“Big Data is like teenage sex: everyone talks about it, nobody really knows how to do it, everyone thinks everyone else is doing it, so everyone claims they are doing it.”
As part of our ongoing quest to enhance our analytics, and to continue to meet our clients requests, we have been spending considerable time over the last few months researching ideas to model the expected cost arising from the forward point component of FX transactions. Such a model would complement our existing framework for estimating expected costs for the spot component.
This research is far from straightforward. The FX Forward market is still a largely voice driven market, often with significant biases in pricing arising from supply and demand or basis factors. This results in a general lack of high quality data with which to perform rigorous modelling. At BestX, however, we do now have a unique data set of traded data that allows for such research and we hope this will provide the foundation for the production of such a model.
We have decided upon a 2 phased approach. Phase 1 will be a relatively simple, yet pragmatic, extension of our existing parametric model for expected spot costs. We plan to launch this in Q1 to meet the initial need for a fair value reference point for the costs arising from forward points. Phase 2 is a longer term project, which will take us down the road of a data-driven approach as there are indications that a parametric model will have limitations when attempting to model the foibles of the forward FX market. We are already planning for this and have started research into using machine learning methods, including nearest neighbour algorithms, to cope with the complexity of this market. As part of this research, one of the initial pieces of work was to try to understand what the key drivers for FX forward costs actually are as we are aware of the risks of utilising machine learning on big data sets without an understanding of the fundamentals. We have summarised the initial findings of this work here.