Signal Live · Premium Trial
Get books →
See courses →
Start trial ↑
Weekly equity picks, governed by a methodology you can audit.
Research quant analysts, not traders.
▸ 397-week walk-forward validated Every Monday at 7am ET, a systematic 9-AI-model panel selects 10 equity picks from a curated 25-ticker universe. Every pick traces to an exact model version and passes a calibration check against a canonical benchmark. If calibration fails, no picks ship. That's the promise.
Framework value-add — 397-week walk-forward validated
SPY baseline (broad market)
+17.81%/yr
+ Universe curation (25 mega-cap quality)
+3.57 pp
+ AI panel selection (over EW-25 baseline)
+3.33 pp
= Total alpha vs SPY
+6.90 pp/yr
We publish two benchmarks every week: SPY (broad market) AND EW-25 (equal-weight of our own universe). Beating SPY is easier — beating EW-25 is the honest test of whether the AI panel adds methodology value. Canonical Sharpe 1.25 vs SPY 0.97 · MDD −30% vs SPY −32%.
What you receive every Monday
✓10 ranked equity picks with confidence scores + panel Agreement Score (per pick)
✓Panel Regime Alert — Frobenius Δ diagnostic, r=+0.23 vs next-week volatility (p<0.0001, N=395 weeks)
✓Full 25×9 model score matrix — every score, every model, every ticker (auditable)
✓4-candidate diagnostic panel with regime-switch monitor (Champion / Consensus / Conviction / Adaptive)
✓Last week's realized returns vs SPY vs EW-25
✓Downloadable PDF for archive
✓Weekly editorial commentary on macro tape + panel behavior
▸ How we're different
Not a copy-trading service. We don't push trades to your brokerage. We publish weekly systematic research — 10 picks, full methodology, walk-forward validated. You place your own trades.
You see the "why" behind every pick. Every pick traces to a specific model version, feature snapshot, and calibration check. If our framework's calibration fails, no picks ship that week.
One methodology, done well. Not a marketplace of hundreds of strategies to browse. Not personality-driven ("invest like a famous trader"). Not entertainment. Systematic research you can audit — closer to institutional quant research letters than a social app.
"Mehrzad brings three decades of quantitative research discipline to Signal Live. The framework is transparent, walk-forward validated, and honest about its regime dependencies." Hossein Kazemi, PhD · Isenberg School of Management, UMass Amherst
Mehrzad Mahdavi, PhD
30 years in quantitative research · Former Executive Director, Financial Data Professional Institute
Start your 30-day Premium trial.
First picks land next Monday. Email-only signup. No password. No credit card.
On day 25, you'll get one email asking whether to continue at $299/yr.
Nothing auto-charges. If you do nothing, your trial ends and you drop to the Free Research Letter (still free, forever).
Nothing auto-charges. If you do nothing, your trial ends and you drop to the Free Research Letter (still free, forever).
Not ready for Premium? Just send me the weekly Free Research Letter →
▸ Also from Digital Hub Insights — Learning Ladder
One framework, three depths. Signal Live is the operational deployment of the methodology taught in our books and courses.
Read — the books ($49 each)
Applied ML in Finance · XGBoost Strategy · foundational theory + interactive notebooks · Podia (Amazon coming soon)
Learn — the courses ($149-$199)
GPT-Augmented Momentum Strategy ($149, includes XGBoost book) · Practical ML in Finance ($199, includes Applied ML book) · full code + walk-forward templates
Watch it run — Signal Live ($299/yr)
Weekly picks + diagnostics. Every Monday 6am CT. The methodology from the books and courses, deployed live with full audit trail. You are here.
Frequently asked
Q: What if I don't like it?
Cancel anytime by replying to any email. Nothing auto-charges without your confirmation on day 25.
Q: How is Signal Live different from copy-trading services?
Copy-trading services give you trades to auto-execute in your brokerage — you're following a trader's decisions in real time, usually without visibility into the "why." Signal Live is systematic research: every Monday you receive 10 ranked equity picks from a nine-model framework, along with the full methodology, diagnostics (Panel Agreement, Regime Alert, Switch trigger), and a 25×9 matrix showing every model's score for every ticker. You place your own trades. The "why" is fully visible. Every pick traces to a specific model version, feature snapshot, and calibration check. If our framework's calibration fails, no picks ship that week.
Q: Are you getting alpha just because you selected 25 high-quality stocks?
Fair question — and we decompose exactly this. Over 397 walk-forward weeks: (1) SPY earns +17.81%/yr baseline; (2) equal-weighting the 25-ticker universe adds +3.57 pp/yr (the "universe curation alpha"); (3) the 9-model AI panel adds another +3.33 pp/yr ON TOP of the EW-25 baseline. So of the +6.90 pp/yr total alpha, roughly half comes from universe selection and half from the AI panel's active picking. The AI panel is validated against the harder benchmark — beating an equal-weight portfolio of the same 25 stocks, not just SPY.
Q: Do you benchmark only against SPY?
No — we publish two benchmarks every week: (a) SPY (broad market baseline) and (b) EW-25 (equal-weighted portfolio of our own universe). Beating SPY is easier because universe selection helps. Beating EW-25 is the real test of whether our model adds methodology value over naive same-universe holding.
Q: Aren't AI models "cheating" since ChatGPT/Claude/DeepSeek already know all these stocks' earnings?
Important — and a common misconception. Signal Live does NOT use large language models (LLMs like ChatGPT, Claude, or DeepSeek) to select stocks. Our 9 models are traditional machine learning classifiers — HistGB, LogReg, Ridge, LinearSVC, Random Forest, XGBoost, LightGBM, CatBoost, and MLP — each trained on quantitative momentum features derived from historical price and volume data. No text, no earnings transcripts, no news, no LLM "knowledge" about the companies. The LLM knowledge-leakage concern applies to research that asks an LLM "which stocks will do well?" — we don't do that. Our models see numerical time-series features and produce a numerical score.
Q: What features do the models use? Time-series or cross-sectional?
Both. We start with 10 base momentum features (multi-horizon returns, risk-adjusted returns, momentum reversals) and derive 10 more (cross-sectional rank versions, acceleration, dispersion-normalized). Time-series features answer "has this stock been rising?" Cross-sectional features answer "is this stock rising more than its peers?" The cross-sectional versions are typically stronger signals because they're less affected by market-wide drift. The full feature list is documented in our ML-for-Finance and GPT-Augmented Momentum courses.
Q: How is look-ahead bias prevented?
Look-ahead bias means accidentally using future information to score past weeks — the single most common way backtests inflate results. We prevent it via three enforced practices: (1) Walk-forward training — every week's model is trained ONLY on data before that week's Friday close. (2) 1-week embargo — the training window ends 1 week before the prediction week to prevent boundary leakage. (3) Immutable feature snapshots — every week's features are frozen at Friday's close and never revised, even if a stock later restates data. The 4 leakage gates shown in every Premium letter verify these constraints hold every week.
Q: Is the strategy sensitive to trade day or time?
The strategy rebalances once per week: picks are formed at Friday's close (using data through that Friday), shipped Monday morning, held for the coming week, and evaluated at the following Friday's close. Actual execution is up to the subscriber — most place orders Monday morning at the open or during the day. Real-time timing within the week is up to you; the framework itself is calendar-based, not intraday.
Q: Why these 25 stocks specifically? Would results hold on a different universe?
The universe was selected in 2019 to satisfy four criteria: mega-cap US equity (top ~2% by market cap for liquidity), sector diversification (no GICS sector exceeds 30%), 5+ years continuous listing history (needed for walk-forward training), and high options-market liquidity. The universe is FIXED — we do NOT rebalance in/out based on which stocks have been winning. No survivorship bias. Whether the +3.33 pp/yr AI-panel alpha persists on a different 25-stock universe is a robustness question on our research roadmap — we'll publish that validation when done.
Q: What markets does Signal Live cover?
25 large-cap US equities — a fixed canonical universe curated for sector diversity and liquidity. Not sector rotation, not international, not options.
Q: Is this investment advice?
No. Signal Live is systematic research for informational and educational purposes only. Not personalized advice. Consult a licensed financial advisor before making investment decisions.
Q: What happens if the framework's calibration fails?
No picks ship that week. You'll receive a notification explaining the failure and when we expect to ship next. Transparency over promises.
Disclaimer: The Signal Live is a systematic research publication for informational and educational purposes only. It is not investment advice, a recommendation to buy or sell any security, or a solicitation. We do not provide personalized investment advice. Past performance does not predict future results. Consult a licensed financial advisor before making investment decisions.