Adverse selection & liquidity supply in a fast electronic ... · Data from Euronext and AMF : 117...
Transcript of Adverse selection & liquidity supply in a fast electronic ... · Data from Euronext and AMF : 117...
Adverse selection & liquidity supply in a fast electronic market
Bruno Biais, Fany Declerck, Sophie MoinasToulouse School of Economics
Very, very, very preliminary! Comments and leniency welcome!
Banque de France, November 2013
Specialists / dealers / designated market makers offer prices (liquidity supply), investors hit those prices (liquidity demand)
In the old days
Market structure changes
1) Transparent electronic limit order books: all investors can monitor, post & hit quotes
=> Facilitates entry for liquidity suppliers
=> “Endogenous” market makers (e.g., hedge funds) instead of only designated market makers
Technological changes
2) High speed computer connections: HFTFaster connections & faster information => lower risks in liquidity supply => HFT endogenous market makers
High FrequencyTrader
Market
How did these changes affect determinants of market quality?
1. LOB: Free entry, increased competition in liquidity supply+ 2. HFT: better monitoring capacities of liquidity suppliers
=> more liquid markets?
But
How is liquidity supplied and by whom?
Is there more or less adverse selection?
HFT adverse selection & liquidity supply
High frequency traders get info before others=> limit orders less exposed to adverse selection=> good at supplying liquidity, posting limit orders
High frequency traders get info before others=> hit slow quotes with market (marketable) orders=> increases adverse selection for slow limit orders=> deters placement of limit orders by slow
Brogaard, Hendershott Riordan 2013: market orders
HFT market order to buy followed by positive return (informed) and comes after negative return (supplies liquidity) unlike non HFT
HFT limit orders exposed to adverse selection, but less thannon HFT
Brogaard, Hendershott Riordan 2013: limit orders
Are high-frequency traders the only ones to follow such strategies?
Open/transparent electronic limit order books: many different categories of market participants can choose to supply liquidity, e.g., hedge funds
Do these traders need to rely on high speed connection to markets, or can they rely on other sources of comparative advantage?
Capital / ability to carry inventoryInformation (more fundamental than high frequency)
If so, how do their strategies compare to those of HFTs?
Data to answer these questions
Data from Euronext and AMF : 117 stocks traded in EuronextParis, in Eurolist 100 + Next 150, from Jan to Sept, 2010
Dropped: 9 with stocks splits, SEO, … & 5 with very few trades
Orders and trades
Best quotes time-stamped at the micro-secondNeed to replay the entire limit order book
Tick-by-tick data including anonymised member IDKnow if agency trading vs prop tradingKnow if order came from member with high speed connection
Agency trading vs prop trading
Pure play prop-trader/market maker20 “principal” traders
Pure broker only placing client orders37 brokers
Placing both prop & client orders82 “dual” traders
Data to answer these questions
Data from Euronext and AMF : 131 stocks traded in EuronextParis, in Eurolist 100 + Next 150, from Jan to Sept, 2010
Dropped: 9 with stocks splits, SEO, … & 5 with very few trades
Orders and trades
Best quotes time-stamped at the micro-secondNeed to replay the entire limit order book
Tick-by-tick data including anonymised member IDKnow if agency trading vs prop tradingKnow if order came from member with high speed connection
MemberchoosesThrouput
Memberchoosescolocation
Ex: 50
Ex: 200
High speed connection: direct measure
Throughput = max number of messages / sec
Speed Capacity = Σ throuputs
Speed
0%
2%
4%
6%
8%
10%
12%
14%
16%
18%
[0 - 25] ]25 - 50] ]50 - 75] ]75 -100]
]100 -150]
]150 -200]
]200 -300]
]300 -400]
]400 -800]
]800 -1300]
]1300 -1600]
]1600 -2400]
]2400 -4600]
% of members
Max # messages per second
Define as “fast” the 16 members with capacity > 1300 mssg/sec
Fast Slow(>1300 mssg/sec) (< 1300 mssg/sec)
Principal 5 15(100% trades as principal)
Dual trader 11 71
Broker(0% trades as principal) 0 37
5 types of players
Sample of stocksSample : 117 stocks
Sub-sample: 20 stocks (more than 4To = 700 CDs)
large cap & financial: 1 stock (3 stocks)large cap & non-financial: 6 stocks (14 stocks)mid cap & financial: 1 stock (1 stock)mid cap & non-financial: 8 stocks (8 stocks)small cap & non-financial: 4 stocks (4 stocks)
Other stocks = hold-out sample
Sample period (Jan 1, Sept 30, 2010)
0
5
10
15
20
25
30
35
40
45
50
04/01/2010 04/02/2010 04/03/2010 04/04/2010 04/05/2010 04/06/2010 04/07/2010 04/08/2010 04/09/2010
VIX
CrisisModeratevolat
April 23: Grece asks for bailout
May 7: June 14: bailout downgrade
Low volat
Descriptive statistics
Descriptive statistics# of
messages - Daily mean
# mess/second - Mean
# mess/most active sec -
Mean
Daily mean fill
rate
Daily mean cancellation
rate
Daily volume
Trade size
Fast proptrader 193 787 9 528 8% 64% 3 610 469 9 116Fast dual trader 115 356 6 228 16% 52% 4 016 132 14 107
Slow proptrader 8 881 2 88 39% 36% 1 012 390 9 772Slow dual trader 4 304 2 34 52% 27% 553 828 12 936
Slow broker 263 1 17 65% 19% 339 289 18 998
Fast traders seem different from slow traders and brokersExcept for trade size: prop are different from dual and broker
-
100 000
200 000
300 000
400 000
500 000
600 000
700 000
800 000
900 000
23/02/2010 23/03/2010 23/04/2010 23/05/2010 23/06/2010 23/07/2010 23/08/2010
16 fast members > ½ messagesIncreased volat => more messages (slow & fast)
Number of messages
0
1
2
3
4
5
6
7
8
9
3/02/2010 23/03/2010 23/04/2010 23/05/2010 23/06/2010 23/07/2010 23/08/2010
Principals & fast have smaller trade to message ratio// Hendershott & al 2011: trade to mssg ratio proxies for algo
Trades to message ratio
Market order
Information content of market order
Signed % change in midquote regressed on aggressor type
sign of market order x [(Mt+k – Mt-) / Mt
-] =
β1,k Indicator that market order stemmed from fast prop
+ β2,k Slow prop + β3,k Fast dual
+ β4,k Slow dual + β5,k Slow broker
+ Fixed Effect stock + Fixed Effect day + error
After market buy, price risesEspecially for prop (fast or slow) traders: 8 bp after 5 minutesProp trade on reversals: supply liquiditySlow non-prop ride short-term momentum
Info content of market orders
-0,0015
-0,0010
-0,0005
0,0000
0,0005
0,0010
0,0015
-1 h
-30
mn
-15
mn
-10
mn
-5 m
n-3
mn
-2.5
mn
-2 m
n-1
.5 m
n-1
mn
-45
s-3
0 s
-15
s-1
s-0
.5 s
-0.1
s 00.
1 s
0.5
s1
s15
s30
s45
s1
mn
1.5
mn
2 m
n2.
5 m
n3
mn
5 m
n10
mn
15 m
n30
mn
1 h
MO = Fast Prop trader
MO = Slow Prop trader
MO = Fast Dual trader
MO = Slow Dual trader
MO = Slow Broker
Info content of market orders in midst of crisis
Larger information content of market (marketable) orders=> More adverse selection for limit orders hit
Momentum stronger, not reversal
-0,0015
-0,0010
-0,0005
0,0000
0,0005
0,0010
0,0015
-1 h
-15
mn
-5 m
n
-2.5
mn
-1.5
mn
-45
s
-15
s
-0.5
s 0
0.5
s
15 s
45 s
1.5
mn
2.5
mn
5 m
n
15 m
n
1 h
MO = Fast Prop trader
MO = Slow Prop trader
MO = Fast Dual trader
MO = Slow Dual trader
MO = Slow Broker
Signed (midquote after trade – price)/midquote before
Profitability of market orders
Prop market orders profitable: 5 bp after 5 minutesMomentum riding market orders of slow brokers not profitable
(spread > info advantage)
-0,0015
-0,0010
-0,0005
0,0000
0,0005
0,0010
0,0015
-1 h
-15
mn
-5 m
n
-2.5
mn
-1.5
mn
-45
s
-15
s
-0.5
s 0
0.5
s
15 s
45 s
1.5
mn
2.5
mn
5 m
n
15 m
n
1 h
MO = Fast Prop trader
MO = Slow Prop trader
MO = Fast Dual trader
MO = Slow Dual trader
MO = Slow Broker
Limit order
Adverse selection for limit orders
Signed % change in midquote regressed on type of memberwho placed limit order
sign x [(Mt+k – Mt-) / Mt
-] =
β1,k Indicator that limit order had been placed by fast prop
+ β2,k Slow prop + β3,k Fast dual + β4,k Slow dual
+ β5,k Slow broker + Fixed effect stock + Fixed effect day + error
Fast limit orders exposed to less adverse selection than slowSlow broker ask hit after decline in price, continued after tradeNo pattern before fast limit hit
Adverse selection for limit orders
-0,0015
-0,0010
-0,0005
0,0000
0,0005
0,0010
0,0015
-1 h
-30
mn
-15
mn
-10
mn
-5 m
n-3
mn
-2.5
mn
-2 m
n-1
.5 m
n-1
mn
-45
s-3
0 s
-15
s-1
s-0
.5 s
-0.1
s 00.
1 s
0.5
s1
s15
s30
s45
s1
mn
1.5
mn
2 m
n2.
5 m
n3
mn
5 m
n10
mn
15 m
n30
mn
1 h
LO = Fast Prop trader
LO = Slow Prop trader
LO = Fast Dual trader
LO = Slow Dual traders
LO = Slow Broker
Signed (midquote after trade – price)/midquote beforeProfitability of limit orders
Losses of limit orders < adverse selection cost: earn spreadSlow brokers & slow prop traders limit orders incur largest lossFast limit order executions profitable if unwound < 1 minuteFast market more profitable than limit // Hendershott et al. 2012
-0,0015
-0,0010
-0,0005
0,0000
0,0005
0,0010
0,0015
-1 h
-15
mn
-5 m
n
-2.5
mn
-1.5
mn
-45
s
-15
s
-0.5
s 0
0.5
s
15 s
45 s
1.5
mn
2.5
mn
5 m
n
15 m
n
1 h
LO = Fast Prop trader
LO = Slow Prop trader
LO = Fast Dual trader
LO = Slow Dual trader
LO = Slow Broker
Number of executed orders per trader
Fast prop traders use both market & limit orders (because relatively immune to adverse selection)Slow prop traders trade rarely with passive limit orders (because exposed to adverse selection)
0
50
100
150
200
250
300
350
400
450
500
Fast prop Slow prop
market (or marketable limit)
(not immediately marketable) limit
Signed % change in midquote after limit order hit
Adverse selection cost largerGreater disadvantage of slow prop and brokers
Adverse selection in midst of crisis
-0,0015
-0,0010
-0,0005
0,0000
0,0005
0,0010
0,0015
-1 h
-15
mn
-5 m
n
-2.5
mn
-1.5
mn
-45
s
-15
s
-0.5
s 0
0.5
s
15 s
45 s
1.5
mn
2.5
mn
5 m
n
15 m
n
1 h
LO = Fast Prop trader
LO = Slow Prop trader
LO = Fast Dual trader
LO = Slow Dual trader
LO = Slow Broker
Number of executed orders per tradermidst of crisis
Fast prop traders use more market orders & less limit ordersNo significant change for slow prop traders
0
50
100
150
200
250
300
350
400
450
500
Fast prop Slow prop
market (or marketable limit)
(not immediately marketable) limit
Conclusion: Prop tradingProp traders’ market orders more informed than non-prop orders
impose more adverse selection costs on others
Slow prop and fast prop market orders equally informed
Info content = 8 bp after 5 minutes
Profitable (= info content – half spread): 5 bp after 5 minutes
Prop traders (fast and slow) execute market orders on reversals:Buy after price declineSupply liquidity with market orders
Conclusion: Fast trading
Fast limit orders less exposed to adverse selection than slow
For prop traders, speed matters more for limit orders than formarket orders
Fast traders, less exposed to adverse selection, execute larger% of trades via passive (limit) orders than slow traders
Slow prop, more exposed to adverse selection, mainly trade withmarket orders (marketable limit)
Non prop slow traders ride momentum (buy after price rise):some information content, but not profitable