Best modelling innovation: CompatibL

Alexander Sokol, CompatibL
Alexander Sokol, CompatibL

To address the need for accurate risk measurement during the Covid-19 pandemic, CompatibL developed a new and innovative software feature – a machine learning-based market generator. With pre-pandemic historical data no longer accurately representing the current levels of risk, its ability to measure risk from a shorter time series has been advantageous during the pandemic.

Banks and asset managers cannot accurately calibrate traditional risk models to the short-length pandemic-era time series. The time series for any given currency, stock or credit name is too short to gain any statistical confidence in the estimate. While sufficient data is available across the entire range, combining it only produces a model for the ‘average currency’ or ‘average name’, which is not sufficiently accurate.

Using unsupervised machine learning, CompatibL’s market generators rigorously aggregate statistical data and then generate data samples for each individual name. The use of machine learning algorithms makes it possible to reduce the inherent statistical uncertainty of the short time series, while preserving and incorporating the differences between the names into the model. This ability was validated by comparing it to out-of-sample data – the gold standard of model validation.

CompatibL’s machine learning-based, market generator-driven models are an industry first, based on pioneering research by quant experts. The models can generate the data for time horizons between the one and 30 years that risk models require. This has made it possible during the pandemic to accurately measure credit risk, limits, insurance reserves and macro-strategy performance. The machine learning algorithms generate samples of market data when historical time series have insufficient length, without relying on any preconceived notions about the data.

Unlike traditional ‘model-free’ techniques that rely on interpolation, model learning can recognise and use patterns in data. And incoming data is continuously incorporated into the model, enabling it to evolve gradually.

CompatibL’s machine learning-based solution addressed the need for an accurate way of measuring risk from short time series during the unfolding Covid-19 crisis, while meeting the regulatory requirement of being free from subjective judgement.

The model is also able to capture and use the full range of data in both crisis and normal regimes for long-term risk projections. It evolves gradually as the crisis subsides over time, so that banks and asset managers do not need to carry out a disruptive switch to another model when regimes change.


Judges said:

  • Machine learning-based market generator, making optimal use of sometimes scarce data.
  • Obvious problem, interesting solution.
  • Reduced the reliance on pre-pandemic historical data.
  • This is truly innovative. Being able to generate time series data adequate for risk management from sparse and short true periods will be crucial given the ever-lower utility of actual historical data as a consequence of events like negative rates, Covid, credit crisis, and so on.


Alexander Sokol, Founder, Executive Chairman and Head of Quant Research at CompatibL, says:

“On behalf of CompatibL, I am honoured to accept the Risk Markets Technology Award for Best modelling innovation. Machine learning is a transformative new technology that will touch every aspect of the front, middle and back office. I am confident that, before the end of this decade, it will become the market standard in financial product valuation and risk management.

Last year, financial markets experienced unprecedented upheaval. Managing the new levels of risk required innovative solutions, and machine learning was able to meet this challenge. The CompatibL team is proud to be at the forefront of integrating machine learning into trading and risk software.”

You need to sign in to use this feature. If you don’t have a account, please register for a trial.

Sign in
You are currently on corporate access.

To use this feature you will need an individual account. If you have one already please sign in.

Sign in.

Alternatively you can request an individual account here