Presentation on High Frequency Trading risks delivered during OpRisk conference in London in June 2012. Content includes an overview of key risks affecting high frequency trading.
1. Failure to meet regulatory and exchange requirements.
2. Removal of human decision making once the algorithms are finished.
3. Extreme market behaviour: Flash Crash (2010).
4. Theft or loss of Intellectual Property.
5. Errors or problems suffered by clients using Direct Market Access and Algo/HFT.
6. Business impact of latency (system errors may increase delays).
7. Limited security controls at the infrastructure level.
8. Failure of hedges. 9. Incorrect/untested strategies.
David Ramirez
IT Audit Director
How tech-is-used-real-time-monitoring-dodd-frank-trade-reporting
Op Risk High Frequency Trading June 14 Final
1. High Frequency Trading
Operational Risk Issues and Mitigation Measures
David Ramirez – Director, IT Audit
14 June 2012 – London 11.10-11.50 am
2. 2
Agenda
1
• Introduction and Key Concepts
2
• Details of Algorithmic Trading and HFT
3
• Key Risks
4
• Mitigating Mechanisms
3. 3
Taxonomy of Algorithmic Trading
“The use of computer algorithms to
Algorithmi automatically make certain trading
c Trading decisions, submit orders, and manage
those orders after submission”.
(Hendershott and Riordan, 2009).
High
Frequency “Employs extremely fast automated
Trading programs for generating, routing,
cancelling , and executing orders in
electronic markets.” (Cvitani and
Kirilenko, 2010)
Trading
Strategies “Market Making, Electronic Liquidity
Provision, Statistical Arbitrage,
Liquidity Detection, Latency Arbitrage,
etc” (Gomber and Arndt, 2011)
4. 4
Agenda
1
• Introduction and Key Concepts
2
• Details of Algorithmic Trading and HFT
3
• Key Risks
4
• Mitigating Mechanisms
5. 5
Latency vs. Position Timeline
High
Traditional
Long-Term
Investment
Latency
Algorithmic Trading
HF
Low
T
Short Long
How Long Position Held
6. 6
Latency? - Key Concepts
Trading Risk Book
Market Data Trade Order
Logic Management Processing
There is some
The Algorithm (algo) Risk Management The order needs to
latency within the
Data from exchange, would need to take checks on the orders: arrive from the
exchange, tends to
news, other decisions based on size, frequency, fat system hosting the
be minimal at
participants. high volumes of fingers, VAR, short algo, to the
selling, etc. around 0.5
data. exchange.
milliseconds.
7. 7
Arbitrage:
•The practice of taking advantage of a price
difference between two or more markets: striking a
combination of matching deals that capitalize upon
the imbalance, the profit being the difference
between the market prices.
Collocation:
•Servers are hosted by the exchange (NYSE, LSE,
NASDAQ) in large data centres; access granted
directly to the exchange infrastructure.
8. 8
HFT Trading Strategies
•Market Making: Earn the •Market Neutral Arbitrage:
spread between bid and ask. Long and short; gain the
difference.
•Rebate Driven Strategies:
Leverage rebates offered by •Cross Asset/Market and
Exchange. Exchange Traded Fund
(ETF) arbitrage: Leverage
•Statistical Arbitrage: Predict
price inefficiencies across
discrepancies in the market.
asset/markets.
•Latency Arbitrage:
Predicting the ‘National Best
Bid and Offer’ value.
9. 9
Agenda
1
• Introduction and Key Concepts
2
• Details of Algorithmic Trading and HFT
3
• Key Risks
4
• Mitigating Mechanisms
10. 10
Key Risks Related to HFT Environments
1. Failure to meet regulatory and exchange
requirements.
2. Removal of human decision making once the
algorithms are finished.
3. Extreme market behaviour: Flash Crash
(2010).
4. Theft or loss of Intellectual Property.
5. Errors or problems suffered by clients using
Direct Market Access and Algo/HFT.
11. 11
Key Risks Related to HFT Environments - cont
6. Business impact of latency (system errors
may increase delays).
7. Limited security controls at the
infrastructure level.
8. Failure of hedges. Incorrect/untested
strategies.
12. 1. Failure to Meet Regulatory and Exchange 12
Requirements
•Regulators and exchanges define message structures that must be
adhered to (regulatory and contractual); this includes specific flags on
the packets (short selling, max order size, frequency on same name,
dealing on restricted names/securities).
•September 2011, the SEC announced that it would start collecting
copies of algorithms for analysis. There is also a plan to collect live
logs from all exchanges.
•Time compliance: Have you closed a trade on time? How is it being
measured? (GPS and the IEEE1588v2 Precision Time Protocol (PTP);
Financial and stock exchange data centers are increasingly deploying
GPS receivers on the roof of the data center and then distributing GPS
timing throughout the data center.)
13. 1. Failure to Meet Regulatory and Exchange 13
Requirements– cont
Securities and Exchange Act 1934 and MAS
•“For the purpose of creating a false or misleading
appearance of active trading in any security registered
on a national securities exchange, or a false or
misleading appearance with respect to the market for
any such security,
14. 2. Removal of human decision making once the
algorithms are finished.
•Algorithms will be executing instructions without
any supervision, the potential for errors increases
significantly.
•Human intervention should be available at all
times, as expected by exchanges.
15. 15
3. Extreme market behaviour: Flash Crash
(2010).
Flash Crash – May 6 2010 – Runaway Algos – Domino Effect? Wikipedia.org
•The Flash Crash, was a United States stock market crash
on May 6, 2010 in which the Dow Jones Industrial Average
plunged about 1000 points—or about nine percent—only to
recover those losses within minutes. It was the second
largest point swing, 1,010.14 points, and the biggest one-
day point decline, 998.5 points, on an intraday basis in
Dow Jones Industrial Average history.
•"'HFTs began to quickly buy and then resell contracts to
each other—generating a 'hot-potato' volume effect as the
same positions were passed rapidly back and forth.'"
16. 3. Extreme market behaviour: Flash Crash 16
(2010). - cont
High volume days tend to be high execution days for HFT – based on
network capacity it can impact traditional trading technology and pipes
assigned to that business.
Volumes can be massive and add up quickly – e.g. a bug in the code
order will become a very large order error and then lead to an error
with the exchange or network or exchange connectivity.
A coding error (which is big and means the Algo is wrong from the
start) can be (mis)understood to be a routing issue with an exchange
(which is small and easier to fix).
17. 4. Theft or loss of Intellectual Property. 17
‘Secret sauce’
• There are examples in the industry of at least four legal
cases in relation to algorithms being stolen.
•These programs are key intellectual property, it is very
easy for staff to leave the firm with the code underlying the
trading strategy.
•Firms struggle with understanding when does an Algo
become an Algo.
18. 5. Errors or problems suffered by clients using
Direct Market Access and Algo/HFT .
•Firms offer Direct Market access to prime clients,
this creates a risk as the activities of clients can
impact the compliance with exchange rules and
regulations.
19. 6. Business impact of latency (system errors 19
may increase delays).
•Latency has direct impact on the P&L, an Ultra-HFT
strategy and some forms of arbitrage will fail if latency is
higher than expected.
•Communications from the servers (collocated or not) to
the exchange must be done over low latency links.
Trading Applications
Packaged Applications Proprietary Applications
Network Network
20. 20
7. Limited controls at the infrastructure level.
•Algorithmic Trading environments tend to have a
very limited number of infrastructure controls, most
are between the local corporate network and the
HFT equipment.
•Operating systems are modified to gain speed
advantages; this has an impact on the security
configuration and layers of security available.
•There is a significant demand increases on the
underlying infrastructure.
21. 8. Failure of hedges. Incorrect/untested
strategies.
•Poorly tested algorithms or interpretation errors
could disrupt the market or drive trading losses.
The magnitude of these will be related to available
liquidity and market conditions.
22. 22
Agenda
1
• Introduction and Key Concepts
2
• Details of Algorithmic Trading and HFT
3
• Key Risks
4
• Mitigating Mechanisms
23. 23
Mitigating Measures
•Increased oversight and •Measuring latency
visibility over algorithms. across applications,
operating systems and
•Built-in and regulatory networks.
algorithmic limits/checks
(e.g., circuit breakers). •Security reviews over
Active data leakage the environment.
controls. • Robust change
management controls
and testing/validation
over new algorithms.
25. Evolution of Order Processing Time (1995-2011)
Source: NYSE Technologies – Eric Bertrand 2011
1200
1000
Latency (microseconds)
800
600
400
200
§¦ ¥¤ £ §¦ ¨¤ £
0 ¢¢ ¡
1995 2000 2005 2006 2008 2009
1 second = 1,000 millisecond =1’000,000 microseconds.
26. How Many Transactions? (Approximate Numbers!)
Number of HFT Transactions For Each Action
Blink of an Eye
Brain Recognises Human Expression
Hard Disk Read
Housefly Wing Flap
0 10,000 20,000 30,000 40,000 50,000 60,000 70,000 80,000
Brain Recognises Human
Housefly Wing Flap Hard Disk Read Blink of an Eye
Expression
Series1 600 800 40,000 80,000