Last month I hoped across the Atlantic to the windy city of Chicago for the annual conferrence about using R in finance. It was a two day conference with a mixture of 5 minute lightning talks, longer sessions and keynote speakers. Over the two days all walks of finance were discussed; equities, bonds even crytocurrencies. The confernece even provided breakfast and lunch which was a nice suprise, usually you pay your massive registration fee and left to fend for meals.

I spoke about FX trading with a lightening presentation on using Hawkes processes to predict FX trades times. It was my first time presenting at a confernece and nice to finally put a slideshow together and sell what I’ve been doing for the last four years of my life.

Matt Taddy of Amazon was there speaking about measuring innovation but he pointed out some best practices for the machine learning workflow. One of which is the importance of generating new data, which means you need to be thinking of experiments to collect new data which can then be used to validate the previous preditions. He also briefly mentiond double machine learning, which is a causal framework for estimating treatment effects using machine learning. Overall, his talk was a great overview of how machine learning should be used in a business setting and what you need to consider before during and after the model has been built. This brief summary doesn’t really do it justice, but, he has a book coming out soon that I will be purchasing.

A talk by Brian Boonstra showed how doing statistics on a high requency financial dataset is trickier than just applying the standard methods. He showed that Bitcoin displays fat tails that cannot be atributed to stochastic volatility which in turn implies that traditional market making techniques can’t be applied to cyrptocurrencies. This is a typically thgeoretical result that might not be true in practise, I’m sure there are plenty of people out there that have proven this. Talks like this highlight how Bitcoin has helped make high-frequency methods more accssible as the data is available for free.

Tactical asset allocation using machine learning by Majeed Simaan. This about using machine learning to predict market returns and then using these predictions to make allocations descions. I liked this talk because it outlined how predictions can be translated directly into asset allocations. Then after googling the speaker I found that he had made an RPubs notebook that mirrors the talk (here). It takes you thorugh the full analysis: downloading the data, preparing the data, modelling the data, extracting the signal and backtesting the strategy. I’m currently implementing my own version heavily inspired by this post.

Overall, the two days really opened my eyes to how broad finance can be and these three talks that I’ve highlighted only scratched the surface of what was on offer. For a relativley `small’ conference it was great to learn so much across a wide variety of things.