Shades of Darkness:
A Pecking Order of Trading Venues
Albert J. Menkveld (VU University Amsterdam)
Bart Zhou Yueshen (INSEAD)
Haoxiang Zhu (MIT Sloan)
May 2015
Second SEC Annual Conference on the Regulation of Financial Markets
Menkveld-Yueshen-Zhu
Shades of Darkness: A Pecking Order of Trading Venues
1
Motivation
Dark (off-exchange) venues account for a large fraction of volume.
European Indices
60
30
55
Dark (off−book) market share (%)
Dark market share (%)
U.S. (Dow 30)
35
25
20
15
10
5
0
2006
FTSE100 stocks
CAC40 stocks
DAX30 stocks
50
45
40
35
30
25
2008
Menkveld-Yueshen-Zhu
2010
2012
2014
20
2008
2009
2010
Shades of Darkness: A Pecking Order of Trading Venues
2011
2012
2013
2014
2
Dark fragmentation
U.S. has 18 stock exchanges, ∼50 dark pools, >200 broker-dealers
Theory: “liquidity begets liquidity” versus “investor self-selection”
“Fragmentation can inhibit the interaction of investor orders and
thereby impair certain efficiencies and the best execution of investors
orders. . . .On the other hand, mandating the consolidation of order
flow in a single venue would create a monopoly and thereby lose the
important benefits of competition among markets. The benefits of such
competition include incentives for trading centers to create new
products, provide high quality trading services that meet the needs of
investors, and keep trading fees low.” —SEC (2010)
We analyze the dynamic fragmentation of U.S. equity markets
Menkveld-Yueshen-Zhu
Shades of Darkness: A Pecking Order of Trading Venues
3
How do we think about fragmentation
Dark Venue Fragmentation
Do venues behave differently?
NO
Consolidation is better
NO
Back to drawing board
YES
Can we explain this difference?
YES
Rationale for fragmentation
Additional implications
Menkveld-Yueshen-Zhu
Shades of Darkness: A Pecking Order of Trading Venues
4
What we find
Unique data: disaggregated U.S. dark volume
Dark Venue Fragmentation
Volume share under urgency shocks (VIX and earnings)
Do venues behave differently?
NO
Consolidation is better
NO
Back to drawing board YES
Can we explain this difference?
YES
A “pecking order” of trading venues
Rationale for fragmentation
Dark venues help reduce investor costs (with calibration)
Additional implications
Menkveld-Yueshen-Zhu
Shades of Darkness: A Pecking Order of Trading Venues
5
Literature on dark venues
Empirical studies
Aggregate dark
O’Hara-Ye (2011)
Hatheway-Kwan-Zheng (2014)
Degryse-de Jong-van Kervel (2014)
Selected dark venues
Hendershott-Jones (2005)
Ready (2014)
Buti-Rindi-Werner (2011)
Boni-Brown-Leach (2012)
Nimalendran-Ray (2014)
Foley-Malinova-Park (2013)
Dark heterogeneity
Comerton-Forde-Putnins (2015)
Foley-Putnins (2014)
Kwan-Masulis-McInish (2014)
Tuttle (2014)
Degryse-Tombeur-Wuyts (2015)
Country
Dark data source
U.S.
U.S.
Netherlands
All TRF
All TRF
All off-exchange
U.S.
U.S.
U.S.
U.S.
U.S.
Canada
Island ECN
Liquidnet, POSIT
11 anonymous dark pool
Liquidnet
One anonymous dark pool
Dark order on TSX
Australia
Canada
U.S.
U.S.
Netherlands
“block” and “non-block dark” on ASX
“dark midpoint” and “dark limit orders”
5 categories that differ from ours
ATS and non-ATS
“hidden order” and “dark venues”
Theory: Hendershott-Mendelson (2000); Degryse-Van Achter-Wuyts (2009); Ye (2011);
Boulatov-George (2013); Buti-Rindi-Werner (2014); Zhu (2014)
Experimental: Bloomfield-O’Hara-Saar (2013)
Menkveld-Yueshen-Zhu
Shades of Darkness: A Pecking Order of Trading Venues
6
1
Pecking Order Hypothesis
2
Data and Econometric Model
3
Venue Pecking Order in the Data
4
A Suggestive Model and Welfare Implications
Menkveld-Yueshen-Zhu
Shades of Darkness: A Pecking Order of Trading Venues
7
Pecking order hypothesis: generic form
We conjecture that investors “sort” venues by cost and immediacy,
along a “pecking order”
Trading activity moves down if demand for immediacy goes up
Low Cost
Low Immediacy
Venue
Type 1
Investor
Order
Flow
Venue
Type 2
Venue
Type n
High Cost
High Immediacy
Menkveld-Yueshen-Zhu
Shades of Darkness: A Pecking Order of Trading Venues
8
Pecking order hypothesis: specific form
Given the recent advance in theories of dark pools, we conjecture the
specific sorting:
Low Cost
Low Immediacy
DarkMid
Investor
Order
Flow
DarkNMid
Lit
High Cost
High Immediacy
Menkveld-Yueshen-Zhu
Shades of Darkness: A Pecking Order of Trading Venues
9
1
Pecking Order Hypothesis
2
Data and Econometric Model
3
Venue Pecking Order in the Data
4
A Suggestive Model and Welfare Implications
Menkveld-Yueshen-Zhu
Shades of Darkness: A Pecking Order of Trading Venues
10
Data
21 trading days in October 2010
A stratified sample of 117 stocks (the same stocks as the 120 stocks
in the “Nasdaq HFT data”)
Five types of dark venues, disaggregated from Nasdaq TRF
I
I
Nasdaq TRF has about 92% of all TRF volume for our sample
FINRA recently starts to publish weekly ATS volumes by venue with a
delay; our data are trade by trade
Limit order book and HFT activity on Nasdaq
Intraday VIX
67 earnings announcements
−→ Stock-day-minute panel (117 × 21 × 390)
Menkveld-Yueshen-Zhu
Shades of Darkness: A Pecking Order of Trading Venues
11
Dark volume shares
DarkMid (2.1%): dark pools focusing on trading at midpoint
DarkNMid (7.7%): dark pools with flexible prices
DarkRetail (10.8%): retail internalization
DarkPrintB (0.9%): average-price trade
DarkOther (5.8%): remainder
DarkNMid (7.7%)
DarkMid (2.1%)
DarkRetail (10.8%)
DarkPrintB (0.9%)
DarkOther (5.8%)
Lit (72.8%)
Menkveld-Yueshen-Zhu
Shades of Darkness: A Pecking Order of Trading Venues
12
Dark data snippet
Alcoa on Oct 1, 2010
date
time
symbol
type
contra
buysell
price
shares
cond1
cond2
cond3
cond4
1-Oct-10
1-Oct-10
1-Oct-10
1-Oct-10
1-Oct-10
1-Oct-10
1-Oct-10
1-Oct-10
1-Oct-10
1-Oct-10
1-Oct-10
1-Oct-10
...
1-Oct-10
9301833
9301834
9301941
9301989
9302005
9302148
9302204
9302224
9302249
9302343
9302540
9302546
...
10100150
AA
AA
AA
AA
AA
AA
AA
AA
AA
AA
AA
AA
...
AA
DP
DP
OT
DP
MP
RT
RT
RT
DP
DP
DP
RT
...
PB
BD
BD
BD
BD
BD
BD
BD
BD
BD
BD
BD
BD
...
B
B
S
B
X
S
B
B
B
B
B
S
...
S
12.2875
12.2875
12.28
12.285
12.285
12.29
12.2701
12.27
12.28
12.28
12.2805
12.29
...
12.285
100
100
100
200
100
9000
300
100
100
100
100
160
...
179379
@
@
@
@
@
@
@
@
@
@
@
@
...
@
...
4
...
...
B
Menkveld-Yueshen-Zhu
Shades of Darkness: A Pecking Order of Trading Venues
13
Table 1: Variable descriptions. This table lists and describes all variables used in this study. All variables
are generated for one-minute intervals. Variables that enter the econometric model (Section 4) are underscored.
The subscript j indexes stocks; t indexes minutes. Type “Y” and “Z” are described in the panel VARX model.
VARX model
Endogenous variables Y :
Type
Variable Name
Description
Panel A: Dark venue trading volumes
Y
VDarkMidjt
VDarkNMidjt
VDarkRetailjt
VDarkPrintBjt
VDarkOtherjt
Volume of midpoint-cross dark pools
Volume of non-midpoint dark pools
Volume of retail flow internalization
Volume of average-price trades (“print back”)
Volume of other dark venues
VLitjt
Total volume minus all dark volume
Panel B: NASDAQ lit market characterization
BASpreadjt
TopDepthjt
HFTinTopDepthjt
HFTinVolumejt
NASDAQ lit market bid-ask spread divided by the NBBO midpoint
Sum of NASDAQ visible best bid depth and best ask depth
Depthjt based on only HFT limit orders divided by Depthjt
NASDAQ lit volume in which HFT participates divided by total NASDAQ lit volume
Panel C: Overall market conditions
TAQVolumejt
RealVarjt
VarRat10Sjt
Z
VIXt
EpsSurprisejt
Menkveld-Yueshen-Zhu
TAQ volume
Realized variance, i.e., sum of one-second squared NBBO midquote returns
Variance ratio, i.e., ratio of realized variance based on ten-second returns relative to realized variance based on one-second returns (defined to be one for a minute with only
one-second returns that equal zero)
One-month volatility of S&P500 index (in annualized percentage points)
Surprises in announced EPS, calculated as the absolute difference in anShades ofEPS
Darkness:
A Pecking
Trading
nounced
and the
forecast Order
EPS, ofscaled
in Venues
share price: |announced EPS −
14
Exogenous variables
VIX shocks are innovations from an AR(1) model of ∆ ln(VIXt ) at
minute frequency:
∆ ln(VIXt ) = α + β∆ ln(VIXt−1 ) + Innovt .
(In the current paper we use VIX level, and results are very similar.)
EPS surprise is calculated as
|Announced EPS − Forecast EPS|
.
Closing price on the day before
Menkveld-Yueshen-Zhu
Shades of Darkness: A Pecking Order of Trading Venues
15
VARX model:
dark volumes, market conditions (spread, depth, volatility, HFT,...)
yj,t = αj +
+
}|
{
z
Φ1 yj,t−1 + · · · + Φp yj,t−p
Ψ1 zj,t−1 + · · · + Ψr zj,t−r
{z
}
|
+εjt .
urgency: VIX shocks and EPS surprise
Optimal lags p = 2 and r = 1 chosen according to BIC
The estimation gives the dynamic interrelation between dark volumes and
market conditions. We focus on the implications on dark venue market
shares.
Menkveld-Yueshen-Zhu
Shades of Darkness: A Pecking Order of Trading Venues
16
1
Pecking Order Hypothesis
2
Data and Econometric Model
3
Venue Pecking Order in the Data
4
A Suggestive Model and Welfare Implications
Menkveld-Yueshen-Zhu
Shades of Darkness: A Pecking Order of Trading Venues
17
Volume share response to VIX shocks
Impulse-response of volume shares to +1% shock to ∆ ln(VIX )
Pecking order predicts: SDarkMid ↓↓, SDarkNMid ↓, SLit ↑
Menkveld-Yueshen-Zhu
Shades of Darkness: A Pecking Order of Trading Venues
18
Volume share response to VIX shocks
Impulse-response of volume shares to +1% shock to ∆ ln(VIX )
Pecking order predicts: SDarkMid ↓↓, SDarkNMid ↓, SLit ↑
SDarkMid
SDarkNMid
2.5
SLit
8
100
1.5
1
0.5
Market share, percent
2
Market share, percent
Market share, percent
7
6
5
4
3
2
80
60
40
20
1
0
1
2
Menkveld-Yueshen-Zhu
3
Minutes
4
5
0
1
2
3
Minutes
4
5
0
Shades of Darkness: A Pecking Order of Trading Venues
1
2
3
Minutes
4
5
18
Volume share response to VIX shocks
Impulse-response of volume shares to +1% shock to ∆ ln(VIX )
Pecking order predicts: SDarkMid ↓↓, SDarkNMid ↓, SLit ↑
SDarkMid
SDarkNMid
2.5
SLit
8
100
1.5
1
0.5
Market share, percent
2
Market share, percent
Market share, percent
7
6
5
4
3
2
80
60
40
20
1
0
1
2
3
Minutes
4
5
0
1
2
3
Minutes
4
5
0
1
2
3
Minutes
4
5
Pecking order hypothesis is confirmed:
∆SDarkMid
∆SDarkNMid
=
SDarkMid
SDarkNMid
∆SDarkNMid
∆SLit
Reject null:
=
.
SDarkNMid
SLit
Reject null:
Menkveld-Yueshen-Zhu
Shades of Darkness: A Pecking Order of Trading Venues
18
Other dark venues’ response to VIX shocks
Impulse-response of volume shares to +1% shock to ∆ ln(VIX )
SDarkRetail
SDarkPrintB
10
0.5
8
0.4
SDarkOther
3.5
6
4
2
Market share, percent
Market share, percent
Market share, percent
3
0.3
0.2
2.5
2
1.5
1
0.1
0.5
0
1
2
3
Minutes
Menkveld-Yueshen-Zhu
4
5
0
1
2
3
Minutes
4
5
0
Shades of Darkness: A Pecking Order of Trading Venues
1
2
3
Minutes
4
5
19
Volume share response to earnings surprises
Consider a 1% shock to earnings surprises
Pecking order predicts: SDarkMid ↓↓, SDarkNMid ↓, SLit ↑
SDarkNMid shares
2.27%
100
90
80
70
60
1% EpsSurprise
Menkveld-Yueshen-Zhu
No shock
110
6.3%
SLit shares
7.06%
100
90
80
70
60
1% EpsSurprise
No shock
Volume share relative to stead state
1.82%
Volume share relative to stead state
Volume share relative to stead state
SDarkMid shares
110
110
80.04%
77.52%
1% EpsSurprise
No shock
100
90
80
70
60
Shades of Darkness: A Pecking Order of Trading Venues
20
Volume share response to earnings surprises
Consider a 1% shock to earnings surprises
Pecking order predicts: SDarkMid ↓↓, SDarkNMid ↓, SLit ↑
100
90
80
70
60
1% EpsSurprise
110
90
80
70
8.83%
100
90
80
70
60
1% EpsSurprise
Menkveld-Yueshen-Zhu
1% EpsSurprise
110
No shock
110
0.33%
90
80
70
1% EpsSurprise
1% EpsSurprise
No shock
80
70
SDarkOther shares
0.41%
100
60
77.52%
90
60
No shock
80.04%
100
SDarkPrintB shares
9.53%
Volume share relative to stead state
Volume share relative to stead state
SDarkRetail shares
110
SLit shares
7.06%
100
60
No shock
6.3%
Volume share relative to stead state
SDarkNMid shares
2.27%
No shock
Volume share relative to stead state
1.82%
Volume share relative to stead state
Volume share relative to stead state
SDarkMid shares
110
110
2.64%
3.17%
1% EpsSurprise
No shock
100
90
80
70
60
Shades of Darkness: A Pecking Order of Trading Venues
20
1
Pecking Order Hypothesis
2
Data and Econometric Model
3
Venue Pecking Order in the Data
4
A Suggestive Model and Welfare Implications
Menkveld-Yueshen-Zhu
Shades of Darkness: A Pecking Order of Trading Venues
21
Model setup
One traded asset with normalized value E (v ) = 0
Two representative investors: a buyer and a seller
Three venues: Lit, DarkNMid, DarkMid
Timing
I
I
I
I
Buyer and seller observe private trading needs Z + and Z − . Size of
each is either Q > 0, with probability φ, or 0, with probability 1 − φ.
They simultaneously choose trading venues, possibly splitting orders
Trade happens in three venues
Unexecuted orders incur inventory cost of
γ
× Inventory2 ,
2
where γ > 0 is a proxy for urgency.
Menkveld-Yueshen-Zhu
Shades of Darkness: A Pecking Order of Trading Venues
22
Venues
Lit: Buyer pays the ask β > 0; seller gets the bid −β; infinite depth
DarkNMid is run by a competitive liquidity provider with inventory
cost: (η/2) · Inventory2 . Restrict to linear prices:
Buyer’s price is p + = δxN+
Seller’s price is p − = −δxN−
DarkMid crosses orders at midpoint price 0. Volume is
+ −
vM = min(xM
, xM ).
Menkveld-Yueshen-Zhu
Shades of Darkness: A Pecking Order of Trading Venues
23
Buyer’s problem is to maximize
price to pay
in DarkNMid
price to pay in DarkMid z
}| { price to pay in Lit
z }|
{
z }| {
δ
π + (z) = −E 0 · VM+ (z) − xN+ (z)2 −β · xL+ (z)
2
2
γ
+E 0 · z − VM+ (z) − xN+ (z) − xL+ (z) − E z − VM+ (z) − xN+ (z) − xL+ (z) .
{z
}| 2
|
{z
}
liquidation value of remaining position
quadratic cost for failing to trade
Similar for the seller.
Menkveld-Yueshen-Zhu
Shades of Darkness: A Pecking Order of Trading Venues
24
Equilibrium
Proposition. If
Q≤∆≡β
1
1
+
(1 − φ)γ
(1 − φ)η
,
then there exists an equilibrium with the following strategies:
xM (0) = 0;
xN (0) = 0;
xL (0) = 0;
δ
Q,
δ + (1 − φ)γ
(1 − φ)γ
xN (Q) =
Q,
δ + (1 − φ)γ
xM (Q) =
xL (Q) = 0.
If Q > ∆, then there exists an equilibrium with the following strategies:
xM (0) = 0;
xN (0) = 0;
xL (0) = 0;
β
,
(1 − φ)γ
β
xN (Q) = ,
δ
xL (Q) = Q − ∆.
xM (Q) =
In both cases, the DarkNMid liquidity provider sets the slope of price schedules
δ = (1 − φ)η.
Menkveld-Yueshen-Zhu
Shades of Darkness: A Pecking Order of Trading Venues
25
Venue pecking order as an equilibrium implication
Proposition. As investor urgency γ increases, lit volume share increases
and dark volume share decreases. Furthermore, DarkMid is more sensitive
to urgency than DarkNMid:
∂sN /sN
∂sL /sL
∂sM /sM
<
<0<
.
∂γ/γ
∂γ/γ
∂γ/γ
Menkveld-Yueshen-Zhu
Shades of Darkness: A Pecking Order of Trading Venues
26
Venue pecking order as an equilibrium implication
Proposition. As investor urgency γ increases, lit volume share increases
and dark volume share decreases. Furthermore, DarkMid is more sensitive
to urgency than DarkNMid:
∂sN /sN
∂sL /sL
∂sM /sM
<
<0<
.
∂γ/γ
∂γ/γ
∂γ/γ
Recall the empirical test:
∆SDarkMid
∆SDarkNMid
∆SLit
<
<0<
,
SDarkMid
SDarkNMid
SLit
after VIX shock or earnings surprises.
Menkveld-Yueshen-Zhu
Shades of Darkness: A Pecking Order of Trading Venues
26
Welfare cost of shutting down dark venues
Two sources of investors’ cost CMNL : spread paid to liquidity
providers and inventory cost
Shut down DarkMid and DarkNMid, recalculate the equilibrium and
the associated CL
β
(VolumeM + VolumeN )
2
≈ $1.43 billion/year
|
{z
}
CL − CMNL =
Calibrated result
Menkveld-Yueshen-Zhu
Shades of Darkness: A Pecking Order of Trading Venues
27
Conclusion
We characterize dynamic fragmentation of U.S. equity markets
A unique dataset on disaggregated U.S. dark trading
A pecking order of trading venues, characterized by heterogeneous
responses to urgency shocks
Evidence supports investor self-selection
Suggestive model with welfare implications
Menkveld-Yueshen-Zhu
Shades of Darkness: A Pecking Order of Trading Venues
28