quantlab: a futures research framework

Background

My early CTA strategies came from parameter optimization, and they fell apart out of sample. So I switched to the validation methods of Advances in Financial Machine Learning and built this framework, with tests guarding every layer: data, labels, cross-validation, backtests.

What I did

• Ingested 16 years of CME minute data; built continuous futures (the ETF trick), information-driven bars, features and event sampling.
• Implemented purged K-fold, combinatorial cross-validation, walk-forward and stress tests, plus triple-barrier labels and meta-labeling.
• Ran research with pre-registered criteria, provenance checks, multiple-testing corrections and independent reviews.

Problems and solutions

• Three traps in long histories: single-digit-year contract codes repeat every decade; codes get reassigned to a new contract in their expiry month; the same instrument_id is reused for other products across decades (silver data would land in S&P futures). Contracts are now identified by code plus validity interval, with a guard before anything is written.
• Bit-identical continuous futures: 1,386 rolls across 16 roots, identical whether built offline or driven live bar by bar.
• Batch/live parity: 113 features match bit for bit between batch and online computation; research-side backtests match the Nautilus engine bar by bar on controlled test data.
• Memory: replaced a dense matrix with a difference array and prefix sums for sample-uniqueness weights — from 81.75 GB to 0.10 GB per process.
• Upstream bug: found that Nautilus Trader 1.222’s catalog query silently ignored its bar-type filter.

The honest result

Across 12 CME futures and 16 years, I tested 11 signal families (77 configurations) and 6 trend filters under pre-registered criteria. None showed a robust, significant excess return over a passive long benchmark, so nothing went live. That is what a backtest audit is for: catching overfitting and leakage before real money is at risk.

What I can do for you

• Data pipelines for futures or equities: ingestion, cleaning, continuous contracts, validity checks
• Backtest audits: look-ahead, overlapping samples, cost assumptions, multiple testing
• Backtests and strategy setup on Nautilus Trader