Machine Learning for Long Risk Horizons: Market Generator Models

Explore a novel type of machine learning techniques in model validation: machine learning-based market generators that can be used to solve a long-standing problem in risk management and demonstrate a better accuracy of generated market scenarios for interest rates and FX compared to traditional techniques

Conference Materials

Market generators are machine learning algorithms for generating realistic samples of market data when historical time series has insufficient length or gaps. While most of the recent research on market generators focused on daily time horizons, the problem of generating realistic market data samples for time horizons from 1 year to 30 years and longer has multiple applications including limit management, insurance (economic scenario generation) and macro investing.

In this presentation, we focus on machine learning in model validation and describe a family of market generators that use machine learning to generate market scenarios with accurate probability distribution over long time horizons from limited time series.

Author

Alexander Sokol

Alexander Sokol is the founder, Executive Chairman, and Head of Quant Research at CompatibL.

Alexander won the Quant of the Year Award in 2018 together with Leif Andersen and Michael Pykhtin, for their joint work revealing the true scale of the settlement gap risk that remains even in the presence of initial margin. Alexander’s other notable research contributions include systemic wrong-way risk (with Michael Pykhtin, Risk Magazine), joint measure models, and the local price of risk (with John Hull and Alan White, Risk Magazine), and mean reversion skew (Risk Books, 2014).