non-stationarity

📈 Synthesize financial time series data adaptively for improved analysis and forecasting using a robust dataflow system.

Calibrated GPD tail inference in Python: continuous and grouped fits, profile intervals, permutation trends, Monte Carlo calibration, and multi-source robustness.

End-to-End Python implementation of bi-level optimization for financial time-series synthesis from "History Is Not Enough" by Xia et al. (2026). Implements cointegration-aware data augmentation, curriculum learning, and meta-learned augmentation policies to immunize quantitative models against concept drift and market non-stationarity.

Most associations in this public semiconductor dataset are artifacts of time. 32 features clear BH-FDR with permutation p=0.0005, but only 2 survive a change of time period. Chronological split, DuckDB/SQL warehouse, SPC, cost-aware baseline, every transform fitted on train only.

Non-stationarity breaks permutation surrogates in multi-agent RL: analysis code and data

Regime-aware memory for agents operating in recurring environments

Adaptive Financial Risk - Self-adaptive forecasting under regime change: drift detection, rolling recalibration and walk-forward backtests