by Joseph Rickert
In a post last week, I offered some first impressions about R/Finance 2013. Apparently, I was way off in estimating that 30% of the attendees were academics. The R/Finance organizers were quick to point out that percentage of academics attending the conference has been a constant 10% over the years; and this year was no different. Why is this important? Well, first off it points to the level of sophistication of the industry attendees who came to hear talks that were mostly very technical, both with respect to the level of mathematics involved and R usage. And, perhaps it explains why there was very little hype at the conference. There were no exagerated claims for Hadoop or any other technology, and discussions involving big data were very "matter of fact". When it comes to technology, the quants are a sober lot, pragmatic, competent and comfortable with the latest technology trends.
The opening keynote presentation: R on the trading Desk (PDF) by Ryan Sheftel, Managing Director of Fixed Income Division at Credit Suisse, underscored this last point. Ryan, an engaging and experienced speaker, provided some extraordinary insight into the stance quants and traders take toward technology: By way of background, Ryan noted that fixed income markets which provide immediate liquidity between buyers and sellers are decentralized. Traders are constantly taking on risk which they try to mitigate by predicting what their clients are going to do. Ryan noted that machine learning ideas, very often implemented in R, have been a “great boon” to the industry. Before the financial crisis, the quants were focused on building complex models to price options, now they are focused on prediction using time series models. These days, the technology on a fixed income desk at Credit Suisse includes:
- SQL and data sources such as OneTick
- R and time series libraries such as zoo and xts
- R libraries for reshaping and manipulating data
- RStudio’s integrated development environment
This combination of technology has helped to “lower the bar to accessibility” meaning timely and high quality models. More people are answering their own questions, testing their own code and doing version control. As a consequence of this, the barrier between quants and traders is breaking down. Both groups are using the same tools and, if I understood Ryan correctly, there are even expectations that traders will do their own unit testing! The idea is to make quality assurance part of the creative process. It is expected that more lines of code will be devoted to testing than calculations and that QA is the responsibility of the person who had the idea. Ryan noted that this way of working and the kinds of tools available are affecting the kinds of people being hired. Although Ryan never used the term, it is clear that at least one organization within Credit Suisse is building its future with "data scientists".
Ryan’s talk covered much more ground than I can explore here; much more even than his slides indicate. However, I would be remiss not to mention Ryan's challenges for R and his warning against complacency. R is apparently deeply entrenched at Credit Suisse, It is now an “acceptable” tool at the bank and quants and traders write their own proprietary packages. However, a disaster linking R to a big loss, such as the Excel error that apparently contributed to the London whale debacle, could dislodge it. In Ryan’s opinion it is just "too easy to do some things" in R. Ryan stated that programming rigor around the language and better error tracking would be very useful here. This is the classic tradeoff between freedom and security, between providing individual users with powerful feature-rich tools and limiting the damage a careless individual can cause. These kinds of concerns help to make the case that some organizations could benefit from a managed distribution of the R language. As for the warning against complacency: Ryan noted that Python with the Pandas library also has a place on the trading desk.
All of the presentations for R/Finance 2013 are online here. Many thanks to the speakers and the conference organizers for making them available. Many of the presentations have snippets of R code that are helpful following the math. (See for example the presentations of Bernhard Pfaff and Sanjiv Das.) For those of you who still want more; the cumulative collection of R/Finance presentations is becoming quite a online library. Have a look at the presentations from previous years:
