How Equity Rank Valuations Work
Three independent signals. One combined fair value. See the Model Diagnostics section below for live factor measurements and recorded directional outcomes — including unflattering ones.
The Three-Layer Formula
Layer 1 — Fundamental Fair Value
A weighted consensus across 8–15 valuation methods (DCF, P/E, P/B, P/S, EV/EBITDA, Graham Number, DDM, PEG). Each method is weighted by how well it has historically performed in the stock's sector. This gives us the baseline intrinsic value before any market sentiment overlay.
Layer 2 — SAVE Sentiment Overlay
Sentiment, Analyst Consensus, Valuation, Earnings Quality — four independent signals measuring real-time market perception. The composite score (−1 to +1) is multiplied by a sector-calibrated weight (0.10–0.25) before being applied to the fundamental fair value. In the design-stage simulation, the model measured 82.7% directional agreement; live diagnostics below are the authoritative readings.
Layer 3 — Innovation Score Overlay
R&D spending, patent velocity, and CapEx discipline vs sector peers. Score 50 = exactly sector average (no adjustment). Higher scores lift fair value for leaders; lower scores apply a discount for laggards. Sector-calibrated alpha range: 0.05 (Utilities) to 0.25 (Biotechnology). In the design-stage simulation, the model measured 64.8% directional agreement; live diagnostics below are the authoritative readings.
Combined Formula
The two signals have near-zero correlation (r = −0.005), meaning they measure genuinely different things. Combining them adds real information rather than doubling down on the same bet.
Model Diagnostics: We Show Our Work
Every day the platform measures whether each factor in the model has actually been predictive, and the readings below are published as-is — including when the answer is no. The information coefficient (IC) is the rank correlation between a factor's scores and stock returns over the following days. It ranges from −1 to +1; a value near 0 means the factor had no predictive relationship over that window.
Why Combine Three Signals?
Redundancy
If one signal is unavailable, the other still works. No black box — each layer has a defined, transparent formula.
Independence
SAVE measures market perception; Innovation measures operational quality. They capture different things, so combining them adds real information.
Calibrated by Sector
Energy stocks get a lower SAVE weight (0.10) because sentiment is less predictive there. Biotech gets a higher Innovation alpha (0.25) because R&D is core.
What Goes Into Each Layer
| Signal | Key Inputs | Source | Update Frequency |
|---|---|---|---|
| SAVE | News tone, search velocity, analyst shifts, earnings call language | AlphaVantage + Google Trends | Live (7-day window) |
| Innovation | R&D/Revenue ratio, patent velocity, CapEx vs peers | SEC filings + AlphaVantage | Weekly batch |
| Fundamental | EPS, book value, FCF, revenue, dividends, growth | AlphaVantage (Supabase cached) | Daily |
See combined valuations for any stock
All three layers applied instantly. Fair value + SAVE overlay + Innovation adjustment.
Start Free Trial — See combined valuations for any stockPer-Method Valuation Derivations
Every method in the consensus blend. Each entry shows the textbook formula, the inputs we use, the assumptions we make, the known limitations, and the sector weights that govern when it counts in the consensus.
1. P/E (Price-to-Earnings) Fair Value
Fair Value = TTM EPS × Sector P/E Multiple
Inputs: trailing-12-month earnings per share, sector default P/E multiple (Damodaran-anchored, recalibrated 2026-05 in Phase 45 BIAS-03). Assumes the earnings base is representative — fails on one-time charges, restructuring quarters, and pre-revenue companies. Limit: sector multiple is a mean, so it under-fits hyper-growth (NVDA-tier) and over-fits restructure (LSPD-tier) — the forward variant below is more reliable for non-mature names.
2. Forward P/E Fair Value
Forward EPS = analyst consensus next-year EPS (preferred), or TTM EPS × (1 + growth_yoy)
Fair Value = Forward EPS × Sector P/E Multiple
Phase 45 D1 finding: forward methods were structurally missing from the screener batch. Wired in 2026-05-26 — every ticker in a forward-eligible sector now has this contribution. Analyst path is preferred (via yfinance forwardPE); falls back to a modelled forward EPS when consensus is unavailable.
3. P/B and Justified P/B
P/B Fair Value = Book Value Per Share × Sector P/B Multiple
Justified P/B = BVPS × min(ROE / Cost of Equity, 30) — capped to prevent extreme outliers
P/B is the asset-side anchor. Justified P/B is the formula-based variant from Stowe et al. (CFA Curriculum L2): a stock's P/B should equal its ROE divided by its cost of equity. Both anchor banks, REITs, and asset-heavy industrials. Cost of equity uses CAPM with a 1.5 beta ceiling (Phase 45 D2, Damodaran Investment Valuation 3e Ch 7) to prevent runaway discount rates on high-beta growth names.
4. P/S and Forward P/S
P/S Fair Value = Revenue Per Share × Sector P/S Multiple
Forward P/S = RPS × (1 + revenue_growth_yoy) × Sector P/S Multiple
Useful when earnings are negative or distorted but revenue is real and recurring. Critical for SaaS, hyper-growth software, and any business model where the company is deliberately reinvesting profit (Amazon era, pre-FCF Tesla). Sector multiple recalibrated in Phase 45 BIAS-03 for Internet Platform + Communication Services.
5. EV/EBITDA and Forward EV/EBITDA
Fair Value = (EBITDA Per Share × Sector EV/EBITDA Multiple) − Net Debt Per Share
Forward = (Forward Revenue × TTM EBITDA Margin × Sector Multiple) − Net Debt
Capital-structure neutral — strips out interest expense and depreciation, useful in cross-company comparisons where leverage and depreciation policies differ. Net debt conversion turns EV → Equity Value. Forward variant holds the TTM margin flat as a conservative no-expansion assumption.
6. DCF and Three-Stage DCF
Discount rate = rf + min(beta, 1.5) × ERP (CAPM with Phase 45 D2 beta cap)
DCF Fair Value = Σ FCF_t / (1+r)^t + Terminal Value
Three-Stage: high growth → linear deceleration → terminal perpetuity (Damodaran)
The intrinsic-value anchor. Beta capped at 1.5 for CAPM — industry-standard practice (Damodaran Investment Valuation 3e Ch 7) — because uncapped raw beta on hyper-growth names (ARM β=3.41, TSLA β=2.2) produces 18-23% discount rates that pulverize future FCF. Three-stage handles transitional decay realistically; single-stage Gordon is a quick check.
7. DDM (Dividend Discount Model)
Fair Value = DPS × (1 + g) / (r − g) (Gordon Growth)
Values the stock as the present value of all future dividends. Only meaningful for established dividend payers (utilities, REITs, banks, mature consumer staples). Sector weights set DDM to ~0 for Cloud Infra / AI Platform / Biotech / EV where most companies don't pay dividends.
8. Graham Number
Graham = √(22.5 × EPS × BVPS) (Benjamin Graham, The Intelligent Investor)
Maximum price a defensive investor should pay. Very conservative; ignores growth. Useful as a floor-value sanity check on value names; sector weights set Graham ~0 for growth sectors where the formula is structurally too conservative.
9. PEG Fair Value
PEG Fair Value = EPS × PEG Ratio × Growth Rate %
Peter Lynch's growth-adjusted P/E rule: a fairly valued stock has PEG ≈ 1 (P/E equals its growth rate). Growth rate capped at 35% to prevent geometric blow-ups on hyper-growth tail. PEG returns DataUnavailableError on zero/negative growth.
10. EPV (Earnings Power Value)
EPV = Normalized EBIT × (1 − tax rate) / Cost of Equity (Bruce Greenwald)
Assumes zero growth — "what is this company worth if it never grows again?" Conservative stress-test floor. Not a price target. The cross-method sanity filter at runtime (ratio FV / price within [0.15, 5.0]) excludes EPV outputs that fall outside that window from the consensus blend — Phase 45 D5 verified this already mitigates the "EPV drags growth names down" concern.
11. P/FCF + EV/FCF
P/FCF Fair Value = FCF Per Share × Sector P/FCF Multiple
EV/FCF = (FCF/Share × Sector EV/FCF Multiple) − Net Debt/Share
FCF-based multiples for the cohorts where reported earnings are noisy (asset writedowns, deferred revenue swings) but cash conversion is strong (mature SaaS, dividend-yielders, tobacco, consumer staples).
12. EV/EBIT
Fair Value = (EBIT/Share × Sector EV/EBIT Multiple) − Net Debt/Share
More conservative than EV/EBITDA — includes depreciation and amortization as a real cost. Better for capital-intensive industries (homebuilders, freight, heavy machinery, steel).
13. FFO (Real Estate / Equity REITs)
FFO = Net Income + Depreciation & Amortization on real property (NAREIT standard)
Fair Value = FFO × Sector P/FFO Multiple
For equity REITs, GAAP EPS understates earnings because non-cash D&A on real property is meaningless. FFO restates on a cash basis. Restricted to Real Estate / Tower REIT / Data Center REIT sectors.
14. EV/Sales
Fair Value = (Revenue/Share × Sector EV/Sales Multiple) − Net Debt/Share
Capital-structure-neutral revenue multiple. Useful for pre-profit names where P/S is meaningful but you want to net out cash and debt. Weighted lightly (0.03) in most sectors as a tie-breaker.
15. Options-Implied Forward Price
F = K + e^(r·T) × (C − P) (put-call parity)
Backs out the market's implied forward price from live options chains (puts and calls at the same strike + expiry). No fundamental assumptions — pure real-time market consensus. Useful as a complementary signal to the fundamental methods above.
16. Reverse-DCF (Implied-Growth Lens) — coming Phase 50
Solve for g such that: Current Price = Σ FCF × (1+g)^t / (1+r)^t + Terminal
Instead of using the model's growth assumption to find fair value, solve for the growth rate the current market price implies. Displays "Market implies X% growth; the model assumes Y%" as a sanity check on every valuation. Implementation lands in Phase 50.
17. Residual Income (RI)
V₀ = BV₀ + Σ [RI_t / (1 + r_e)^t] + Gordon Terminal Value
RI_t = (ROE_t − r_e) × BV_{t-1}
BV_t = BV_{t-1} × (1 + ROE_t × retention)
Terminal: RI_5 × (1 + g) / (r_e − g), capped when r_e − g < 2% → TV = 0
Formula derivation (Ohlson, 1995): The Residual Income model values equity as book value plus the present value of future excess returns — earnings above the cost of equity multiplied by the book value base. This is mathematically equivalent to the DCF model under clean surplus accounting, but far more stable for companies where free cash flow is volatile: banks, insurance, capital-intensive industrials, utilities, and mortgage REITs.
Model structure: A 5-year explicit stage uses trailing ROE linearly fading toward the cost of equity (competitive equilibrium assumption). Year-5 excess returns carry into a Gordon-growth terminal stage under these assumptions: g ≤ 2.5% and r_e − g ≥ 2% (perpetuity blow-up guard). Book value rolls forward each year using the retention ratio (1 − dividend payout). The model fair value is clamped to a 2.5× upper ceiling relative to current price.
Sector applicability: Strongest for Banks, Insurance, Mortgage REITs (weighted 0.20 in consensus); useful for capital-intensive Industrials, Utilities, Energy (0.08–0.10); smallest weight for asset-light Technology and Software (0.04) where intangibles dominate book value. Returns no model estimate for negative book value (leveraged buybacks, accumulated losses) or when book equity is immaterial relative to market price.
Limitations: Assumes clean surplus accounting (no off-balance-sheet items distorting book value). Terminal RI = 0 is deliberately conservative — quality compounders with sustained excess returns will show a model fair value differential below intrinsic value under this assumption. ROE figures are trailing; forward-ROE forecasting is not yet incorporated.
80% Confidence Intervals (Phase 47)
Every per-method fair value and the consensus blend carry an 80% model interval — a band that combines two uncertainty sources. Not a forecast or expected range.
spread_A = IQR(per-method FVs) × 1.282 / 1.349 (cross-method disagreement)
spread_B = consensus_fv × decile_multiplier × 0.05 (sector volatility)
half_width = √(spread_A² + spread_B²)
80% interval = [consensus_fv − half_width, consensus_fv + half_width]
spread_A uses IQR (Q3 − Q1) — robust to DCF terminal-value tail outliers (which inflate standard deviation). The 1.282/1.349 conversion turns a normal-distribution IQR into a 80% z-score one-tail half-width.
spread_B uses sector volatility deciles calibrated 2026-05. Utilities and consumer staples → multiplier 1.0 (5% half-width per FV). EV Technology, Biotech, AI Platform → multiplier 3.0 (15% half-width). Default multiplier 1.5 for unlisted sectors.
Edge cases: if only 1 method fires, spread_A falls back to 15% of consensus_fv. If fv_low would go below zero, it clamps at 0. If half_width is below 2.5% of consensus_fv, it floors at 2.5% (zero uncertainty is not credible).
What We Never Claim
Equity Rank is not a registered investment adviser (federal or state). Per IAA §202(a)(11), FINRA, and SEC anti-fraud rules, we never use the following kinds of language and have an automated CI gate (Phase 46 LABEL-04) blocking it from any UI, email, or AI prompt:
- "buy" / "sell" as verbs directed at the reader — we use overweight / underweight / attractive / unattractive
- "you should", "we recommend", "our recommendation" — we use "the model surfaces" / "model output"
- "price target", "expected return", "upside potential" — we use "projected fair value range" / "model fair value differential"
- "will outperform", "guaranteed to", "beat the market", "generate alpha" — we describe relative model output, not forward returns
- "forecast horizon" — we use "model horizon"
- Firm-name comparisons (Bloomberg, Goldman, Morgan Stanley, JPMorgan) as quality claims
Document Provenance
Authored by the Equity Rank engineering team
Methodology version: v2.0 (Phases 44–49 substantive ship)