How We Score Stocks
Every morning before the bell, Wall Street Analyst reads 30+ live data sources across 1,000+ stocks and scores each on a 15-dimension engine. No single signal decides anything — every dimension is measured, scored, and rolled into one composite. Here is exactly how that works.
The 15 dimensions
RSI, MACD, moving averages, volume surges, mean-reversion and multi-timeframe momentum.
P/E, P/B, PEG, ROE, free cash flow, margins, growth rates and debt ratios.
Trend strength and acceleration across timeframes — what is moving before the crowd notices.
Analyst consensus and revision trends.
Institutional ownership changes, whale detection and conviction scoring.
Unusual volume, implied-volatility skew, put/call ratios and whale-sized positioning.
EPS surprises, revenue beats and post-earnings drift.
Estimate-revision velocity and surprise trajectory, tracked through a second independent earnings feed.
Whether reported earnings are backed by real cash and durable margins.
CEO/CFO/board purchases and sales, aggregated and scored over 90 days.
STOCK Act filings and institutional 13F moves, cross-referenced.
Reddit, StockTwits and social platforms, sentiment-scored.
Factor exposures — value, quality, size, volatility.
Price-target implied upside and target dispersion (conviction).
Return scored against the risk taken to earn it.
From dimensions to a score
Each dimension produces a 0–100 sub-score. Those combine into a single composite score (0–100) — most through adaptive weights, the rest as targeted confirmation bonuses — then rank into a percentile against the full screened universe — a 90 means top 10%. The composite maps to a plain-English signal, STRONG BUY → STRONG SELL.
Convergence detection
When 3+ independent signals cluster on the same stock at the same time — a technical breakout, unusual options flow, and insider buying together — that's a convergence event, flagged automatically. Agreement across unrelated sources is harder to fake than any single indicator.
Scored in the open — including losses
Every call is logged and scored from day one — winners and losers, timestamped, no cherry-picking. The track record is real and built in the open, the opposite of a polished screenshot you can't verify.
Why 1,000 stocks — not 10,000?
Some services advertise scanning 10,000+ tickers. That number should worry you, not impress you: only about 6,000 common stocks trade on the major US exchanges. To claim ten thousand, you have to sweep in OTC pink sheets, shell companies, and micro-caps whose bid-ask spreads would eat a retail order alive — names with no options market to read and barely any volume to analyze. Passing a symbol through a filter isn't analysis.
We do the opposite. The engine screens a universe of 1,000+ names with real liquidity, real options markets, and real institutional footprints — the overwhelming majority of US market capitalization — and scores every one of them 15 dimensions deep, every trading morning. Below that line, data quality collapses and execution costs eat whatever edge a signal might carry. Coverage stops being information and starts being noise.
An inch deep across everything, or a mile deep where the money actually is. When a service leads with the size of its scan, read the fine print under its examples. Ours don't need any — every score is dated, public, and graded on the track record.
Predict less. Re-read more.
Most systems study the past and bet tomorrow looks like it. Ours starts over every morning: the full universe re-scored from live data — earnings, flows, rates, the headlines included. A prediction goes stale overnight. A fresh read doesn't. And every fresh read lands on the public record, graded — on the track record.
Rules we tested — and rejected
Every rule that could change how the portfolios trade gets backtested before it touches the book — and we publish the result either way. Most services only show you the rules that worked. These are ones the data killed.
Mechanical stop-loss enforcement REJECTED · JUN 2026
The hypothesis: automatically sell any position that trips its trailing stop, hard stop, or profit target, instead of letting the scoring engine decide at rebalance. We simulated 12 months of daily enforcement on every strategy, with proceeds held in cash until the next rebalance and the same trading-cost drag as every backtest we run.
| STRATEGY | BASELINE RETURN | WITH STOPS | RETURN IMPACT |
|---|---|---|---|
| Momentum Edge | +223.6% | +122.6% | −101.1pp |
| Value Discovery | +29.6% | +34.6% | +5.0pp |
| Catalyst Trader | +83.7% | +72.5% | −11.2pp |
| Turbo Growth | +51.6% | +46.0% | −5.5pp |
| Blue Chip Fortress | +14.3% | +14.3% | 0 |
The verdict: mechanical stops gutted the momentum book, cut Catalyst Trader's and Turbo Growth's returns, did nothing on the capital-preservation book (its holdings never hit a stop), and only helped Value Discovery. Stops did shave the worst drop on three of the five books — but at the return cost above, and Catalyst Trader's worst drop actually got slightly deeper. The engine's nightly re-score already exits genuine breakdowns, so stops mostly sold winners on pullbacks before recoveries. The risk thresholds remain monitored and reported daily, but the model makes the final exit call. (12-month total-return simulation on current index membership and the July 2026 strategy constructions, data through August 10, 2026 — survivorship-biased, illustrative, not a forward-return guarantee.)
Track-record continuity: the five strategy definitions were sharpened on July 3, 2026 (distinct ponds, hard risk caps, contrarian value pool). Like a fund that changes mandate, each portfolio keeps one continuous record across that date — the performance chart marks the change, with day-by-day sequential tracking of the actual held baskets from July 3 forward and basket-based history before it. We show the seam instead of hiding it.
Turnover dampening (hysteresis) REJECTED · JUN 2026
The hypothesis: keep incumbent holdings unless a challenger beats them by a clear margin, to cut churn. The A/B showed it reduced trading but hurt returns in most configurations — the names it "protected" were the ones the model was right to replace. We shipped a rebalance cadence gate instead, which cut daily churn without touching the model's picks.
Earnings-blackout entry gate REJECTED · JUL 2026
The hypothesis: block new entries within 1–5 trading days of a scheduled earnings report, to sidestep binary event risk the model can't price. We A/B tested four blackout widths on the three momentum-family books over 12- and 24-month windows — 24 gated configurations against baseline. Turbo Growth paid up to −5.3pp of return for the gate; Catalyst Trader and Momentum Edge were coin-flips inside noise (−0.4pp to +2.6pp, no width consistent across windows — the widest gate rejected 2,300+ would-be entries on the daily-rebalance book and barely moved its result). The gate bought no protection either: the worst drawdown was unchanged or slightly deeper in nearly every configuration, with a best case of 0.1pp shallower. Why it fails: these books earn part of their edge from post-earnings drift — skipping the event skips the payoff along with the risk. (Total-return simulations on current index membership, data through July 15, 2026 — survivorship-biased, illustrative, not a forward-return guarantee.)
Wall Street Analyst is financial software and education — not personalized investment advice, and we are not a registered investment advisor. Markets carry risk; past results don't guarantee future ones.