0% found this document useful (0 votes)

92 views36 pages

High-Frequency Trading Analysis

The document analyzes high-frequency traders as middlemen in electronic markets. It finds that while middlemen can theoretically solve no-trade situations by quickly processing public news, they can also create adverse selection if all investors are uninformed. Empirically, a 2007-2008 sample showed adverse selection was a major part of passive order markups, and the introduction of a new venue friendly to middlemen in the Netherlands increased liquidity supply by 29% while lowering volume by 13%. A new highly active middleman preferentially traded when news was important and traded passively 80% of the time.

Uploaded by

borelcantelli

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

92 views36 pages

High-Frequency Trading Analysis

Uploaded by

borelcantelli

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

electronic markets and high frequency

trading: same wine in new jars?

albert j menkveld
tinbergen institute, vu university amsterdam

december 2010

the nyse in 1873

the nyse today

literature
algorithmic trading: Foucault and Menkveld (2008), Hendershott,
Jones, and Menkveld (2010), Chaboud, Chiquoine, Hjalmarsson, and Vega
(2009), Hendershott and Riordan (2009), Biais and Weill (2010)

high frequency trading: Brogaard (2010), Kirilenko, Kyle, Samadi, and

Tuzun (2010), Jovanovic and Menkveld (2010), Menkveld (2010)

limit order books and information asymmetry: Foucault (1999),

Rosu (2006), Hollifield, Miller, Sands, and Slive (2006), Goettler, Parlour,
and Rajan (2009)

middlemen in search markets: Rubinstein and Wolinsky (1987),

Masters (2007), Li (1996), Duffie, Garleanu,

and Pedersen (2005), Lagos,
Rocheteau, and Weill (2009), Duffie, Malamud, and Manso (2009)

questions

1. does algorithmic trading improve liquidity?

Hendershott, Jones, and Menkveld (2010)
2. high-frequency traders as middlemen
Jovanovic and Menkveld (2010)
3. high-frequency trading and market structure
Foucault and Menkveld (2008), Menkveld (2010)

questions

1. does algorithmic trading improve liquidity?

questions

1. does algorithmic trading improve liquidity?

questions

1. does algorithmic trading improve liquidity?

Hendershott, Jones, and Menkveld (2010)

timeSpreads
trend bid-ask
spread
for the
long

run

relative bid-ask spread

Dow Jones stocks
Figure 1. Bid-ask spreads on Dow Jones stocks
(all
DJ
stocks
1900-1928,
DJIA 1929-2000)
stocks 1929-present)
(all stocks 1900-1928, DJIA
stocks
1.60%
1.40%

Proportional spread

1.20%
1.00%
0.80%
0.60%
0.40%
0.20%
0.00%
1900

1910

1920

1930

1940

1950

1960

1970

1980

1990

2000

!"#$%"%&'()*+%'((,%"-%+*&*#-%./)-(+0
time
trend bid-ask spread (ctd)
nyse
value-weighted
average
effective
spread
Figure
1. NYSE
value-weighted
proportional
effective
spreads
0.25%

Proportional effective spread

0.20%

0.15%

0.10%

0.05%

0.00%
1994

1995

1996

1997

1998

1999

2000
11

2001

2002

2003

2004

autoquote was phased in, i.e. December 2, 2003, through July 31, 2003. We use the exogenous no

to identify
causality from algorithmic
for the endogenous
algo trad t/o,
iv regression
including
volatility,
price,trading
andto liquidity.
size We est
it

Lit = i + t + Ait + Xit + it

where Lit is a spread measure for stock i on day t, Ait is the algorithmic trading measure algo tra
share turnover,
market
Lit volatility,
= i +1/price,
t +and
Alog
Xitcap.
+ We
it always include fixed effects and time
it +
all explanatory variables, except that we replace algo tradit with auto quoteit . We regress by q
that are robust to general cross-section and time-series heteroskedasticity and within-group auto

Coefficient on algo tradit

Panel A: quoted spread, quoted depth,

qspreadit
-0.52** -0.42**
(-3.23)
(-2.21)
qdepthit
-3.47**
-1.43
(-2.50)
(-1.16)
espreadit
-0.18** -0.32**
(-2.65)
(-2.23)
Panel B: spread decompositions
rspreadit
0.35**
0.76**
(3.53)
(3.97)
adv selectionit -0.53** -1.07**
(-3.57)
(-4.08)
#observations: 1082*167 (stock*day)

Coefficients on
vola
share
turnoverit tility

and effective spread

-0.43
-0.16
9.92
(-1.44)
(-0.05)
(1.22)
-1.99
15.49
0.61
(-1.07)
(0.39)
(0.19)
-0.35
-1.63
4.65
(-1.56)
(-0.42)
(1.16)

-2.80**
(-3.01)
-5.16
(-0.64)
-1.01**
(-2.32)

0.90*
(9.70
-1.64
(-1.87
0.69*
(9.51

1.03**
(2.06)
-1.39**
(-2.06)

3.13*
(1.92)
-4.12**
(-2.24)

-1.06*
(-2.15
1.75*
(3.29

14.26
(0.46)
-15.48
(-0.47)

15.88
(1.36)
-11.21
(-1.33)

*/**: Significant at a 95%/99% level.

a
: We use quintile-specific coefficients for the control variables and time dummies. For brevity,
cap-weighted coefficient for the control variables and its t-statistic.
b
: We report the Dickey-Fuller test statistic based on the residuals in order to diagnose nonstati
that the series contains a unit root, i.e. it rejects nonstationarity.
12

details). As the LSB decomposition limits persistence to that of an AR(1) process, we also estimate a H
of the trade-related (stdev tradecorr compit ) and trade-unrelated (stdev nontradecorr compit ) comp
transactions (see Section 3.3.2 and Hasbrouck (1991a, 1991b) for details). For the regressions, we use
autoquote to instrument for the endogenous algo tradit to identify causality from algorithmic trading
estimate
M = + + A + X +

iv regression for lsb and hasbrouck decompositions

Mitor=Hasbrouck
i + component
+stock
Xiti +
t + Ait for
it t, Ait is the algorithmic trading me
where Mit is a LSB
on
day
including share turnover, volatility, 1/price, and log market cap. We always include fixed effects and ti
in the Hasbrouck component regressions. We regress by quintile and report t-values based on standar
and time-series heteroskedasticity and within-group autocorrelation (see Arellano and Bond (1991)).
Coefficient on algo tradit
Q1

Panel A: Lin, Sanger, and Booth (1995)

LSB95 f ixedit
0.26** 0.59**
(3.63)
(4.16)
LSB95 adv selit
-0.26** -0.61**
(-3.46) (-3.80)
LSB95 order persistit
-0.18** -0.30**
(-3.06) (-3.10)
Panel B: Hasbrouck decomposition
stdev tradecorr compit
-0.22** -0.26**
(-2.62) (-3.08)
stdev nontradecorr compit 0.13** 0.13**
(2.48)
(2.36)
#observations: 1082*167 (stock*day)

0.69**
(2.26)
-0.84**
(-2.14)
-0.21
(-1.60)

9.91
(0.46)
-12.19
(-0.46)
0.66
(0.28)

8.97
(1.36)
-7.72
(-1.32)
3.30
(1.21)

-0.30*
(-1.69)
0.13
(1.47)

-3.39
(-0.30)
1.03
(0.28)

-0.57**
(-2.73)
0.13
(1.12)

Coefficien
v
share
turnoverit t
2.35**
(2.07)
-2.58*
(-1.85)
-0.82**
(-2.33)

*/**: Significant at a 95%/99% level.

a
: We use quintile-specific coefficients for the control variables and time dummies. For brevity, we o
cap-weighted coefficient for the control variables and its t-statistic.
13

2. high-frequency traders as middlemen

Jovanovic and Menkveld (2010)

main findings
middlemen quickly process public news which in theory
solves no-trade deadlock between uninformed early investor
and informed late investor who trade through take-it-or-leave-it
(limit) orders (+)
creates adverse selection if both investors are uninformed (-)

empirically 07-08 sample belgian and dutch stocks shows

adverse selection is largest part of passive order markup
in-sample, chi-x (a venue friendly to middlemen) is introduced
for dutch stocks; coincidentally a highly active new middleman
shows up who
participates more when public news is important
trades primarily passive (80%)