Bernard S. Donefer Distinguished Lecturer Baruch College, CUNY

Transcription

1 Bernard S. Donefer Distinguished Lecturer Baruch College, CUNY Principal, Conatum Consulting LLC Bernard S. Donefer. All rights reserved. May not be reproduced by any means without prior written consent.

2 Wall Street s computer scientists and linguists keep trying to find quicker ways to react to the news by creating ever-more complicated algorithms, the mathematical formulas that execute stock trades automatically based on such criteria as headlines and news stories. The idea is to buy or sell at a faster clip than the guy or computer at a rival trading desk. 2

3 All this can go horribly wrong, as United Airlines learned last week when a six-year-old story about the company s 2002 bankruptcy filing gained new life on the Internet, triggering a cascade of stock sales. In a matter of about 12 minutes more than $1 billion in stock-market value evaporated. Human error seems to have played only a minor role. The financial damage was mostly the result of the interplay between the algorithms that search and compile information from the Web and the ones that Wall Street firms and hedge funds use to make trades automatically. 3

4 New York Times Sept 14 4

5 Witness another recent case that had the potential to cause a stock market wipeout, but benefited from serendipitous timing: after the close of trading on Aug. 27, Bloomberg News inadvertently released an obituary of Steve Jobs, the chief executive of Apple who, despite frequent rumors of ill health, was, and is, very much alive. The story was quickly retracted. Remember its not just news, it could also be erroneous or late quotes, or any other algo parameter driving the trading strategy. 5

6 The Japanese arm of Credit Suisse Group may get punished by local regulators over problems involving an automated trading system used by its major clients, the Yomiuri Shimbun daily said on Friday Japan's Securities and Exchange Commission (SESC) is considering asking the Financial Services Agency (FSA) to punish the company for a series of problems with an algorithmic stock trading system, which include mistakenly placing large share orders, the paper said. Should the FSA find it necessary, it may order Credit Suisse to improve its business management and operations, the Yomiuri added, without giving details.

7 An official at the SESC also declined to comment but said that in general terms administrative punishment could include suspension of business or ordering a company to improve its operations. The algorithmic trading system relies on computer algorithms to decide when to trade stocks and in what amount, based on data such as price movements. Using these calculations, the system then places orders on its own. Kajino said Credit Suisse began algorithmic trading in Japan in 2003 but declined to give further details except to say that this business was growing. 7

8 'When I use a word,' Humpty Dumpty said, in a rather scornful tone,' it means just what I choose it to mean, neither more nor less.' 'The question is,' said Alice, 'whether you can make words mean so many different things.' OK Humpty, what is electronic trading? 8

9 An STP FIX connected trade to an electronic market for execution An order directly from a buy side OMS Retail or institutional A high frequency trading strategy executed autonomously by computer About which are we and the regulators most concerned? 9

10 The Liquidity Algo Trade Break up an institutional sized order and use an automated strategy across multiple market places and over time, to minimize market impact You know the asset, side and size to trade High speed search for liquidity at a price The Alpha Seeking Algo Trade Implement a strategy using historical models and real time market data, find and exploit profitable trading opportunities stat arb / pairs trading / news Do not know in advance the security or timing of the trade Aka black box, quant, etc. 10

11 Market Making Given a list of securities, stocks, options, etc. and a source of order flow, adhere to the order handling rules and trade at the spread Automate the expertise of a Series 55 licensed trader under all market conditions Most NASDAQ and ISE market makers are fully automated The NYSE is offering ability to interface algo capabilities to floor specialists 11

12 12

13 Buy side Sell Side Market Place 13

14 Buy Side Client Am I using the correct strategy? Did I get the trade done at the benchmark cost? Algo Vendor Sell Side Firm or ISV Do I have the capacity, speed, reliability to meet my client s needs? Does my Algo work as advertised? Intraday client exposures Market Places Do I have the capacity, speed, reliability to meet my client s needs? Can an algo roil the market? 14

15 The FAT finger Remember Mizuho Market Data historical and real time Quality and speed (latency) Network latency trading and data LAN and WAN Pre-trade process TCA process assumptions Choosing the right strategy and parameters Executing the strategy Increased volatility, trending markets, news Basket correlations Predators sensing your orders Trading battlebots earn est. $15-25B annually 15

16 The Algo The model s logic and implementation The historical data to create the model The real time data controlling its execution DMA and smart router infrastructure Connectivity, logic, latency FIX message correctly representing the algo strategy New order types and market specific orders The market s trading system correctly interpreting and executing the order Is FIX ADTL enough? 16

17 Who is now responsible for Best Execution? The buy side trader with DMA and an assortment of algo tools Who s job is it to ensure the algo strategies are correctly understood? What responsibilities are taken on by the sell side providing algo s and infrastructure? Are the rules for disaster recovery, losses due to technology related losses clearly understood by all parties? How will internalized markets, (ATS s) and ECN s address risks from algo s? Will a100 share retail quote move the NBBO and result in millions of shares executed at that price? 17

18 Most risk management systems provide a daily view of client and firm positions Can they respond to high frequency trading? While flat EOD, can intraday positions exceed credit limits? Are clients violating short sale rules? Impact of high frequency trading on firm capital requirements Can we stop a client s trading mid strategy??? Hedge funds using multiple prime brokers hide their strategy and positions What risks are held by prime brokers in the event of a hedge fund failure? Clearance, settlement, loans, etc. With buy side firms creating their own algo s, how does their broker handle Algo s Gone Wild? 18

19 Wasn t 1987 algo s gone wild? Will Algo trading be the next crash s portfolio insurance? Is momentum and alpha trading adding to systemic risk? Slicing and dicing further decreasing bid/offer size Reducing liquidity and furthering market fragmentation High speed trading increasing quote frequency, flicker and adding noise to price discovery Are retail orders being disadvantaged? Brokers unable to control client trading Potential for market manipulation it wasn t me, it was the algo! Complex clearing and settlement 19

20 Supervision of Algorithmic Trading Jan 18, 2008 Market Integrity Notice use of an algorithmic trading system and certain limitations on the ability of Market Regulation Services Inc. to intervene to vary or cancel trades arising from a malfunctioning algorithmic trading system. the source of the order or the means by which an order is entered on a marketplace does not relieve a Participant of responsibility for the supervision of such orders. RS is also of the view that orders entered on a marketplace without the involvement of staff of the Participant, such as in the case of orders transmitted to a marketplace by means of an algorithmic trading system, present heightened risks to both the integrity of the markets and to the financial position of the Participant. 20

21 Participant should develop and implement fail-safe mechanisms for the supervision of proprietary algorithmic trading systems that are adequate to prevent the entry of orders and execution of trades that, based on market conditions, are unreasonable. Source: 21

22 More proprietary Algo s More complex strategies More competition More asset classes and cross asset trading More electronic market places More data More speed More risk 22

23 Future Traders 23