← All insights

Advances in Volatility Modelling; The Costs and Benefits of Synthetic Data for Global Banks

Published Research · 16 July 2026 · 2 min read

---

Published in Computational Economics (Springer), this paper examines what banks gain — and what they give up — when synthetic data stands in for real market history in market-risk estimation.

Scope

The study evaluates synthetic data for estimating Value at Risk (VaR) and Expected Shortfall (ES) for banks trading three major equity indices: the S&P 500, the FTSE 100 and the EuroStoxx 350.

Why it matters for practitioners

Market-risk models are constrained by the history available. Real series are finite, contain structural breaks, and cannot be shared freely between institutions or with regulators. Synthetic data offers a route around those constraints — but only if the generated series preserve the properties risk measures actually depend on: the shape of the tail, volatility clustering, and the dependence structure across assets.

That is the trade-off this research quantifies. Synthetic data is not free: it inherits the assumptions of the generator, and a series that looks right in the body of the distribution can still misprice the tail where VaR and ES are measured. The contribution is to set out the costs alongside the benefits, rather than treating synthetic data as a straightforward substitute.

How this connects to our work

This paper is the methodological foundation for our Synthetic Data assessment: copula and block-bootstrap generation, paired with a validation suite that tests exactly what the research says must be tested — distributional fit, correlation preservation, tail behaviour, and VaR/ES accuracy against the real series — plus a privacy assessment.

If you are considering synthetic data for model validation, benchmarking, or sharing data outside your institution, the honest starting point is knowing where it breaks down.