walk-forward-validation

A quantitative trading strategy backtester with an interactive dashboard. Enables users to implement, test, and visualise trading strategies using historical market data, featuring customisable parameters and key performance metrics. Developed with Python and Polars.

2026 FIFA World Cup prediction research system — 7-component fusion (Dixon-Coles, Elo, Pi-Rating, Weibull, Market, Stacking, Conformal), auditable walk-forward backtesting, bilingual reports

Multi-agent trading platform on an Alpaca paper account: a CLI places orders on request; scheduled routines manage trailing stops, alerts and daily records. Agents on Microsoft Agent Framework. Research uses walk-forward, out-of-sample and multiple-testing checks; no tested signal (MACD+RSI, TSMOM, GEX) has an edge yet. Not financial advice.

ML portfolio: Persian RAG assistant (hybrid BM25+RRF retrieval, RAGAS-style eval), stock movement with walk-forward backtest, recommender system, AI-text & SMS-spam classifiers — all with tests, CI and real metrics

Kiploks Trading Robustness Engine is an open-source TypeScript engine for deterministic backtest and walk-forward analysis (WFA) of algorithmic trading strategies, published as @kiploks/engine-* packages under Apache 2.0.

Reproducible Python time-series forecasting benchmark with walk-forward validation, leakage-safe backtesting, classical models, lag-based machine learning, tests, and CI.

Calibrated multi-sport forecasting, audited in public. Live analytics site + a claim-by-claim evidence index with the rejects and retractions included. Calibration rigor, not edge claims.

End-to-end ML system for prediction market trading — 521K markets, 78 features, 7 model architectures, walk-forward validation, live VPS A/B across 7 configs. Honest research-stop on alpha decay (NO-GO verdict). AFML methodology: Purged K-Fold, Deflated Sharpe Ratio, meta-labeling, focal loss.

A systematic intraday long-only strategy on XAU/USD using volume profile Point of Control (POC) retest signals, validated on 8 years of 5-minute MT5 data.

NexaBask — Production-grade basket trading system with risk-first architecture, MT5 integration, walk-forward validation, and ML/LLM roadmap

Historical RV15 forecasts for six U.S. stocks. Option state improves both families; the mixed B2 block improves the linear model only. Source-time assumptions, placebo and final-window limits remain material. Public aggregate checks are not an independent licensed-data replication.

ML-based edge extraction framework for MT5 — mean reversion, pairs, volatility edges. Walk-forward + bootstrap validation. HTML reports.

Machine-learning trading research and paper-inference engine for ETH market-data experimentation.

ML investigation into whether public data can generate profitable betting signals on European football markets

Temporal partitioning and backtesting for time-correlated datasets

Pre-game NFL win probability, spread and total forecasts: walk-forward models, a leakage-proof day-before information lock, pre-registered evaluation, and a broadcast-style FastAPI + HTMX dashboard.

Walk-forward-validated ML pipeline for 21-day swing-trade selection across AI/semiconductor stocks, with an options-strategy overlay — benchmarked honestly: the edge is sector beta, not stock-selection alpha.

Multi-model time-series forecasting with Bayesian Optimisation (Optuna TPE): SARIMA, Random Forest, XGBoost, LightGBM, Prophet, LSTM, and QuantileML probabilistic forecasts behind a unified ModelSpec protocol. Walk-forward validated; supports monthly, weekly, daily, and hourly data.

PPO reinforcement learning agent for multi-asset portfolio allocation, evaluated with walk-forward validation, multi-seed reporting and a shuffled-price control

F1 pole, winner and podium predictions — published before every race, scored in public, and measured against baselines a one-line rule could beat.