COMMENT: How to use R to make forecasting predictions, by a data scientist at Point72
Quantitative hedge funds are secretive beasts. Strict non-disclosure agreements and non-compete agreements are used to ensure that their proprietary algorithims do not escape into the outside world.
But sometimes the veil is lifted. Many hedge funds publish research papers, with AQR an especially prolific example. Others have released some of their code as open source software, such as the high performance Arctic datastore released by my former fund, AHL.
Now Dr Chris Weiss, a London based employee of US hedge fund Point 72, has published a chapter of the PhD dissertation that likely helped him get a job in finance . Dr Weiss created a forecasting package for R, a language popular with data scientists in finance and elsewhere. His dissertation explaining in some detail how the package works, and why it is useful.
Forecasting is a tricky business, and a variety of techniques have evolved to try and improve the accuracy of predictions. An important insight from academic research is that there is unlikely to be one perfect forecast, and the best results can be obtained from combining forecasts based on different inputs. For example, an equity trader trying to forecast share prices might want to combine predictions from different accounting ratios such as dividend yield and price: earnings ratios, as well as price momentum. Using just use one of these inputs will be inferior to a carefully chosen combination.
There are numerous ways to combine forecasts, and the key contribution of this package is to package over a dozen of these techniques into an easy to use toolbox. The choice of technique to combine forecasts is often an afterthought, with most quant researchers sticking to tried and tested methods, such as linear regression.
The LinkedIn profile of Dr Weiss is also interesting. He has a Phd in Economics on ‘Essays in Hierarchical Time Series Forecasting and Forecast Combination’. At Point72, his job title is listed as ‘Global Macro (Senior Analyst)’.
Global macro funds, like George Soros’ legendary Quantum fund, try and predict high level movements in financial markets and currencies. Important sources of information for global macro traders are historic and predicted changes in economic indicators, such as inflation and unemployment. If you’re trying to replicate the success of Soros, then hiring an economist with expertise in forecasting economic time series makes a lot of sense.
In his article, Weiss uses R to build a forecast of UK electricity supply. Point72 could yet harness it to predict the price of UK power (although this is a tricky market to trade)? Or maybe it could use UK electricity as one input into a model used to trade other assets, such as the share prices of firms that run UK power plants? Either way, if you want to use R for forecasting, Weiss clarifies how it's done.
Robert Carver is the former head of fixed income at quantitative hedge fund AHL, and the author of 'Systematic Trading' and 'Smart Portfolios’. Whilst working at AHL Robert also developed a systematic global macro trading strategy, which was implemented in R.