vincent
09-19-2008 05:19 PM
Robert Almgren did the next keynote, with an excellent tutorial on
Best Execution using CEP, and why the investment banks need it for things like market data analytics,
smart order routing, and trading limit checks, and why equities will surely be followed by futures, FX and fixed income trading. He mentioned the various generations of market analytics in best execution being:
- simple tactics
- optimal execution based on arrival price
- exploiting time-varying liquidity
He also explained that adaptive algorithms were used to account for market volatility (for example, general patterns of trading that occur every day) in some equities, but which become much more variable (/less predictable) in medium and small cap equities.
Some of the interesting comments I thought were:
- His company purchased a CEP tool even though his IT team had built an event platform.* Basically, the vendor was easier to deal with than the IT department. On build vs buy, the cost of the CEP system was < 1 days revenue. Sensible thinking at work.
- Smart Order Routing could possibly be done using a stats package like R (or presumably S+). But it was easier to use a CEP tool.
- Some of the biggest costs were in configuring market adapters and configuring the trading engine links. The message here is that, where available, pre-built adapters help.
- There is no requirement for on-line updates - the day’s trading window is small, so there is plenty of time for out-of-hours updates to the decisioning mechanisms.
David Luckham moderated the follow-up panel on “Event Processing as seen by customers: where does it help, what is the potential, what is missing”.
- Arkady Godin of MITRE commented on the need to consider the data world in EP.
- Ewan Silver and Alex Kozlenkov of Betfair commented on their extreme (Google/Amazon/Ebay size, 24×7 worldwide, highly regulated) betting exchange application. Challenges include:
- Billions of events per sec
- High security in a extremely distributed environment
- Low latency, but with a preference for accuracy over speed
- Long running patterns over months (such as betting on events taking place a year in the future)
- Need to re-use rules, while SQL was insufficient as a rules representation
- Jeff Farr from Dycom Industries is involved with telco and tackling distributed event processing - but as nothing had been announced yet he couldn’t talk about his app!
- Ian Koenig talked about content aggregation and management, and event-based interfaces.
Source...
09-20-2008
