This page is the whole method. Fintch has no proprietary black box: the factors, the weights, the bands and the adjustments are all set out below, and every user gets the same number for the same company.
Publishing it is deliberate. A score you cannot inspect is a score you should not trust.
One number from 0 to 100. Five pillars, each scored 0–100 from its own sub-factors using piecewise-linear bands, then weighted into a composite and adjusted by cross-signal rules. Factors with no data are skipped and their weight is redistributed within the pillar, so partial coverage still produces a score rather than a blank.
| Pillar | Balanced weight | What it measures |
|---|---|---|
| Technical & momentum | 24% | Trend, RSI, MACD, moving averages, position in the 52-week range |
| Valuation | 22% | P/E, P/B, P/S, EV multiples — or a pre-profit track where earnings are negative |
| Quality & balance sheet | 20% | Margins, returns on equity and capital, leverage, interest cover, liquidity |
| Risk-adjusted return | 18% | Sharpe, Sortino, maximum drawdown, beta, volatility, correlation |
| Growth | 16% | Revenue and earnings expansion, multi-year and trailing |
Labels map to bands: 75+ Strong Buy, 64–75 Buy, 52–64 Hold, 42–52 Reduce, below 42 Sell. These name a band on a scale. They are not instructions.
The same indicator means different things in different tapes. An early version scored RSI on a fixed mean-reversion curve, which argued against buying at exactly the moment momentum was strongest — a specification error, not a tuning problem.
Fintch now classifies the tape using Wilder's ADX and directional indicators, then selects the RSI and 52-week-position bands to match:
Averaging factor scores destroys the information in their combinations: cheap-and-rising and cheap-and-falling can average to the same number and are opposite situations. Rules fire on combinations and adjust the composite within a bounded range of −22 to +14, so they can shade a conclusion but never manufacture one.
They include value-with-a-catalyst, possible value trap, buyable dip, falling knife, quality compounder, momentum without support, parabolic extension, balance-sheet distress, an earnings-quality flag, and high or low accruals.
Computed from filed statements and shown as an independent check on the composite: the Piotroski F-Score (nine tests of year-on-year improvement), the Altman Z-Score (distress), the Beneish M-Score (earnings-manipulation indicators), and the Sloan accruals ratio.
Earnings multiples are meaningless when earnings are negative, so for loss-making companies the valuation pillar switches entirely to EV/sales, EV/gross profit, the Rule of 40, and cash runway in quarters. Runway is weighted heavily, because a discounted raise is usually what actually destroys the return.
Fintch also detects when the balanced weighting is simply the wrong lens for a company and says so, rather than quietly producing a structurally misleading number.
The same evidence, three weightings, each with a holding window taken as the minimum of competing constraints — the binding one is named as the reason the call expires.
| Horizon | Weighting |
|---|---|
| Short (days–weeks) | Technical 60, risk 22, valuation 10, quality 5, growth 3 |
| Medium (weeks–months) | Technical 32, risk 20, valuation 20, growth 16, quality 12 |
| Long (1–3 years) | Quality 30, valuation 27, growth 26, risk 12, technical 5 |
Company fundamentals are taken from SEC EDGAR XBRL — filed statements, not a vendor's summary — using the most recently filed value for a period, so restatements are picked up. Price history and quotes come from third-party market-data providers. Analyst consensus, peer groups and earnings dates, where shown, come from a market-data provider and are supplementary; the score does not depend on them.
Stated plainly, because a methodology page that only lists strengths is marketing:
Every indicator — SMA, EMA, Wilder RSI and ATR, MACD, Bollinger %B, ADX and the directional indicators, annualised return and volatility, Sharpe, Sortino, maximum drawdown, beta, Jensen's alpha and correlation — is checked against an independent implementation in pandas and numpy to floating-point precision, as part of the test suite that runs before each release.