Not random, and not a forest: black-box ML turns white

Bayesian analysis can replace forest with a single, powerful tree, writes UBS’s Giuseppe Nuti

The machine learning (ML) tidal wave is sweeping finance alongside most other industries. Our quants are busy applying new models to various (often old) problems: reinforcement learning for option pricing, deep neural networks for alpha generation, and so on.

Alas, colleagues in model validation – and possibly our regulators – are less enthusiastic, and likely with good reason: these models are often black boxes, making it close to impossible, for example, to explain why an algorithm was short