incorporating news into algorithmic market trading presented by philip gagner, vice president...

20
Incorporating News into Algorithmic Market Trading Presented by Philip Gagner, Vice President RavenPack International, S.L. With the kind assistance of Dow Jones Newswires

Upload: raymond-sparks

Post on 05-Jan-2016

212 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Incorporating News into Algorithmic Market Trading Presented by Philip Gagner, Vice President RavenPack International, S.L. With the kind assistance of

Incorporating News into Algorithmic Market Trading

Presented by

Philip Gagner, Vice President

RavenPack International, S.L.

With the kind assistance ofDow Jones Newswires

Page 2: Incorporating News into Algorithmic Market Trading Presented by Philip Gagner, Vice President RavenPack International, S.L. With the kind assistance of

Algorithmic Trading

• The use of formal computer models to trade markets has a long history

• Algorithmic execution works but is limited• Technical trading strategies don't work• News content and market events correlate• News is available online, but not in the right form for

computer programs

Page 3: Incorporating News into Algorithmic Market Trading Presented by Philip Gagner, Vice President RavenPack International, S.L. With the kind assistance of

Formal Computer Models• Start with a model of group behaviors, generally in the

area of behavioral finance• Think about why news events might correlate with

market changes– Real world events drive price, volatility, and volume changes– Good news organizations report relevant events quickly,

before markets react– Markets are sensitive to rumor

• Think about reasonable time frames for leading or lagging correlations

Page 4: Incorporating News into Algorithmic Market Trading Presented by Philip Gagner, Vice President RavenPack International, S.L. With the kind assistance of

Building a Market Modelthat Incorporates News

• Market Indexes and news generally, the problems and why one might believe it possible– Indexes combine very different companies– But, equities all share common characteristics for investors– Investors have choices: money might flow to/from equity

markets generally,– News might drive volatility, as well as (or instead of) price

• As a matter of observation, most exchange-traded companies highly correlate on stock price changes

• We could trade a derivative against Index

Page 5: Incorporating News into Algorithmic Market Trading Presented by Philip Gagner, Vice President RavenPack International, S.L. With the kind assistance of

Disney News/Citibank NewsDisney Stock/Citibank Stock

Page 6: Incorporating News into Algorithmic Market Trading Presented by Philip Gagner, Vice President RavenPack International, S.L. With the kind assistance of

What do we know about news?Reasons to be discouraged

• News is highly periodic – Markets are not

• News is not structured, except for primitive tags• News contains meaning, computer programs do not

deal well with meaning• News is language: ambiguous and fuzzy• Numbers are easier to work with than words, in any

algorithmic way• There is too much news

Page 7: Incorporating News into Algorithmic Market Trading Presented by Philip Gagner, Vice President RavenPack International, S.L. With the kind assistance of

What do we know about news?News is Periodic, Markets Not So

Page 8: Incorporating News into Algorithmic Market Trading Presented by Philip Gagner, Vice President RavenPack International, S.L. With the kind assistance of
Page 9: Incorporating News into Algorithmic Market Trading Presented by Philip Gagner, Vice President RavenPack International, S.L. With the kind assistance of

What do we know about news?Reasons to be encouraged

• News is relevant to value. Price should reflect value

• Traders read news and believe they act on it• Advances in AI, understanding language and

complexity• Success (for example) in spam filtering• Statistics works• There is so much news and it is so alike

Page 10: Incorporating News into Algorithmic Market Trading Presented by Philip Gagner, Vice President RavenPack International, S.L. With the kind assistance of

The RavenPack Approach

• Statistics, not meaning, is within our grasp– Engineers use thermodynamics, not quantum mechanics, to

study systems– Thermodynamics is (nothing but) statistical quantum mechanics

• Discover patterns in news: Make them explicit• Normalizing the News• Use AI technology: Reduce News to Numbers

– Classifier systems– Neural Networks– Image Processing technologies (compression, anti-aliasing)– Other

Page 11: Incorporating News into Algorithmic Market Trading Presented by Philip Gagner, Vice President RavenPack International, S.L. With the kind assistance of

What is Required to Turn News into Numbers?

• More data is better, but quality data is best– Quote from customer: “I never believe in back-testing or

historical results, but I can't live without them”– Huge problems with in-sample testing and apparently true

observations

• Normalizing News requires gathering every news event in a stream—missing data are fatal

• New technologies, especially for time series, must be developed

Page 12: Incorporating News into Algorithmic Market Trading Presented by Philip Gagner, Vice President RavenPack International, S.L. With the kind assistance of

Simple Things that Work• Counting works

– Simply counting certain types of news events over time, correlates with some market behaviors! But you must normalize and filter first.

– Why? Probably because nobody is looking at it. Without equal access to the information that drives market changes, arbitrage is possible.

– DJ Elementized News Feed, a different approach

• Stochastic Methods work• Common Sense works

– Simplistic assumptions don't: If a certain type of story has a market impact within an hour, will it have that effect at 2 AM, market time? No!

– Can it safely be ignored? No!

Page 13: Incorporating News into Algorithmic Market Trading Presented by Philip Gagner, Vice President RavenPack International, S.L. With the kind assistance of
Page 14: Incorporating News into Algorithmic Market Trading Presented by Philip Gagner, Vice President RavenPack International, S.L. With the kind assistance of

An Example: Predicting Price from Corporate Press Releases• Companies issue to:

– Boost short term profits and (therefore) short term stock price

– Regulatory compliance– Name recognition and long term profits

• Empirical question: Do they boost short term stock prices?

• Empirical answer:

Page 15: Incorporating News into Algorithmic Market Trading Presented by Philip Gagner, Vice President RavenPack International, S.L. With the kind assistance of
Page 16: Incorporating News into Algorithmic Market Trading Presented by Philip Gagner, Vice President RavenPack International, S.L. With the kind assistance of

Tools Available inDow Jones News Analytics

• Historical News Archive (20 years) designed for algorithmic traders

• Historical Market Archive• The MIS Application

– Display Tools– Mathematical Tools– Retrieval Tools

Page 17: Incorporating News into Algorithmic Market Trading Presented by Philip Gagner, Vice President RavenPack International, S.L. With the kind assistance of

Some Results

• IBM Stock v. IBM Press Releases

• Baysian MarketTalk and Equities (BMQ)• Volatility Index ($VIX) and News Volatility• Latent Semantic Analysis in Petroleum Market

Page 18: Incorporating News into Algorithmic Market Trading Presented by Philip Gagner, Vice President RavenPack International, S.L. With the kind assistance of
Page 19: Incorporating News into Algorithmic Market Trading Presented by Philip Gagner, Vice President RavenPack International, S.L. With the kind assistance of
Page 20: Incorporating News into Algorithmic Market Trading Presented by Philip Gagner, Vice President RavenPack International, S.L. With the kind assistance of