with
the sole intention of slowing down competing algorithms.
– Quote dangling: Sending quotes that force a squeezed trader to
chase a price against her interests.
– Pack hunting: Predators hunting independently become aware
of each others activities, and form a pack in order to maximize
the chances of triggering a cascading effect.
11
Electronic copy available at: https://ssrn.com/abstract=2150876
 Slow chess may be harder than you think (1/2)
• O’Hara [2011] presents evidence of their disruptive activities.

• A quote dangler forcing a desperate trader to chase a price

up. As soon as the trader gives up, the dangler quotes back
at the original level, and waits for the next victim.
12
Electronic copy available at: https://ssrn.com/abstract=2150876
 Slow chess may be harder than you think (2/2)
• NANEX [2011] shows what appears to be pack hunters forcing
a stop loss.

• Speed makes HFTs more effective, but slowing them down

won’t change their basic behavior: Strategic sequential
trading.
13
Electronic copy available at: https://ssrn.com/abstract=2150876
 The PIN Theory
Easley & O’Hara [1996] PIN estimates the probability that
market makers are being adversely
selected (i.e., provide liquidity to an
informed trader).
E[ Si | t ]  Pn (t ) Si*  Pb (t ) Si  Pg (t ) Si
 Pb (t )
B ( t )  E [ Si | t ]   E [ Si | t ]  S i 
   Pb (t )
 Pg (t )
A(t )  E[ Si | t ]   S  E[ Si | t ]
   Pg (t )  i
 Pg (t )  Pb (t )
(t )   Si  E[ Si | t ]   E[Si | t ]  Si 
   Pg (t )    Pb (t )

1 𝛼𝜇
𝑖𝑓 𝛿 = ⇒Σ= 𝑆 − 𝑆𝑖
2 𝛼𝜇 + 2𝜀 𝑖

PIN 
  2
14
Electronic copy available at: https://ssrn.com/abstract=2150876
 Estimating PIN in High Frequency
• Suppose that we divide the market activity in n volume buckets of
equal size V. We can index these buckets as 𝜏 = 1, … , 𝑛.
• Let 𝑉𝜏𝐵 be the proportion of volume in a volume bucket 𝜏 associated
with buying pressure, and 𝑉𝜏𝑆 associated with selling pressure.
• We know from Easley, Engle, O’Hara and Wu (2008) that the
expected arrival rate of informed trades is E 𝑉𝜏𝑆 − 𝑉𝜏𝐵 =
𝛼𝜇 2𝛿 − 1 , and E 𝑉𝜏𝑆 − 𝑉𝜏𝐵 ≈ 𝛼𝜇. The expected arrival rate of
total trade is
1 n
 
n  1

VB  VS  V   1                  1          2
         
volumefrom up event volumefrom down event volumefrom no event

• From the values computed above, we can derive the Volume-

Synchronized Probability of Informed Trading (VPIN) as
𝑛 𝑆 − 𝑉𝐵
𝛼𝜇 𝛼𝜇 𝑉
𝜏=1 𝜏 𝜏
𝑃𝐼𝑁 = = ≈ 𝑉𝑃𝐼𝑁 =
𝛼𝜇 + 2𝜀 𝑉 𝑛𝑉

15
Electronic copy available at: https://ssrn.com/abstract=2150876
 Bulk Volume Classification
𝑉
• For each volume bucket 𝜏, we can form J volume bars of size .
𝐽
• For each bar j, T% of the volume is classified as buy and (1-T)%
as sell (denoted “bulk classification”). Caution: Not all the
volume of a single trade or bar is classified as buy or sell (some
researchers are confused by this). Then:
𝐽
𝑉 𝑃𝜏,𝑗 − 𝑃𝜏,𝑗−1
𝑉𝜏𝐵 = 𝑇 , 𝑑𝑓
𝐽 𝜎∆𝑃
𝑗=1
𝐽
𝑆
1 𝑃𝜏,𝑗 − 𝑃𝜏,𝑗−1
𝑉𝜏 = 𝑉 1 − 𝑇 , 𝑑𝑓 = 𝑉 − 𝑉𝜏𝐵
𝐽 𝜎∆𝑃
𝑗=1

where 𝑃𝜏,𝑗 is the last price in bar j within bucket 𝜏, T is the CDF of
the t-distribution with df degrees of freedom, and 𝜎∆𝑃 is the
estimate of the standard derivation of price changes between bars.
16
Electronic copy available at: https://ssrn.com/abstract=2150876
 Bulk Volume Classification vs. Tick Rule (1/4)
• The Tick Rule (TR) and the Bulk Volume Classification
(BVC) algorithms have different goals:
– TR attempts to classify trades as buy-initiated or sell-initiated.
– BVC determines the proportion of volume associated with
buying or selling pressure.
• TR was designed for a time when most informed traders
were aggressors.
• With the advent of high frequency, informed traders are
increasingly relying on limit orders.
• A critical advantage of BVC is that it incorporates:
– Buying (selling) pressure from orders resting in the bid (ask).
– Buying (selling) pressure from cancellations in the ask (bid).
17
Electronic copy available at: https://ssrn.com/abstract=2150876
 Bulk Volume Classification vs. Tick Rule (2/4)

• Market makers adjust to order imbalances, so BVC and

TR should have explanatory power over high-low ranges.
• Let’s define:
𝑉𝜏𝐵 −𝑉𝜏𝑆 𝑉𝜏𝐵
 𝑂𝐼𝜏 ≡ = 2 − 1 is the estimated order imbalance.
𝑉𝜏 𝑉𝜏
 𝐻𝜏 − 𝐿𝜏 is the difference between high and low in volume
bucket 𝜏.
• Then, we can fit the following regression model to 𝑂𝐼𝜏
derived from BVC and TR, and apply the Newey-West
HAC correction:
𝐻𝜏 − 𝐿𝜏 = 𝛽0 + 𝛽1 𝐻𝜏−1 − 𝐿𝜏−1 + 𝛾 𝑂𝐼𝜏 + 𝜉𝜏
18
Electronic copy available at: https://ssrn.com/abstract=2150876
 Bulk Volume Classification vs. Tick Rule (3/4)
Regression Stats for BVC on WTI Regression Stats for TR on WTI
Vol. Bar aR2 NW lags Coeff(α0) Coeff(α1) Coeff(γ) t-Stat(α0) t-Stat(α1) t-Stat(γ) Vol. Bar aR2 NW lags Coeff(α0) Coeff(α1) Coeff(γ) t-Stat(α0) t-Stat(α1) t-Stat(γ)
1000 0.4170 17 5.8920 0.3143 37.8563 36.9490 43.8899 99.0193 1000 0.1971 17 12.7006 0.4174 -5.2172 70.4226 46.7589 -25.5985
2000 0.4656 14 7.5671 0.3310 53.1076 26.6550 35.0893 74.2852 2000 0.2110 14 15.3334 0.4558 -2.1625 48.6918 39.7110 -4.5423
3000 0.5045 13 7.9809 0.3560 65.7965 19.3087 33.5315 67.0455 3000 0.2414 13 16.5738 0.4927 2.2671 37.1431 36.6620 2.6547
4000 0.5124 12 8.8928 0.3554 76.2373 18.0799 31.1926 58.1366 4000 0.2451 12 18.3786 0.4968 6.0838 34.2202 35.5162 4.8603
5000 0.5186 12 9.4361 0.3648 84.7154 13.8771 25.2215 53.9255 5000 0.2514 12 19.7551 0.5032 10.6620 25.9718 27.8923 6.3465
6000 0.5317 11 9.7246 0.3716 93.9735 13.1009 25.3969 49.2206 6000 0.2634 11 20.5196 0.5134 17.4270 24.2252 28.7296 7.4789
7000 0.5332 11 9.9700 0.3771 101.8469 11.4000 24.0834 46.4617 7000 0.2618 11 22.2337 0.5119 19.3449 22.7484 26.9841 6.9339
8000 0.5319 10 10.5324 0.3711 110.4512 11.2616 23.1319 40.9419 8000 0.2558 10 23.7416 0.5047 24.6784 21.0508 24.6193 6.8123
9000 0.5311 10 11.1319 0.3641 119.0141 10.5247 21.5767 40.1135 9000 0.2524 10 25.2300 0.5026 28.3805 20.9909 24.1256 6.9782
10000 0.5351 10 11.5727 0.3657 124.8904 10.0351 21.5811 37.8392 10000 0.2445 10 26.9771 0.4928 30.7460 19.5195 21.7657 6.3642

• BVC’s estimation of Order Imbalance has significant

explanatory power over high-low ranges (Note: It would be
even better with a power specification).
• TR’s Order Imbalance has inconsistent explanatory power
(note the inconsistent signs associated with TR)
• Question: Why does Aggressor-Side Imbalance fail to explain
the trading range?
19
Electronic copy available at: https://ssrn.com/abstract=2150876
 Bulk Volume Classification vs. Tick Rule (4/4)
• Answer: When an informed trader slices and sequentially
executes her buy order passively, sell-initiated trades coexist
with her persistent buy order flow. Informed traders are not
necessarily aggressive traders, thus Aggressor Side-
Imbalance is a deficient estimator of Order Imbalance.
High-Low range and BVC’s OI High-Low range and TR’s OI

20
Electronic copy available at: https://ssrn.com/abstract=2150876
 Does the PIN Theory work in practice?

• Multiple empirical microstructure studies have found

that order flow imbalance impacts trading ranges (e.g.,
Eisler et al. [2012])
• VPIN formalizes that empirical finding by providing the
theoretical connection between order flow imbalance
( 𝑉𝜏𝑆 − 𝑉𝜏𝐵 ) and the range at which market makers
provide liquidity (Σ).
• Through VPIN, we can apply the PIN theory to study:
– Bid-ask dynamics and liquidity crises.
– Toxicity-induced volatility.
– Transaction cost functions and execution strategies.

21
Electronic copy available at: https://ssrn.com/abstract=2150876
 E-mini S&P500 futures on 05/06/10
1 60000 By 11:56am, the realized
0.9
value of the VPIN metric
59000 was in the 10% tail of the
0.8
distribution (it exceeded a
0.7
58000
90% CDF(VPIN) critical
0.6
value). By 1:08pm, the
57000
realized value of VPIN was

Market Value
Probability

0.5
in the 5% tail of the
56000
0.4 distribution (over a 95%
0.3
55000
CDF(VPIN)). At 2:32pm the
crash begins according to
0.2

54000
the CFTC-SEC Report time
0.1
line. Link to video.
0 53000
5/6/10 2:31
5/6/10 5:28
5/6/10 8:27
5/6/10 9:16
5/6/10 9:40
5/6/10 9:55

5/6/10 13:56

5/6/10 15:19
5/6/10 10:18
5/6/10 10:35
5/6/10 10:52
5/6/10 11:07
5/6/10 11:18
5/6/10 11:36
5/6/10 11:51
5/6/10 12:12
5/6/10 12:40
5/6/10 13:12
5/6/10 13:29

5/6/10 14:09
5/6/10 14:17
5/6/10 14:24
5/6/10 14:34
5/6/10 14:39
5/6/10 14:43
5/6/10 14:45
5/6/10 14:48
5/6/10 14:52
5/6/10 14:56
5/6/10 15:03
5/6/10 15:09

5/6/10 15:29
5/6/10 15:41
5/6/10 15:49
5/6/10 15:57
5/6/10 16:02
5/6/10 19:37
Note: The May 6th 2010
‘Flash Crash’ is just one of
VPIN CDF(VPIN) Market value
hundreds of liquidity
events explained by VPIN!
22
Electronic copy available at: https://ssrn.com/abstract=2150876
 Price

4
4.2
4.4
4.6
4.8
5
5.2
5.4
5.6
5.8
7/30/12 9:31
7/30/12 16:03
7/30/12 16:25
7/31/12 9:50
7/31/12 15:15
8/1/12 9:30
8/1/12 9:32
8/1/12 9:35
8/1/12 9:36
8/1/12 9:40
8/1/12 9:42
8/1/12 9:44
8/1/12 9:45
8/1/12 9:47
8/1/12 9:48
8/1/12 9:49
8/1/12 9:50

ARC US Price
8/1/12 9:53
8/1/12 9:54
8/1/12 9:55
8/1/12 9:56
8/1/12 9:57
8/1/12 9:58

CDF(VPIN)
8/1/12 10:02
8/1/12 10:15
8/1/12 10:37
8/1/12 10:53
8/1/12 11:57
8/1/12 14:22
8/1/12 15:01
8/1/12 15:14
8/1/12 15:45
8/1/12 15:59
8/1/12 16:58
8/1/12 17:13
0
1

0.1
0.2
0.3
0.5
0.6
0.7
0.8
0.9

0.4

CDF(VPIN)

Price
10
10.5
11
11.5
12
12.5
13

7/30/12 9:33
7/30/12 10:15
7/30/12 11:10
7/30/12 12:18
7/30/12 13:26
7/30/12 14:20
7/30/12 15:26
7/30/12 15:59
7/31/12 9:52
7/31/12 10:04
7/31/12 10:13
7/31/12 10:21
7/31/12 10:44
7/31/12 11:06

Electronic copy available at: https://ssrn.com/abstract=2150876

7/31/12 11:39
7/31/12 12:27
7/31/12 13:31
7/31/12 14:09
GT US Price

platforms should have picked this up and pulled orders automatically.

7/31/12 14:32
7/31/12 15:08
7/31/12 15:30
The “Knight-mare” of 08/01/12

7/31/12 15:50
7/31/12 17:12
8/1/12 9:37
CDF(VPIN)

8/1/12 9:46
8/1/12 9:53
8/1/12 10:00
8/1/12 10:11
8/1/12 10:38
8/1/12 11:26
8/1/12 12:36
8/1/12 13:45
8/1/12 14:36
8/1/12 14:54
jumped, but the relevant piece is that the price jump occurred as a result of persistent
order imbalance. It was the result of overwhelming and uninterrupted buying pressure

8/1/12 15:10
both cases, CDF(VPIN) jumps to high levels within a few minutes of the open. Prices also

8/1/12 15:48
8/1/12 18:42
Trades for ARC US (American Reprographics) were cancelled, not for GT US (Goodyear). In

23
0
1

(which lasted for 44 minutes), rather than a price adjustment to new information. Knight’s
0.1
0.2
0.3
0.5
0.6
0.7
0.8
0.9

0.4

CDF(VPIN)
 SECTION III
Forecasting (and understanding) Volatility

Electronic copy available at: https://ssrn.com/abstract=2150876

 Forecasting Toxicity-induced volatility (1/4)
• An event e occurs every time that 𝐶𝐷𝐹 𝑉𝑃𝐼𝑁 𝜏 ≥ 𝐶𝐷𝐹 ∗
while 𝐶𝐷𝐹 𝑉𝑃𝐼𝑁 𝜏 − 1 < 𝐶𝐷𝐹 ∗ . We can index those
events as 𝑒 = 1, … , 𝐸, and record the volume bucket at
which 𝐶𝐷𝐹 𝑉𝑃𝐼𝑁 𝜏 crossed the threshold 𝐶𝐷𝐹 ∗ as 𝜏 𝑒

• For each particular e, Event Horizon h(e) is defined as

𝑃𝜏 𝑒 +ℎ1
ℎ 𝑒 = ℎ0 𝑒 , ℎ1 𝑒 = 𝑎𝑟𝑔 max −1
0≤ℎ0 <ℎ1 𝑃𝜏 𝑒 +ℎ
0
1≤ℎ1 ≤𝐵𝑝𝐷

• Similarly, Maximum Intermediate Return MIR(e) is defined

𝑃𝜏 𝑒 +ℎ1 𝑒
𝑀𝐼𝑅 𝑒 = −1
𝑃𝜏 𝑒 +ℎ0 𝑒

25
Electronic copy available at: https://ssrn.com/abstract=2150876
 Forecasting Toxicity-induced volatility (2/4)
0.45
We have computed
two distributions of
0.4 probability: One for
MIRs following an
0.35
event e (in blue), and
0.3
another one for MIRs
at random starts (in
red).
Frequency

0.25

0.2
Following an event e,
0.15 most MIR (blue) fall
within one of the two
0.1
tails of the
0.05
unconstrained
distribution (red). High
0 volatility occurred after
-0.1 -0.05 0 0.05 0.1 0.15
MIR VPIN crossed the
designated threshold
26
Electronic copy available at: https://ssrn.com/abstract=2150876
 Forecasting Toxicity-induced volatility (3/4)
0.04 This qq-plot
shows that both
distributions are
0.02
clearly different:
VPIN events are
0 not random and
-0.06 -0.05 -0.04 -0.03 -0.02 -0.01 0 0.01 0.02
indeed have
consequences in
-0.02
terms of non-
standard MIR).
-0.04
This is consistent
with most (blue)
-0.06 𝑀𝐼𝑅 𝑒 falling at
the tails of
unconstrained
-0.08
MIR (red).

27
Electronic copy available at: https://ssrn.com/abstract=2150876
 Forecasting Toxicity-induced volatility (4/4)
BDP D Events eMean eStd uMean uStd KS_Stat KS_CDF
25 0.2 239 0.02819 0.01628 0.01860 0.011903 0.19171 1.00000
25 0.5 144 0.03149 0.01597 0.01860 0.011903 0.24984 1.00000
25 1 81 0.03554 0.01727 0.01860 0.011903 0.31130 1.00000
50 0.2 257 0.03051 0.01716 0.01951 0.012073 0.24169 1.00000
50 0.5 124 0.03392 0.01681 0.01951 0.012073 0.26478 1.00000
50 1 73 0.03499 0.01570 0.01951 0.012073 0.32145 1.00000
75 0.2 241 0.03166 0.01707 0.01969 0.012319 0.25774 1.00000
75 0.5 121 0.03552 0.01759 0.01969 0.012319 0.26242 1.00000
75 1 64 0.03761 0.01713 0.01969 0.012319 0.32163 1.00000
100 0.2 244 0.03127 0.01667 0.02010 0.012236 0.19684 1.00000
100 0.5 142 0.03470 0.01809 0.02010 0.012236 0.26960 1.00000
100 1 88 0.03912 0.01913 0.02010 0.012236 0.32815 1.00000

The intermediate returns that follow high VPIN events have a mean (eMean) that is
up to 100% greater than the intermediate returns at random events (uMean). The
table above compares those means and standard deviations (eStd, uStd), for the
Events identified through various combinations of Buckets per Day (BPD) and Days
of Sample (D).

28
Electronic copy available at: https://ssrn.com/abstract=2150876
 SECTION IV
What can Low Frequency Traders do?

Electronic copy available at: https://ssrn.com/abstract=2150876

 If you cannot defeat them… (1/5)
• Volume-time is a particular case of “subordinated
stochastic process”, which can be traced back to
Mandelbrot and Clark’s work in the early 70s.
• Any concentration of information per unit of trading
is susceptible of being recognized and taken
advantage from. We have seen this with TWAP algos
and round-number orders, but there are many more
examples.
• Part of HFT’s success is due to the reluctance of LFT
to adopt their paradigm. LFT Choice #1: Where
possible, adopt the HFT paradigm.
30
Electronic copy available at: https://ssrn.com/abstract=2150876
 If you cannot defeat them… (2/5)
• There is some evidence that “big data” is not
necessarily an advantage in all instances.
• For example, Easley et al. [2012b] show that “bulk
volume classification” determines the aggressor side
of a trade with greater accuracy than the tick rule
applied on tick data!
• The same authors show that low-frequency statistics
(like VPIN) can detect the presence of informed
traders and determine the optimal trading horizon.
• LFT Choice #2: Develop statistics to monitor HFT
activity and take advantage of their weaknesses.
31
Electronic copy available at: https://ssrn.com/abstract=2150876
 If you cannot defeat them… (3/5)
• Over 50% of the trades on Index Futures in 2011
were for 1 contract. Trades of 100 contracts are 17
times more frequent than trades of size 99 or 101.
HFT algos can easily detect
when there is a human in
the trading room, and take
advantage. We have seen
that the cost of not
concealing trading
intentions could be up to
40% of a trade’s profit target
(forecasted alpha).
• LFT Choice #3: Use “smart algos”, specialized in
searching for liquidity and avoiding a footprint.
32
Electronic copy available at: https://ssrn.com/abstract=2150876
 If you cannot defeat them… (4/5)
• Participation rate strategies do not take into account
the market’s state, leaving an identifiable footprint.
Comparison of the
probabilistic loss
values for the OEH
algorithm versus a
scheme that
participates in 5% of
the volume.
𝑣 𝐵 = 0.4.

• LFT Choice #4: Do not target a participation rate.

Instead, determine the optimal execution that fits
the prevalent market conditions. 33
Electronic copy available at: https://ssrn.com/abstract=2150876
 If you cannot defeat them… (5/5)
• Toxic order flow disrupts the liquidity provision
process by adversely selecting market makers.
• An exchange that prevents such disruptions will
attract further liquidity, which in turn increases the
corporate value of its products.
• A way to avoid disruptions is to make harder for
predators to operate in that exchange.
• LFT Choice #5: Trade in exchanges that incorporate
smart circuit breakers and matching engines.

34
Electronic copy available at: https://ssrn.com/abstract=2150876
 Conclusions
• Orders from informed traders generate a persistent
order flow imbalance.
• Market makers adjust their trading range accordingly, in
order to avoid adverse selection.
• Market makers operate in a Volume Clock, an are
particularly susceptible to imbalances in that frequency.
• The key to optimal execution is to minimize the
footprint of your trades on the order flow in volume
clock.
• The Optimal Execution Horizon (OEH) algorithm
determines the amount of volume needed to conceal
the intentions of an informed trader.
35
Electronic copy available at: https://ssrn.com/abstract=2150876
THANKS FOR YOUR ATTENTION!

36
Electronic copy available at: https://ssrn.com/abstract=2150876
 Bibliography (1/2)
• Almgren, R. and N. Chriss (2000): “Optimal Execution of Portfolio Transactions”,
Journal of Risk (3), 5-39.
• Almgren, R. and G. Burghardt (2011): “A window into the world of futures market
liquidity”, CME Group Research Note, March 30th.
• Easley, D., Kiefer, N., O’Hara, M. and J. Paperman (1996): "Liquidity, Information,
and Infrequently Traded Stocks", Journal of Finance, September.
• Easley, D., R. F. Engle, M. O’Hara and L. Wu (2008): “Time-Varying Arrival Rates of
Informed and Uninformed Traders”, Journal of Financial Econometrics.
• Easley, D., M. López de Prado and M. O’Hara (2011a): “The Microstructure of the
Flash Crash”, The Journal of Portfolio Management, Vol. 37, No. 2, Winter, 118-
128. http://ssrn.com/abstract=1695041
• Easley, D., M. López de Prado and M. O’Hara (2011b): “The Exchange of Flow
Toxicity”, The Journal of Trading, Vol. 6, No. 2, Spring, 8-13.
http://ssrn.com/abstract=1748633
• Easley, D., M. López de Prado and M. O’Hara (2012a): “Flow Toxicity and Liquidity
in a High Frequency World”, Review of Financial Studies, Vol. 25 (5), pp. 1457-
1493: http://ssrn.com/abstract=1695596
37
Electronic copy available at: https://ssrn.com/abstract=2150876
 Bibliography (2/2)
• Easley, D., M. López de Prado and M. O’Hara (2012b): “Bulk Volume Classification”,
Working paper: http://ssrn.com/abstract=1989555
• Easley, D., M. López de Prado and M. O’Hara (2012c): “The Volume Clock: Insights
into the High Frequency paradigm”, Journal of Portfolio Management
(forthcoming). http://ssrn.com/abstract=2034858
• Easley, D., M. López de Prado and M. O’Hara (2012d): “Optimal Execution Horizon”,
Working paper: http://ssrn.com/abstract=2038387
• Eisler, Zoltan, J.-P. Bouchaud and J. Kockelkoren (2012): “The Impact of order book
events: Market orders, limit orders and cancellations”, Quantitative Finance, 12(9),
1395-1419. Available at http://ssrn.com/abstract=1373762
• Leinweber, D. (2009): “Nerds on Wall Street: Math, Machines and Wired Markets”,
Wiley.
• López de Prado, M. (2011): “Advances in High Frequency Strategies”, Ed.
Complutense University. http://tinyurl.com/hfpin
• NANEX (2011): “Strange Days June 8'th, 2011 - NatGas Algo”,
http://www.nanex.net/StrangeDays/06082011.html
• O’Hara, M. (2011): “What is a quote?”, Journal of Trading, Spring, 10-15.
38
Electronic copy available at: https://ssrn.com/abstract=2150876
 Bio
Marcos López de Prado is Senior Managing Director at Guggenheim Partners. He is also a Research
Affiliate at Lawrence Berkeley National Laboratory's Computational Research Division (U.S. Department
of Energy’s Office of Science).

Before that, Marcos was Head of Quantitative Trading & Research at Hess Energy Trading Company (the
trading arm of Hess Corporation, a Fortune 100 company) and Head of Global Quantitative Research at
Tudor Investment Corporation. In addition to his 15+ years of trading and investment management
experience at some of the largest corporations, he has received several academic appointments,
including Postdoctoral Research Fellow of RCC at Harvard University and Visiting Scholar at Cornell
University. Marcos earned a Ph.D. in Financial Economics (2003), a second Ph.D. in Mathematical
Finance (2011) from Complutense University, is a recipient of the National Award for Excellence in
Academic Performance by the Government of Spain (National Valedictorian, 1998) among other awards,
and was admitted into American Mensa with a perfect test score.

Marcos is the co-inventor of four international patent applications on High Frequency Trading. He has
collaborated with ~30 leading academics, resulting in some of the most read papers in Finance (SSRN),
three textbooks, publications in the top Mathematical Finance journals, etc. Marcos has an Erdös #3 and
an Einstein #4 according to the American Mathematical Society.

39
Electronic copy available at: https://ssrn.com/abstract=2150876
 Notice:

The research contained in this presentation is the result of

a continuing collaboration with

Prof. Maureen O’Hara

Prof. David Easley

The full paper is available at:

http://ssrn.com/abstract=1695596

For additional details, please visit:

http://ssrn.com/author=434076
www.QuantResearch.info

Electronic copy available at: https://ssrn.com/abstract=2150876

 Disclaimer

• The views expressed in this document are the authors’

and do not necessarily reflect those of the
organizations he is affiliated with.
• No investment decision or particular course of action is
recommended by this presentation.
• All Rights Reserved.

41
Electronic copy available at: https://ssrn.com/abstract=2150876