▲ stockoverlap.com

Methodology & data sources

Last updated October 2026

This page exists so you don't have to take a score on faith. It documents exactly where the numbers come from, how often they update, how the combined score is calculated, and where the limits are.

Data sources

Scores are built from three providers. Financial statements (revenue, gross profit, net income, assets, debt, equity and EPS — the basis for ROE, earnings growth, margins and leverage) come from SEC EDGAR filings. Analyst price targets and ratings, earnings surprises, insider transactions and sector labels come from Finnhub. Prices and price history, market cap, cash flow (for free-cash-flow yield), short interest and institutional ownership come from Yahoo Finance, which is also the fallback when EDGAR or Finnhub is missing a figure. We do not use fund-holdings data — StockOverlap screens individual stocks on fundamentals; it does not compute ETF/fund overlap.

Update frequency

The scan runs once per trading day, in the evening after the US market close (Monday–Friday, skipping weekends and running on whatever data is available around holidays). There is no midday scan. Every notification email and the terminal header show the scan's actual timestamp so you always know how fresh a score is. Between scans, open picks' prices and upside % refresh live from Yahoo Finance each time you load the page — fundamentals themselves (score, grade) update on the next scan. Fundamental data is cached for up to about four days, because it only changes when companies report.

How the combined score is calculated

Two independent 12-point screens run against every stock in the scan universe, each built from individually scored factors so the total is reproducible rather than a black box:
  • Return on equity — up to 2.0 pts (≥15% worst reported year)
  • EPS growth (CAGR) — up to 2.0 pts (≥10%)
  • EPS trend consistency (R²) — up to 1.0 pt (≥0.85)
  • Gross-margin stability (variability, lower is better) — up to 2.0 pts (≤5% variation)
  • Valuation vs. sector (P/E ratio) — up to 1.0 pt (at/below sector median)
  • Balance-sheet strength (debt/equity) — up to 2.0 pts (≤0.50)
  • Insider buy/sell balance — up to 1.0 pt (net insider buying ≥25% of insider share volume)
  • Last two quarters' earnings surprises — up to 1.0 pt (both beats)
Each factor scores in tiers between its floor (0) and the max shown above, not just pass/fail — a stock at 12% ROE earns partial credit, not zero. Only stocks trading within ±5% of their 200-week moving average are graded, and a stock needs at least 6 of 12 points to stay on this screen. Stocks that fail a hard floor are dropped entirely regardless of their other scores: zero or negative ROE in any recent year, EPS shrinking more than 2% a year, debt/equity over 2.0, more than 20% of float sold short, insider selling above 90% of insider volume, or both of the last two quarters missing estimates by more than 10%. Short interest is not scored on this screen — it only applies as that knockout here, and is scored on the value screen below.The value screen runs on every stock in the universe, regardless of the 200-week average, but only considers stocks priced at $5+ that are not in the top 12% of their 52-week range. It is scored the same way — nine factors, each in tiers up to a max — and a stock needs at least 5 of 12 points. Floors drop a stock entirely if breached (revenue growth below -5%, gross margin under 8%, debt/assets over 0.80, or EPS CAGR below -20%):
  • Revenue growth (YoY) — up to 2.0 pts (≥15%)
  • Asset growth (YoY) — up to 1.0 pt (≥10%)
  • Free-cash-flow yield — up to 2.0 pts (≥6% of market cap; operating cash flow is used when free cash flow isn't available, which overstates the yield for capital-intensive companies)
  • Earnings yield, inverse of P/E — up to 1.0 pt (P/E ≤15)
  • Position in the 52-week range — up to 2.0 pts (in the bottom 30%, i.e. closer to the low)
  • 52-week range tightness — up to 1.0 pt (high-to-low spread ≤40%)
  • Short interest (% of float) — up to 1.5 pts (≤4%)
  • Institutional ownership — up to 1.0 pt (≥65%)
  • Analyst sentiment not already euphoric (consensus rating not near unanimous "strong buy") — up to 0.5 pt
A stock becomes a dual pick only when it scores on both screens the same day, the combined score (sum of both, out of 24) is 15+, and consensus analyst upside is 30%+ for the live Dual Picks list in the terminal (the public track record logs every otherwise-qualifying stock from 25%+ upside, so it is slightly broader). A combined score of 18+ earns the top-tier badge. This is not an overlap-of-holdings calculation — it's agreement between two independently-run fundamental screens on the same ticker.

Historical accuracy — what's frozen vs. live

When a pick first appears, its entry price and entry date are frozen — that's the trade record (the only adjustment is for stock splits). While a pick is open, the score, grade, and analyst target shown are the current scan's numbers wherever the latest scan still covers the stock, not the ones from entry day — so you see where it stands today. Return % is always calculated from the frozen entry price to the live (or, once closed, the exit) price. Closed picks keep their exit price and are not re-priced. If a closed pick later qualifies again as a fresh Dual Pick, it is logged as a new pick with that day's entry date and price, and its earlier closed record keeps its own entry and exit. (Until October 2026 the earlier record was reopened instead; those exits have been restored.)

How a pick is closed

Every open pick is re-scored on both screens after each scan, even if it no longer passes an entry filter (the 200-week band, the 52-week range gate, the value cutoff, a knockout, or the market-cap and price floor). If a score can't be computed on a scan because data is missing or a data fetch failed, the pick keeps its last score and stays open; missing data never closes a pick on its own. The scanner closes a pick by rule, in one of three ways:
  • Delisted — the price data shows no trades in the stock for 5 trading days, for example after an acquisition or a delisting. It is closed at its last traded price; if the price data no longer returns the stock at all, at the last price a scan recorded for it (or its entry price if none was ever recorded). A scan with unusually little data never closes a pick this way, and if more than two picks, or more than 10% of open picks, would close this way on the same scan, none of them are closed; a single delisting on its own always closes.
  • Score falls from its peak — its combined score stays 3+ points below its peak for 3 consecutive scored scans, unless it still qualifies as a new Dual Pick that day. A run where the whole book's scores shift at once (see Corporate actions & methodology changes) resets that count, and a scan with unusually little data doesn't advance it.
  • Target reached — while the score rule isn't firing, its price comes within 2% of the consensus analyst target, unless it still qualifies as a new Dual Pick that day. No multi-scan confirmation is needed.
The score and target rules only close a pick that was last held open within the past 60 days, so an old backlog isn't closed with today's date. Drifting away from the 200-week average, failing another entry filter, or analyst upside falling below the entry bar does not close a pick. Exit rules have changed over time, so older closed picks may reflect earlier rules. Track-record figures are hypothetical results of a rules-based screen, not actual trades.

Data-quality limits

Market data providers occasionally return delayed, missing, or incomplete fields — this is a real limitation of any third-party feed. Where sector or company name is unavailable for a ticker, we fall back to the most recent known value rather than showing blank data, which we flag here for transparency; we do not claim 100% completeness on every field, every day. If a provider errors on a ticker, the scan falls back to another source where one exists. Values that are still missing are not guessed: they earn no credit, or a neutral score for a few quality signals such as insider activity and earnings surprises.

Survivorship & look-ahead bias

The scan universe is rebuilt daily from a live market screener, not a fixed list — a stock that gets delisted, acquired, or goes bankrupt simply stops appearing in future scans rather than being retroactively removed. Since October 2026, the logging code never deletes a pick from the public track record, whether or not the ticker is still in the current universe. Before that, the record was edited by hand: in July 2026, 6 closed picks were removed and 17 closed picks were reopened (disclosed on the track record page). Separately, until October 2026 a closed pick that qualified again had its record reopened, which took its earlier exit off the public closed list and out of the realized stats; those exits were restored in October 2026. Live daily scans only ever use data available as of that day, so the scores you see were not computed with hindsight. Backtested performance numbers (used in our research, not shown as a live promise) are a different case — they apply today's analyst targets and current company fundamentals against historical dates, not the targets/fundamentals that actually existed then. That is a real source of look-ahead bias in backtests specifically, and any backtest figures we reference should be read as directional, not as a clean historical simulation.

Corporate actions & methodology changes

Returns are automatically adjusted for stock splits — a split can make a flat position look like a large loss or gain if the entry and current/exit prices aren't on the same share basis, so we check each open and closing pick against real historical split data before computing its return.

Returns are price-only and do not include dividends. For dividend-paying stocks this modestly understates the true total return an investor holding the actual shares would have earned.

Delisted, acquired, or bankrupt companies are never silently dropped from the track record: an open pick is closed once the price data shows no trades for 5 trading days, at its last traded price (or, if the price data no longer returns the stock, the last price a scan recorded for it) — see How a pick is closed. A pick that simply can't be scored for a while because of a data gap stays open with its last score.

If the scoring model or rubric itself changes, every open pick's score can shift at once — that looks like a data event, not individual businesses deteriorating. We detect this (the whole book's scores moving together in the same run) and skip score-drop exits for that run, resetting each pick's weak-score count, rather than closing a batch of picks purely because the yardstick changed; normal exits resume on the next scan.

Analyst-target reliability

The 30%+ consensus-upside requirement for the live list (the track-record log uses 25%+) is a filter, not a guarantee — analyst price targets are opinions, revised often, and not an independent predictor of future returns on their own. In backtesting, the 30% threshold (vs. a looser 25%) showed a similar win rate and average return, meaning the extra filtering mainly narrows the list rather than materially improving odds — and, per the caveat above, that backtest itself carries look-ahead bias since it used current analyst targets against past dates. Treat the upside requirement as one input among the screen's several factors, not as a standalone signal.

Independent validation

We don't ask you to trust the score in isolation — every dual pick, win or loss, is logged the day it's called, and the public track record page counts every pick — open ones marked to market and after an assumed trading cost — against SPY over the same holding windows; closed picks are listed there with their frozen entry price, exit price, and return (open picks are summarized without tickers). The record isn't untouched: in July 2026, 6 closed picks were removed and 17 were reopened, by hand; until October 2026 a closed pick that qualified again was reopened at its original entry price (those exits have since been restored), and exit rules have changed over time — all disclosed there. That history — not a marketing win-rate number — is the actual evidence for whether the methodology works.

Not financial advice

This is a research and screening tool. Scores and picks are not a recommendation to buy, sell, or hold any security. See our Terms of Service for the full disclaimer.
Related guides
Methodology & Data Sources — StockOverlap