Oct 2, 2019
In this episode Adam Butler and Rodrigo Gordillo host ReSolve's
Head of Quantitative Research, Andrew Butler to discuss how ReSolve
employs tools from the field of machine learning to produce
meaningful and practical improvements in investment outcomes.
We start with Andrew's background in applied mathematics and in particular his experience applying ML tools to solve complex real-world problems in the physical sciences. It was fascinating to hear Andrew recount how he came to understand that the tools that work well to model physical systems are much less useful in a financial context. This was a consistent theme throughout the discussion.
Our objective was to offer a high-level overview of the ML toolset so we started by defining what ML is and digging into three traditional classes of ML: unsupervised learning, supervised learning, and reinforcement learning. We make each method accessible with simple examples and discuss how ReSolve uses the respective techniques to improve outcomes at virtually every step in the investment process.
At many points the group paused to reflect on the myriad ways in which financial markets are distinct from other problem categories. We explore why it is critical to view financial markets through the prism of ML for any statistical inference, and discuss several tools that should be handy in the toolbox of every modern financial analyst.
Of critical importance, we reinforced the fact that the ML toolset is useless – if not downright dangerous – if deployed naively without the direction and support of experienced operators. Without a deep understanding of the unique properties and pitfalls of financial markets ML tools are likely to do much more harm than good to portfolios.
We also discussed why the most important step – by far – in data-driven research is the validation and online learning step – the sentinel – where trader intuition and experience can amplify results by orders of magnitude.
There was some debate about the role of machines and humans in finance and more broadly, and how those roles may evolve. Rodrigo held out hope for sustained human dominance in complex tasks while Adam argued that machines could be playing a much larger and positive role in society already if humans would just get out of the way!
There is a lot of marketing around the field of machine learning at the moment but very little nuanced, practical wisdom. We hope you take something of practical relevance from our conversation.