Models, engineered.
Research built to be challenged. We test market hypotheses, publish the evidence, and release the code to reproduce it.
Questions, specified.
We start with precise questions about market behavior, model assumptions, and measurable outcomes. Each study defines its terms before the data can shape the story. The result is research that stays focused, falsifiable, and open to inspection.
Models, challenged.
Models are tested beyond the data that shaped them. We compare them with simple baselines, evaluate them across changing regimes, and measure where their conclusions hold or fail. A result matters only when it survives a serious challenge.
Evidence, inspectable.
The method stays connected to the result. Papers, diagnostics, and source code make the work easier to verify, reproduce, and question. Clear records turn each finding into something others can build on.
01
What we publish.
M5 Research is an independent lab working on quantitative methods for financial markets. We publish working papers, their methods, and reproducibility evidence. Where code is available, the archive links to it.
Infrastructure| Date | Title | Domain | Status |
|---|---|---|---|
| Certifying Regime Detectors Before Use | Volatility | Working Paper | |
| Penalty-Based No-Arbitrage Enforcement for Neural Volatility Surfaces Under Sparse Strike Data | Volatility | Working Paper |
From paper to evidence.
A result you can trace.
Explore the paper, implementation, and certification records behind our regime-detector research.
Reported results can be reproduced from included artifacts. Regenerating artifacts requires separately obtained vendor data.
Inside the infrastructure.
m5-backtest-engine
Python strategies and support code for replaying market data through the C++ backtest engine.
Explore infrastructure