How We Select Our Picks
Every SaturdaySignal signal goes through a 3-stage process. Here's exactly how it works…
Stage 1: Stock Universe Screening
We start with the entire universe of publicly traded U.S. stocks, thousands of securities. From there, we apply basic quality filters to narrow the universe down to the names worth deeper analysis.
What we filter for:
- Robust trading liquidity: only names with institutional-grade depth make the cut, so subscribers can enter and exit cleanly
- Accessible share-price range suitable for typical retail brokerage execution
- U.S. exchange listings (NYSE, NASDAQ)
This initial filter ensures we're only looking at stocks that our subscribers can actually trade: no thinly traded names, no liquidity traps, no names where a single order would move the market. Just stocks with the depth and daily volume to execute cleanly.
Stage 2: AI Quantitative Analysis
100% Quantitative Analysis. No human judgment. No fundamental analysis.
Our proprietary AI-powered system analyzes the pre-qualified universe using purely quantitative methods. The algorithm focuses on identifying potential buying opportunities where multiple quantitative models converge on a favorable setup.
What the AI looks for:
The system is an ensemble of complementary models, each measuring a different dimension of the tape. None of them is asked to predict price by itself. Each one votes, and a candidate only advances when a configured quorum of models agree on the same setup at the same time on the same name. That cross-model agreement, confluence, is the discriminator, and it is the single hardest thing for noise to produce by accident.
The models look at the things a serious systematic trader looks at: cyclical structure, regime classification (trending, ranging, and volatility-driven conditions), volume-price relationships, accumulation and distribution signatures, momentum exhaustion, and statistical reversion behavior. Each input is normalized to its own historical context per name, so a setup on a $40 mid-cap and a setup on a $300 large-cap are evaluated on comparable ground rather than on raw price.
Model weights are adaptive: each model is reweighted by its rolling prediction accuracy. A regime layer running underneath the ensemble means the same raw measurement is interpreted differently in a quiet market than in a volatile one. The point is to make the system fail to fire when the regime is wrong for it, not to find a reason to take a trade in every condition.
The exact model composition, weighting scheme, lookback windows, and thresholds are proprietary and stay that way. What gets published every Saturday is the output the system produces after every gate has cleared.
No emotions. No opinions. Just the measurements and what they agree on.
- ✓ Multi-AI model convergence
- ✓ Machine learning signals
- ✓ Statistical price patterns
- ✓ Volume anomalies
- ✓ Proprietary price patterns
- ✓ Proprietary technical indicators
Stage 3: AI Fundamental Analysis with Human Review
AI does the fundamental work. A human reviews it. The signal ships only when the review confirms.
Stage 2 produces quantitative candidates. Stage 3 layers fundamental analysis on top (performed by a separate AI process running structured research on earnings, filings, recent communications, and the items below), and then a human reviews the AI's fundamental output before the signal ships. The reviewer can override the AI at any point: kill a signal, hold it, or send it back. Final release is editorial, not automatic.
These are considerations, not a checklist. A pre-revenue biotech is not read against the same questions as a mature industrial. A regional bank is not read like a software company. The AI weighs the considerations that actually matter for the name and the setup in front of it, giving weight to whichever factors are load-bearing for that specific case.
The considerations menu:
- Earnings trajectory — trend, surprises, quality of the print
- Recent company communications — press releases, filings, guidance changes
- Growth profile — revenue, margins, the line items that drive the model
- Competitive position — pricing power, market share, share-shift signals
- Management track record — capital allocation, prior guidance accuracy
- Balance sheet — debt structure, runway, working-capital health
- Sector and end-market context — cyclical position, demand drivers, headwinds
- Additional proprietary review factors
No single item disqualifies a candidate by itself, and no single item is sufficient on its own. The fundamental layer reads the whole picture and produces a recommendation. The human reviewer then decides whether to release, hold, or kill the signal, based on the full record in front of him.
This is the third leg of the confluence. Stage 2 reads the tape. Stage 3 reads the business. Stage 3's recommendation has to clear the human review before the signal goes out. The reviewer is not a rubber stamp; he can and does override the AI when something doesn't look right.
The result: only candidates that clear all three stages (quantitative screen, AI fundamental layer, and human editorial review) make it into the Saturday signal.
We're disciplined at our final filter. AI handles the analytical heavy lift; a human signs off before each pick goes out. We value our subscribers, and we want them getting the best of what the model produces.
Position Mechanics
Scaled entries (DCA). When the model identifies an attractive setup at a lower price on a position we already hold, the strategy may issue up to three additional buy signals on that position over the trade cycle, layering up to four tiered entries that improve the average cost basis. ("DCA" here refers to scaling into a position at a better price, also called averaging-down, not strict fixed-interval DCA.) Follow-on entries are flagged as DCA in the signal email and on the track record, and they do not count against your tier's monthly signal allotment.
Maximum Holding Period. Positions close on the model's exit signal or at a hard five-year cap, whichever comes first. The five-year cap is a capital-discipline rule: at the strategic horizon, the position is closed and the capital becomes available for the next opportunity the system identifies. For context, in our published backtest spanning March 2013 - April 2026, the longest hold is 4.95 years — the cap has been approached but never exceeded.
Ready to Put This to Work?
See how the methodology has performed historically, or learn how subscribers use the alerts day-to-day.
Compensation & Independence
WE DO NOT ACCEPT COMPENSATION, PAYMENT, REMUNERATION, OR ANYTHING OF VALUE FROM ANY COMPANY, ISSUER, OR THIRD PARTY IN EXCHANGE FOR RECOMMENDING, FEATURING, OR FAVORABLY DISCUSSING THEIR SECURITIES.
This is our Section 17(b) anti-touting commitment under the Securities Act of 1933. We derive all revenue from subscriber fees. Our independence is the single most important feature of our research.
We trade these signals with our own money, and we wait at least one business day after publication before entering any new Buy Alert (including DCA add-on entries). Subscribers always get first crack at the entry price. You move before we do on every Buy Alert, every time. Sell timing is at our discretion.
SaturdaySignal provides general educational content only, not personalized investment advice. No methodology, algorithm, or investment strategy can guarantee results. All investing involves risk of loss, including total loss of principal. Past performance is not indicative of future results. You are solely responsible for your own investment decisions.