Factor discovery, backtests, and market-stress research in one AI quant lab.
FactorForge is an AI-powered stock strategy research platform for factor discovery, cost-aware backtesting, market-stress analysis, hotspot monitoring, and simulated model-portfolio observation — with every number computed from real OHLCV and clearly labeled when it falls back.
Research only. No order execution, no live trading, no investment advice — a transparent lab for inspecting quant evidence, not a trading account.
New to stocks? Start with Stocks 101 — every term in plain EnglishNo hardcoded returns. Metrics are calculated from OHLCV backtests, optional keys degrade to labeled fallback/template mode, and maintainer docs cover CI, releases, security policy, issue triage, and PR review.
First time on FactorForge? Pick a thread — each surface flows into the next.
Model portfolio vs SPY/QQQ since May — simulated, deterministic, fully labeled.
ExploreOpen a backtest end to end: signals, next-open fills, costs, drawdown, and trades.
ExploreAI-style memo and hotspot signals built from live factor breadth, not vibes.
ExploreCombine ranked strategies and watch diversification and concentration diagnostics.
ExploreRisk-On Regime Active
Constructive tape: 19/28 names hold their 200-day trend with contained volatility (24% annualized). Risk signals are quiet, but drawdown discipline still applies.
Model Portfolio Performance Since May
May 1, 2026 → October 23, 2026 · 122 trading days · 5 active strategies
A deterministic, equal-weighted blend of the platform's top-ranked research strategies, normalized on the start date and compared with SPY/QQQ. Simulated research portfolio only — not a real-money trading account, and historical performance does not indicate future results.
Since May 1, 2026, the simulated model portfolio returned -0.4% versus +6.0% for SPY and +4.7% for QQQ, with a max drawdown of -0.9%. Results are based on research backtests and paper-observation logic, not a live trading account.
Each leg’s own return over the same window. The blended portfolio return above is the equal-weight average of these legs — individual contributions are not a real-money allocation.
Simulated research portfolio — not a real-money account and not investment advice. Equal-weighted blend of 5 top-ranked research strategies, each normalized to 1.00 on 2026-05-01. Based on available historical market data and research backtests, not live execution.
Themes from the demo catalyst engine — proxy baskets and estimated scenario ranges. Research-only.
Yahoo Finance daily bars with provider and fallback metadata.
Momentum, volatility, volume, RSI, and trend breadth snapshots.
Rule-based VCP, ATR breakout, pullback, EMA continuation, and rotation logic.
Next-open fills, fees, slippage, stops, drawdown, and trade logs.
Annualized return, Sharpe, drawdown, win rate, sample size, and risk penalties.
Only radar-approved strategies enter simulated observation; no real orders.
Workbench guardrails
- What it is: an OSS research workbench for inspecting factor and backtest evidence.
- Who it is for: researchers, contributors, and maintainers reviewing quant logic and data provenance.
- What it does: fetches daily OHLCV, computes factors, runs rule-based backtests, ranks candidates, builds portfolio diagnostics, and drafts memos.
- Why it is safe: no order execution, no live trading, and no financial-advice workflow.
- Why it is maintainable: CI, tests, issue templates, release checklist, security policy, and maintainer backlog are documented.
- Fallback policy: real data is preferred; fallback/demo data and template memos are labeled.
| Strategy | Annual | Max DD | Sharpe | Score | Curve | Status |
|---|---|---|---|---|---|---|
| Defensive Trend Pullback AAPL | +3.2% | -2.1% | 1.40 | 81 | radar candidate | |
| Low-Volatility Rotation WMT | +4.2% | -3.4% | 1.26 | 79 | continue observing | |
| EMA Continuation Signal CAT | +5.8% | -5.7% | 1.15 | 77 | continue observing | |
| Quality Momentum Breakout CAT | +4.3% | -2.0% | 1.42 | 74 | continue observing |
Composite stress score is 19/100 from breadth, volatility expansion, momentum, and breadth-of-decline. 19/28 names hold their 200-day trend.
Track whether the regime persists before changing candidate priorities.
Average 20-day realized volatility is 24% and short-horizon volatility is near its 60-day baseline.
Review strategy drawdown and stop behavior before interpreting recent backtest strength.
Average 20-day momentum is 2.5%. Momentum leadership is intact but should be confirmed by breadth.
Avoid over-ranking strategies solely by trailing returns while momentum is unstable.
In a risk-on tape, lower-volatility and trend-holding names (19/28 above SMA200) tend to show improving relative strength as higher-beta leaders de-rate.
Compare defensive vs high-beta strategy drawdowns to confirm the rotation in your own backtests.
Average volume surge is 0.91x the 20-day baseline. Participation looks orderly, but gap risk rises if volume spikes into weakness.
Stress-test stop placement against overnight gaps rather than intraday fills.
68% of the watchlist holds its 200-day trend and 79% is lower today. Participation is reasonably broad.
Confirm whether candidate strategies depend on the few remaining leaders.
AI Research Memo: Market Selloff Review
Constructive tape: 19/28 names hold their 200-day trend with contained volatility (24% annualized). Risk signals are quiet, but drawdown discipline still applies. Composite stress score is 19/100 (risk-on). Breadth is broad and momentum is leading; short-horizon volatility is roughly in line with its 60-day baseline. Conditions support continuation research, but keep stops and position sizing documented in case the regime turns.
Breadth is broad (19/28 above SMA200), momentum is leading at 2.5% avg 20-day, and volatility is normal (24% annualized). Factor signals are not flashing broad stress, but drawdown discipline still governs observation admission.
3 of 5 screened strategies read as resilient and 0 as under stress. Strategies with positive historical returns but high downside sensitivity should be moved from "candidate" to "watch" until volatility normalizes.
Stress-adjusted scoring repriced 3 of 5 strategies versus their base score, penalizing high drawdown and downside volatility while rewarding benchmark-relative resilience and smoother equity. 1 strategies still clear the radar-candidate gate.
No strategy is currently live in paper observation, so there is no simulated book to stress-test this session.
- Re-run the radar shortlist with the stress-adjusted score as the primary sort key and compare observation admissions.
- Measure each candidate's worst 5-day return and benchmark-relative drawdown against the current regime.
- Stress-test stop placement against overnight gaps rather than intraday fills.
- Track whether breadth broadens or volatility keeps expanding before changing candidate priorities.
Generated from deterministic engine metrics (regime, breadth, volatility, drawdown, radar scoring). No order instructions. Historical backtests do not represent future returns. Research only — not investment advice.
19/28 symbols in the default watchlist are above SMA200. Average 20-day return is +2.5% and average annualized 20-day volatility is about 23.7%.
Volatility is within an observable range, but concentration and drawdown still need monitoring.
5 screened strategies behave like ~3.2 independent bets (low overlap).
XOM leads 20d return at +14.1%.