Page 478 - ISC PROCEEDINGS 21.4

joint distributions and inter-series dependencies; however, they also bring practical
challenges such as training instability, mode collapse, difficulty reproducing heavy tails
and non-stationarity, and potential privacy leakage. Addressing these challenges requires
evaluating GANs not only on statistical fidelity but also on operational criteria like privacy
guarantees, validation procedures, and integration into production-grade data
infrastructure.
This paper surveys and categorizes Financial GAN architectures with an emphasis on
the stylized facts each family tends to capture and illustrates how GAN-driven synthetic
data can function as a component of institutional digital infrastructure to support stress-
testing, regulator sandboxes, and privacy-preserving model development.
The remainder of the paper is organized as follows. Section 2 discusses the role of
data and digital infrastructure in economic transformation and why governed access
matters for supervision and innovation. Section 3 gives a technical overview of Financial
GAN architectures (autoregressive/recurrent, conditional/hybrid, transformer/attention-
based) and highlights privacy-focused variants and maps model classes to the stylized
facts they tend to capture. Section 4 describes datasets and the
experimental/benchmarking setup used to evaluate the frameworks for infrastructure-
ready synthetic financial data. Section 5 discusses trade-offs and policy recommendations
for deployment, and Section 6 concludes with a summary and directions for future work.
2. Role of data & infrastructure in economic transformation
Data is the raw material of modern financial markets and macroeconomic
policymaking. When access to timely, high-quality data is reliable and well governed,
markets work more efficiently, regulators act more effectively, and new digital services
can scale. Below is a focused explanation of why data access and sharing matter for
markets and regulators, and what infrastructural capabilities are required to deliver those
benefits.
2.1. Why data access and sharing matter
Reduce information asymmetry and improve price discovery. Better access to
transaction, order-book, and balance-sheet data reduces the information gap between
market participants. That leads to more accurate pricing of risk and assets, tighter spreads,
and improved allocative efficiency across the economy.
Enable evidence-based regulation and faster macroprudential response. Regulators
rely on diverse datasets to detect build-ups of systemic risk (credit concentrations,
liquidity stress, correlated exposures). Broader data sharing shortens detection time and
enables targeted interventions (e.g., countercyclical capital, sectoral guidance).
Support stress testing and scenario analysis. Shared data allows supervisors and
institutions to run realistic counterfactual scenarios and systemic stress tests. These
exercises depend on rich historical and cross-institutional datasets to model contagion
channels and tail risks.
Accelerate innovation while managing risk. Fintech innovators need access to
representative datasets to build, test, and validate new services (credit scoring,
automated market-making, robo-advisors). Controlled data sharing lets innovation
proceed without exposing sensitive customer records.
Promote financial inclusion and competition. When data is available, new entrants
can offer tailored products for underserved populations, increasing competition and
inclusion while reducing concentration risk in markets.

477