CompatibL’s founder and Head of Quant Research, Alexander Sokol, sat down with Risk.net’s Phil Harding to talk about CompatibL’s machine learning-based autoencoder market models and the challenge of getting buy-in from internal stakeholders and regulators.
In this Q&A session, Alexander highlights the distinctive characteristics of this brand new class of machine learning models, expanding on the success of CompatibL’s award-winning market generator models, and speculates about how businesses might avoid the difficulties associated with model validation and regulatory compliance.
Click here to watch the video.
What Are Autoencoder Market Models?
Autoencoder market models are a new type of interest rate models based on machine learning algorithms called autoencoders. CompatibL uses these algorithms to represent the historical yield curve and volatility surface shapes optimally, using the smallest number of model variables and without any preexisting notion of what their behavior should be.