Download Disputation, August 4th, 2009, Ryan Riordan

High-Frequency Trading and
Price Discovery
Terrence Hendershott
(Jonathan Brogaard
Ryan Riordan)
Overview

Technology has been good for markets




Is every use of technology good?
How do we think about (evaluate) HFT?
What are costs/benefits of those closest to the market?
Analysis of Nasdaq HFT data in terms of price discovery




Statistical model of HFT and “information” and “noise” in prices
Challenges in interpreting these results
Cross-correlations between HFT and price changes
Explore what types of information HFT has (not in paper yet):
 Macro news announcements
 Market-level (as opposed to individual stocks) results
 Limit order book imbalances
High-Frequency Trading and Price Discovery
2
The old trading floor, NYSE 1873
More recent trading floor
High-Frequency Trading and Price Discovery
4
Current trading floor
Securities Market
Algorithmic & High-Frequency Traders/Systems
Algorithmic and High-Frequency Trading
Algo Trade Definition: “The use of computer algorithms to manage the trading process.‘‘
(Hendershott et al. 2011); HFT similar, but for short holding periods
High-Frequency Trading and Price Discovery
5
Is the evolution good?
Some News Articles
“...rule changes are need to control risks...”
“…development have had negative effects…”
“HFT firms are accused of flooding markets
with orders that are cancelled…, leading to
volatility”
-
Ultra fast trading needs curbs – global regulators
-
July 7, 2011 - London
-
-
“...the stock market is more prone than
ever to large intraday moves with little or
no fundamental catalyst.”
“locusts … feeding off the equity
market.”
It's hard to imagine a better illustration (of
social uselessness) than high-frequency
trading. The stock market is supposed to
allocate capital to its most productive uses,
for example by helping companies with good
ideas raise money. But it's hard to see how
traders who place their orders one-thirtieth
of a second faster than anyone else do
anything to improve that social function.
Paul Krugman August 2, 2009 - New York Times
-
“High-frequency traders generated about
$21 billion in profits last year.”
“...use rapid-fire computers to essentially
force slower investors to give up profits,
then disappear before anyone knows what
happened. ”
6
Longer perspective on technological changes
affecting liquidity
Figure 1. Bid-ask spreads on Dow Jones stocks
(all DJ stocks 1900-1928, DJIA stocks 1929-present)
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
Technological advances have made the markets better:
Evidence from trends and specific events/introduction
High-Frequency Trading and Price Discovery
7
Liquidity (Trading Costs/Spreads) on NYSE
Decline is from lower averse selection (less information in trading) -- Hendershott et al. 2011
How well can we measure liquidity as trade size -> 0? Transitory impact of large orders?
High-Frequency Trading and Price Discovery
8
Algorithmic Trading on NYSE
Hendershott et al. 2011
High-Frequency Trading and Price Discovery
9
Correlation is not causality: we need a
“natural experiment”: Autoquote
High-Frequency Trading and Price Discovery
10
Algo Trading is a superset of HFT
What is HFT?
HFT is always AT – but AT is not always HFT
Typical properties of HFT:
–
High-speed trading
–
Sophisticated computer programs
–
Use of co-location services and data feeds
–
Short time frames for establishing and liquidating positions, high trading volume and intensity
(SEC, 34-61358, Concept Release on Equity Market Structure)
•
HFT is a mixture of the use of technology and trading strategies (do they differ?)
•
What is new is direct access, increased electronic information sources, and more computing
High-Frequency Trading and Price Discovery
11
HFT Strategies (SEC)
• Make money at the
bid-ask spread and
liquidity rebates
Passive Market Making
Arbitrage
• Cross asset and
cross market
arbitrage
• Statistical arbitrage
Directional
• News Trading
• Liquidity Detection
(order anticipation)
• Momentum trading
and ignition
Strategies
• Latency Arbitrage
• Flash orders
•
–
Structural
Anything new here?
Short-lived strategies with tight risk management
– Based on public information, technology, and sophisticated tools
Is the focus the technology, strategies, or interaction?
High-Frequency Trading and Price Discovery
12
HFT and Price Discovery (beyond liquidity)
United Airlines files for Ch.
11 to cut costs - Bloomberg
United Airlines Share Price
Sept. 19th, 2002
Sept. 6th, 2008
High-Frequency Trading and Price Discovery
10:58 Sept. 8th, 2008
Sept. 8th, 2008
13
Price Discovery: Efficient Price & Noise

Explicitly model observed prices each minute as efficient
price (martingale) and noise (pricing error)

Observed price is martingale (efficient price) plus noise:
pi,t =mi,t + si,t

mi,t = mi,t-1 + wi,t.

Trading makes cov(s,w) <> 0
Standard market microstructure approach: Innovations in
martingale are related to innovations in trading:
𝐴𝑙𝑙
wi,t = 𝜅𝑖𝐴𝑙𝑙 𝐻𝐹𝑇𝑖,𝑡
+ 𝜇𝑖,𝑡

Transitory component also relates to trading:
𝐴𝑙𝑙
si,t = 𝜙𝑠𝑖,𝑡−1 + 𝜓𝑖𝐴𝑙𝑙 𝐻𝐹𝑇𝑖,𝑡
+ 𝜐𝑖,𝑡

Identification from Cov(m,u)=0 (Hasbrouck (1993) and others)
High-Frequency Trading and Price Discovery
14
Price Discovery: Interpretation
𝐴𝑙𝑙
wi,t = 𝜅𝑖𝐴𝑙𝑙 𝐻𝐹𝑇𝑖,𝑡
+ 𝜇𝑖,𝑡
𝐴𝑙𝑙
si,t = 𝜙𝑠𝑖,𝑡−1 + 𝜓𝑖𝐴𝑙𝑙 𝐻𝐹𝑇𝑖,𝑡
+ 𝜐𝑖,𝑡

k trading’s (order flow) relation to efficient price



+ informed about change in efficient price
- adversely selected on efficient price
y trading’s (order flow) relation to noise

+ direction of trading correlates with more noise


Transitory price impact, risk mgt., manipulation, order anticipation
- trading against noise, making prices more efficient

Statistical arbitrage via identifying transitory effects
Decompose overall HFT order flow into Liquidity Demand and Supply
Examine highest volatility days (stability?)
High-Frequency Trading and Price Discovery
15
Nasdaq HFT Data


120 stocks: small, medium, large cap
Identifies 26 independent HFT firms



Via unique “port” which firms use
Does not identify large integrated firms, e.g., Goldman
All data is aggregate across all HFT firms

Is this a representative sample? If not, why not?

HFT firms worried about regulatory response
Data is available via NDA with Nasdaq
Nasdaq had 20-40% market share
HFT is 42% of volume in large stock, 12% in small stocks
Market/limit order volume similar in large, less limit in small




High-Frequency Trading and Price Discovery
16
Nasdaq HFT (SSM) Price Discovery Results

Overall HFT trade to make prices more efficient
Results remain (and are stronger) on high-volatility days

Market order trading is responsible for this




+ correlation with efficient price, - with pricing error/noise
Consistent with forecasting both parts of returns
Limit order trading coefficients have opposite signs






- correlation with efficient price, + with pricing error/noise
Efficient price: standard adverse selection of liquidity providers
Noise: risk mgt., adverse selection, manipulation, order anticipation?
Do these passive trades make money?
Yes, earn the spread and liquidity rebates
Overall HFT profitability per $ is low: ~$0.03/$10,000 traded
High-Frequency Trading and Price Discovery
17
Interpretation of Statistical Model









Is HFT being informed good or bad?
HFT liquidity demand imposes adverse selection
But, also trade against noise in prices
It is possible to do one without the other?
How important is this information? How long-lived?
Would it get into price anyway?
HFT liquidity supply looks a lot like market making (good)
How could we measure excess intermediation (bad)?
Next, cross-correlations for individual stocks and market-wide
and macro news announcements
High-Frequency Trading and Price Discovery
18
Individual Stocks:
HFT and Future Stock Returns (1sec periods)
HFT Demand/Supply/Allt - Rtnt, t+10
0.25
0.20
Correlation Coefficients
0.15
0.10
0.05
HFT Demand
0.00
HFT Supply
t
t+1
t+2
t+3
t+4
t+5
t+6
t+7
t+8
t+9
t+10
HFT All
-0.05
-0.10
-0.15
-0.20



Time (in seconds)
HFT Demand positively predict stock returns for a few seconds
HFT Supply negatively predict stock returns for a few seconds
Demand effect dominates overall
High-Frequency Trading and Price Discovery
19
Event Study: Macro Announcements
800.00
0.20%
600.00
0.15%
400.00
0.10%
200.00
0.05%
0.00
0.00%
-200.00
-0.05%
-400.00
-0.10%
-600.00
-0.15%
-800.00
Returns
HFT Order Flow ( $10,000)
Negative Macro News
-0.20%
-10 -9
-8
-7
-6
-5
-4
-3
-2
-1
0
1
2
3
4
5
6
7
8
9
10
Seconds Around Macro News Announcements
HFT Demand




HFT Supply
HFT All
VW Return
Aligning return and HFT times is challenge (Nasdaq BBO for subset)
HFT Supply is adversely selected; HFT Demand is transmitting info
Overall Supply dominates; how to think about this HFT Demand?
Regressions show HFT Demand is not associated with noise
High-Frequency Trading and Price Discovery
20
Event Study: Macro Announcements
800.00
0.20%
600.00
0.15%
400.00
0.10%
200.00
0.05%
0.00
0.00%
-200.00
-0.05%
-400.00
-0.10%
-600.00
-0.15%
-800.00
Returns
HFT Order Flow ($10,000)
Positive Macro News
-0.20%
-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0
1
2
3
4
5
6
7
8
9 10
Seconds Around Macro News Announcements
HFT Demand



HFT Supply
HFT All
VW Return
Similar to Negative announcements
Predominantly HFT is providing liquidity at most stressful time
209 Announcements in: Construction Spending, Consumer Confidence, Existing Home Sales,
Factory Orders, ISM Manufacturing, ISM Services, Leading Indicators, Wholesale Inventories
High-Frequency Trading and Price Discovery
21
Market-wide:
HFT and Future Stock Returns (1sec periods)
HFTtD, S, and All - VW Rett, t+10
0.50
Correlation Coefficients
0.30
0.10
HFT Demand
HFT Supply
t
t+1
t+2
t+3
t+4
-0.10
t+5
t+6
t+7
t+8
t+9
t+10
HFT All
-0.30





HFT Demand positively predict market-wide returns for a few seconds
HFT Supply negatively predict market-wide returns for a few seconds
Supply effect dominates overall (also true in a market-level SSM)
Past market returns predict HFT similarly (E-mini predicts HFT Demand)
HFT has a longer lasting role in market-wide price discovery
-0.50
High-Frequency Trading and Price Discovery
22
Another source of HFT Information

Limit order book imbalance LOBI (Size at offer minus Size at bid)
 LOBI predicts future returns (-), HFT Demand (-), HFT Supply (+)
 Not much in the small stocks
Panel A: 𝑹𝒆𝒕𝒕+𝟏
𝑳𝑶𝑩𝑰𝒕
(t-stat)
Large
Medium
Small
All
-0.01
-0.01
0
-0.01
(-16.36)
(-4.35)
(0.01)
(-1.19)
-0.19
-0.21
-0.13
-0.18
(-15.17)
(-7.55)
(-2.05)
(-8.49)
0.07
0.05
0.06
0.06
(5.90)
(1.86)
(1.18)
(3.33)
-0.06
-0.11
-0.03
-0.07
(-12.36)
(-11.81)
(-0.63)
(-4.20)
𝑫
Panel B: 𝐇𝐅𝐓𝐭+𝟏
𝑳𝑶𝑩𝑰𝒕
(t-stat)
𝑺
Panel C: 𝐇𝐅𝐓𝐭+𝟏
𝑳𝑶𝑩𝑰𝒕
(t-stat)
𝑨𝒍𝒍
Panel D: 𝐇𝐅𝐓𝐭+𝟏
𝑳𝑶𝑩𝑰𝒕
(t-stat)
High-Frequency Trading and Price Discovery
23
Sources of HFT Information

Other sources of “Public” information in past prices, orders?



In other assets (market-wide, announcement results)
Large stock leading small stocks
HFT Demand predicts returns




How to think about short-lived nature of (public) information?
 Foucault, Hombert, and Rosu (2012)
Reducing the transitory part is good
Trading on (soon to be) public information is not
How do we know what information will get into price without HFT?
High-Frequency Trading and Price Discovery
24
Conclusions/Thoughts

Technology has improved markets


Algorithms: prices more efficient and markets more liquid
HFT is technology applied to certain strategies



Make prices more efficient; true on high-volatility days also
 What are the benefits of getting info into prices in seconds?
 Does this improve financing and investment decisions?
Should traders closest to the market mechanism be informed?
 Focus attention on HFT liquidity demanding strategies?
 Are long-term investors losing out (zero sum trading)?
 If so, will HFT become competitive and offer technology services
to long-term investors?
 Will market structures evolve to support LFT (without HFT)?
What is the optimal configuration of intermediation sector?
 Free entry or regulated monopoly/oligopoly?
 Specific regulations on liquidity supply and demand?
High-Frequency Trading and Price Discovery
25