SBBTS: A Unified Schr\"odinger-Bass Framework for Synthetic Financial Time Series

A team of researchers including Alexandre Alouadi, Grégoire Loeper, Célian Marsala, Othmane Mazhar, and Huyên Pham has introduced SBBTS (Schrödinger-Bass Bridge for Time Series), a novel framework that addresses a persistent challenge in financial machine learning: generating synthetic time series that accurately reproduce both marginal distributions and temporal dynamics. Previous approaches like diffusion-based methods typically fix volatility parameters, while martingale transport models ignore drift components entirely. SBBTS overcomes these limitations by extending the Schrödinger-Bass formulation to multi-step time series, constructing a diffusion process that jointly calibrates both drift and stochastic volatility through a tractable decomposition into conditional transport problems.

In numerical experiments, SBBTS demonstrated superior performance on the Heston model, accurately recovering stochastic volatility and correlation parameters that prior SchrödingerBridge methods failed to capture. When applied to real-world S&P 500 data, the framework generated synthetic time series that significantly enhanced downstream forecasting tasks. Using SBBTS-generated data for augmentation consistently improved model performance, yielding higher classification accuracy and Sharpe ratios compared to training exclusively on real data. These results position SBBTS as a practical and effective solution for realistic financial time series generation, with immediate applications in data augmentation for trading algorithms, risk management systems, and financial forecasting models where data scarcity or quality limitations pose significant challenges.