AI Quant turns quantitative research into systematic strategies, tests them under explicit assumptions, and trades selected candidates in a live portfolio. We publish the methods, limitations, failures, and live results.
The portfolio trades in real time through a brokerage account, so this is a live forward record rather than a backtest. Returns are net of the commissions and slippage of the portfolio’s execution, and results may differ from those achievable in another account. Everything here is research and education, not investment advice and not an offer to invest.
Examples of systematic strategies developed and tested by AI Quant Forge. Each case study shows the hypothesis, implementation, results and limitations — not a recommendation or a list of our best-performing models.
New case studies are published here regularly.
The track record we point to is the forward portfolio, because it trades in real time: each result is recorded when it happens and cannot be touched up afterwards. Backtests appear on this site too, but only as separate research examples, because simulated history is far easier to make look good.
Trades in real time through a brokerage account; each day's result is recorded at the close and never revised. Reported returns are net of commissions and slippage.
Explore the full portfolio →Historical simulations of individual strategies, published as research examples. They stay separate from the live portfolio's record.
Paper or hypothesis → implementation → audit and backtest → selection → forward testing → portfolio monitoring. The AI engine reads and structures the research, formalises the testable claims, helps write the code, runs the tests and audits, and records every result with its limitations. Weak candidates get cut along the way, and the ones that survive are managed together as one portfolio.
An AI agent turns a research idea into a concrete strategy: trading rules, instruments, and code that runs.
The code is backtested on historical data, with transaction costs and risk limits applied. Implementation errors are corrected and the test is rerun. A failed hypothesis remains a failed result and may still be published.
Candidates that survive validation may be selected for live trading in the forward portfolio.
Live results are recorded daily and compared against what the backtest promised. Portfolio changes follow the published rules and are timestamped, never revised after the fact.
Why we can publish all this. The core research platform remains proprietary, allowing us to publish its output openly while keeping access controlled.
Choose a daily Radar of newly identified papers or a weekly digest of the most interesting research.