Quants should take care with synthetic data – Lehalle

Synthetic data creates an illusion of certainty and risks messing up portfolio construction, says quant

A top quant has cast doubt on one of the hottest new practices in investing — the use of synthesised data to train machine learning models and to backtest strategies.

In a new working paper, Charles-Albert Lehalle, a professor in applied math at École Polytechnique, together with Adil Rengim Cetingoz, a researcher at Université Paris 1 Panthéon-Sorbonne, shows that synthetic data cannot reduce uncertainty in models – no matter how much of the new data quants might create.

Many machine learning