Fischer Francis Trees & Watts Defines and Evaluates Systematic Investment Strategies
- Backtest analysis time reduced from days to minutes
- Reliable strategy assessment ensured
- Multiple asset classes supported
As the global fixed-income manager of BNP Paribas Investment Partners, Fischer Francis Trees & Watts (FFTW) manages assets for institutional investors around the world. The company’s quantitative team develops systematic investment strategies that rely on rule-based models combined with their colleagues’ judgment, opinions, and analysis. This FFTW quantitative team uses MATLAB® for modeling and evaluating its investing strategies.
“In many large firms, researchers who have questions about new strategies rely on other teams to answer them,” says Ben Steiner, quantitative researcher at FFTW. “MATLAB gives the people who identify problems the ability to solve them. At FFTW, that means we are not sitting around waiting for answers. We can test more ideas, invest in the best ones, and keep pace with inevitable changes in financial markets.”
The team also wanted to ensure that it correctly interpreted backtesting results. “Backtesting can be very dangerous,” says Steiner. “When you overfit data, for example, it is easy to misinterpret the results and come up with a strategy that is perfectly tuned to the past but useless for the future.”
To help avoid overfitting, FFTW needed a way to visualize the results of backtests and gain a clear understanding of how the underlying strategy was working.
FFTW analysts used MATLAB to develop tools for collecting and transforming investment data, modeling strategies, creating backtests, and analyzing the results of backtests.
They used Datafeed Toolbox™ to access market data from the Bloomberg® financial data service, and used Database Toolbox™ to retrieve proprietary and historical data stored locally in a Microsoft® SQL Server® database.
Using MATLAB and Statistics and Machine Learning Toolbox™, they developed tools for cleaning and transforming the data. This process involved identifying outliers, converting price data into returns, calculating standard deviations and moving averages, and converting dates with Financial Toolbox™.
The team developed a backtesting framework for using the resultant data to evaluate fixed-income investment strategies. These strategies can be modeled in MATLAB as a set of positions or a historical time series, or they can be defined in Microsoft Excel® or other statistical software packages.
The analysts used MATLAB to create a graphical interface for evaluating investment strategies using a wide range of measures, including profit and loss, maximum drawdown, sensitivity to transaction costs, turnover, and diversification.
The interface, which incorporates techniques the team learned from user-contributed files on the MATLAB Central File Exchange, enables the team to compare a portfolio for a particular strategy against hundreds of random portfolios with similar characteristics and against portfolios already in production.
FFTW analysts have deployed profitable systematic strategies based on analysis completed in MATLAB.
Backtest analysis time reduced from days to minutes. “When we were working with spreadsheets and other tools, analysts typically started each project with a blank slate,” says Steiner. “With our MATLAB based tools we can apply the same process across different strategies. It used to take days to analyze backtest results. Now we can do it in minutes—and more thoroughly.”
Reliable strategy assessment ensured. “Because we standardized on the tools we built with MATLAB, for the first time we can put all our models through the same analysis,” notes Steiner. “We’re comparing apples to apples. That consistency enables us to reliably assess different strategies and to detect problems when strategies behave differently than expected.”
Multiple asset classes supported. “The characteristics of bond futures, inflation linked bonds, emerging market debt, and exchange traded funds vary widely, yet MATLAB enabled us to build a common set of tools that we can apply to them all,” says Steiner. “Supporting different asset universes with a single platform lets us spend more time on research.”