Proprietary research software, live capital testing, and risk reporting across liquid markets. Signal logic stays private. Method and evidence do not.
Cumulative percentage return from inception. All figures reflect live capital test results, not backtested performance. Drawdown measured peak-to-trough intra-period.
Each cell represents one trading day. Colour intensity corresponds to magnitude. Data covers combined Multibot and VWAP strategy P&L. No absolute values shown.
Signal status reflects current model output across active strategies. Instrument coverage and strategy assignment change with regime conditions.
Python-based research stack for data ingestion, feature construction, hypothesis testing, and backtest evaluation. Every notebook is versioned. Every published figure traces to a commit.
Separate reporting workflow journals every test trade, tracks drawdown against pre-declared limits, publishes monthly. Reports are plain language, archived with their underlying data.
Selected research is forward-tested with capital at small size under the same reporting discipline. No client capital, no solicitation, no investment products.
Statistical learning models evaluate regime context and generate position signals. Every output is auditable against its input features.
LLM-assisted synthesis of market conditions, published daily. A reading tool, not a trading tool.
Automated drawdown checks, correlation monitoring, and regime detection feed into position sizing logic.
We do not predict markets. We measure the conditions under which our models perform, and size accordingly.
Monolith ResearchThe latest research note takes the Top-5 cross-asset OU sleeve into a stricter final holdout. The selected turn filter gives up a little return, keeps the path quieter, and records a 2.19 Sharpe with maximum drawdown held to -0.93% after stressed costs.
The cleaner statistic is the one that matters: less violence in the path, nearly the same destination. From the final OOS noteContinue reading →
The turn filter lifted held-out Sharpe from 1.62 to 1.83 while reducing maximum drawdown from -1.20% to -1.00%.
Six stressed-cost folds: 1.216 average Sharpe, 47.32% summed test return, and a worst fold of -3.19%.
The broad FX transfer failed; the XAU-heavy branch survived as a separate modest candidate with 7.63% total return.
Thirty-six market-tone features were built; AUC stayed near coin-flip, so the branch remains a warning layer rather than live alpha.
UK-based researcher and software builder, studying liquid markets across ETFs, FX, and commodities. The desk runs on a small, deliberate stack: a Python research environment for data ingestion and hypothesis testing, and a companion reporting workflow that journals every test trade against pre-declared limits.
Working skills: Python, data analysis, backtesting, risk reporting, systematic research. The interest is method first — building the kind of evidence record that a careful reader can audit a year later, not the kind that survives only while the screenshots are fresh.
Current focus: building the research infrastructure and the live capital evidence record that supports it. The company does not employ a sales function, does not raise capital, and does not publish performance projections. What it publishes, it can defend.
Meet the teamCorrespondence welcome — for research, technology, and principal-backing conversations.
bilal@monolithresearch.uk
Monolith Research