Problem
A fundamental investor was looking to take the next step in their investment strategy by leveraging untapped sources of data they had collected, but faced with a set of hundreds of instruments, it was difficult to distinguish sources of predictive power from spurious, noisy correlations. The customer’s aim was to build a refined set of signals as a starting point from which to continue their analysis.
Solution
We started with a selection of just over 300 raw signals relating to the prices and trading activity of coffee, oil and gold over a period of years, on a cadence between hourly and daily. Signal to noise ratio was low, making it challenging to detect causal relationships between instruments and the target variables of future price movements for each commodity.
The client was able to use Mind Foundry to screen many raw and derived features rapidly, and so was able to explore more signals than would normally be accessible to them. A variety of techniques including automated stationarisation analysis, spectral decomposition, and pattern recognition were used to generate in excess of 10,000 signals from the initial data. These were refined using a range of factor, model, and signal driven methods to deliver 30 leading indicators which warranted further analysis by hand.
Our client used this information to guide their in depth hand curated analysis and successfully drive a profitable trading strategy on the three commodities combining fundamental and quantitative methods.
