So McKinsey gets the joke on machine learning in banks: it will soon be embedded everywhere. Here’s a link to a presentation I gave to some investment banks a couple of months ago. It’s focussed on credit trading desks, but applies generally to any fixed income OTC product. As always with the banks, it will take some leadership from the likes of Goldman and then machine learning will become the ‘new new thing’.
The key points are:
- The main organising principal of OTC trading desks will become asset liquidity and data richness rather than credit quality or industry sector.
- There are going to be Data Scientists on trading desks. This is a natural home for these guys, but they will also need to be trained in finance and get their hands dirty in the markets.
- There are going to be fewer traders. The liquid stuff will lumped into a central portfolio and managed by a couple of senior traders with the help of a ML algo that will decompose the risk into (tradeable) liquid factors.
- Individual firms will decide how far along the liquidity spectrum they want to position themselves. Full service ‘megabanks’ will continue to offer the full spectrum. Smaller European / Asian banks will manage a central portfolio and avoid anything tricky.
- ML algos will do everything from portfolio decomposition, market impact prediction, client activity prediction & classification, market regime classification, automatic market making, market neutral trade replacement recommendation, ‘red flag’classification… and other things we haven’t thought of yet.
- Machine learning will allow capital intensive trading operations to slash costs and optimise capital use – which is why its implementation is only a matter of time.
What will really speed this process along is that banks are slowly switching development into Python (from Java / C++). With the Scikit learn and pandas modules for python you can knock together and embed machine learning algos into institutional grade systems very easily.
Take a look at the presentation and feel free to use anything you want, but please credit me if you do.