Christian Hesse, Quant, Deutsche Bank
"Very generic data-driven techniques are capable of picking out interesting features that are interpretable..."
London - A variety of financial industry players were in attendance for MATLAB's computational finance conference yesterday to discuss complex financial models and implementation techniques.
Delivering the keynote address, Matthew Waldron from the Bank of England's research hub said that future plans are in place to enhance the forecasting platform to provide more functionality around policy analysis - forward guidance being one area.
Forward guidance has been a controversial policy ever since it was announced in August last year. The BoE's monetary policy committee is feeling the heat over speculation on when interest rates are going to rise after a faster-than-expected drop in unemployment.
Waldron added that there's also a view to devote more resources to research, particularly around big data techniques.
"The (BoE) is uniquely placed in its role as regulator to take advantage of some very large and interesting proprietary data sets to shed light on interesting analytical questions that the monetary policy committee or the financial policy committee may have," he said.
Ultimately, MPC decisions are judgemental and informed by models - the role of which is to create a stylised representation of how the world might work, Waldron said. And the more transparent the inner workings of those models, the more valuable they are.
"There is a huge premium on models that elucidate economic mechanisms and stories that can be related to theory rather than just black box statistical forecasting," he said.
The event served as a showcase of the ways in which MATLAB clients were putting the software to use. In the case of the BoE, it is part of the forecasting platform - COMPASS, MAPS, EASE and the suite of models.
So why such a poor showing for forecasting unemployment figures? Speaking to Automated Trader on the sidelines of the conference, Waldron said that volatility of data means that a drop in unemployment of some 0.3% isn't unlikely, and some leeway needs to be taken into account.
Other topics of the day included machine learning and quantitative investment.
Christian Hesse, quant for Deutsche Bank's Autobahn Equity Europe and researcher at University College London, talked about the variety of machine learning techniques applied to finance, for example portfolio optimisation.
As part of his talk, he showed how to analyse more closely the intra-day market structure important to algorithmic trading and product.
"Very generic data-driven techniques are capable of picking out interesting features that are interpretable, make sense and therefore (give) confidence in being able to...automate a lot of the clustering or assessment that you are trying to do," he said.
Hedge funds too are incorporating the software into their workflow. Edward Hoyle, a director in Fulcrum Asset Management's research team, explained how the firm identifies where sources of returns exist, works with models in MATLAB and puts its trading strategies into production in a Java environment.
An important component, he added, is code optimisation to improve speed. "We are medium frequency, we are not a high frequency trading firm, so speed is not our primary concern but we do like things to run quickly and cleanly."
In terms of working with models, he said that if simple works, then more complicated may improve results. But if you can't get a simple model to work, don't bother trying anything more complicated.
Simple, however, does not mean it will be easy. The trouble is finding signals that give you some information about future returns. "Once you find the relationships...then the modelling step should be straightforward but finding the relationships to begin with and where to look for them is definitely the hard part," he said.
Moreover, transaction costs need to be taken into account. This is particularly important for shorter horizon strategies, the shortest of which is a couple of hours for the firm.
"We tend to be conservative in our estimation of costs. It's critical," he said.