Stock Analyzer

📊 Valuation Methodology

Our intrinsic value model builds company fundamentals only from publicly available SEC EDGAR filings. The only non-EDGAR inputs are macro rates — the 10-year Treasury yield and FRED inflation expectations — used in the discount rate. Market prices never enter the valuation; they appear only for comparison (the hype meter and benchmark). This ensures our valuations are independent and can identify market mispricing.

🔬 Core DCF Framework:

• Risk-Free Rate: the live 10-year U.S. Treasury rate, fetched daily from the U.S. Treasury's published par yields (falling back to FRED DGS10, then a last-good cached value, then a fixed last-resort constant) — shown in the discount-rate inputs above. A degraded source is flagged on the print and lowers its reliability grade. This is the "time value of money" baseline.
• Discount Rate (r = g + spread + K): We build the rate up from terminal growth plus a fundamentals-based risk spread, shifted by a calibration constant K struck so the median S&P operating name lands on a market-neutral, bond-anchored baseline (currently 6.24%, struck 2026-07-04), then floor it at that anchor (and never below Risk-Free + 1.5pp). No beta, no CAPM, and no market prices anywhere in the rate.
• Cash-Flow Base (owner earnings): the stream we perpetuate is operating cash flow minus maintenance CapEx, taken from one coherent window — reconciled trailing-twelve-months when every stream passes the checks, else the audited fiscal year, never mixed; cyclical names use a mid-cycle 3-year average. Suspected one-off cash receipts (a single-period receipt ≥15% of the base) are flagged, never silently subtracted.
• Projection Period: 5 years of detailed cash flow projections, then a terminal value for all future years (the insurance lane uses a 10-year fade).
• Terminal Growth Rate (g = inflation + pricing power): Long-run inflation expectation (FRED 5y5y forward; a ~2% fallback is flagged) plus a pricing-power score from gross-margin stability and above-inflation real revenue growth. Bounded −1% to 3.5%. Durability routing decides whether the terminal grows, holds flat, or declines ("harvest") — growth the filings haven't corroborated is not credited in perpetuity — and terminal growth is never free: a reinvestment charge (the g = reinvestment × return identity, run in reverse) pays for it.

📈 Growth Rate Prediction (15-Year Historical Analysis):

We analyze 15 years of SEC filings to dilute the impact of short-term anomalies like COVID recovery:

• Model Selection: Automatically chooses the best-fitting model (linear, quadratic, or exponential) by AIC, with BIC, R², and RMSE reported as fit diagnostics. Loss years are retained: the exponential model is excluded when history contains losses, rather than silently fitting only the profitable years.
• Recent-Trend Consistency: The growth rate is read off the fitted trend at its most recent point (not a backward average) and cross-checked against the latest filed actuals — a fit whose forward rate contradicts the data's own recent shape is flagged and routed to a more conservative rate. The long history prevents overfitting to recent noise.
• Outlier Detection: Identifies and handles unusual periods (e.g., COVID disruptions) to avoid distorting growth projections.
• Revenue vs. Earnings Consistency: If earnings grow much faster than revenue, we blend the rates to avoid unrealistic projections.

🎯 Risk Spread from Business Fundamentals:

The spread over terminal growth is built from accounting fundamentals (not stock-price beta), starting from a 3pp base and bounded 1.5–10pp:

• ROIC quality: Higher, sustained returns on invested capital tighten the spread (a true value-creator is lower-risk).
• ROIC vs. r: Comfortably earning above the cost of capital tightens the spread. Destroying value does not widen the rate — it is a floor question, handled by the conservative-book floor, never by inflating the discount rate.
• Earnings cyclicality (the dominant driver): High net-income variability and frequent year-over-year declines widen the spread — a stable utility lands near zero, a cyclical semiconductor name earns a multi-point premium.
• Forward-decline guard: Earnings in secular decline (recent half of history vs. the earlier half) widen the spread by up to 2pp — the forward risk that trailing stability misses.
• Leverage: High net debt relative to equity widens the spread by up to 1.5pp.

The resulting rate (g + spread, shifted by the calibration constant) is then floored at the bond-anchored market-neutral anchor (currently 6.24%): names whose fundamentals cannot honestly place them above the anchor price AT the anchor. Downside durability is measured separately as a floor-strength score — a selection and trust signal, never a discount-rate adjustment. A diagnostics layer flags any result where the rate, spread, or terminal-value share looks unreliable.

Why fundamentals over beta? Stock prices can be irrational; earnings and returns on capital reflect actual business performance. Academic research (Ball & Brown 1968, Beaver et al. 1970) shows earnings volatility predicts business risk well — and the forward risk that trailing stability misses is handled by the fundamentals-based secular-decline guard, not a CAPM floor. Market prices never enter the rate.

🏰 Competitive Advantage (Moat) Analysis:

We analyze 15-20 years of SEC data to identify sustainable competitive advantages:

• Scalability: Can the company grow without proportional increases in assets or working capital? (Asset turnover, working capital efficiency)
• Margin Persistence: How stable are profit margins over 15+ years? Consistent high margins suggest pricing power.
• Market Saturation Signals: Declining revenue growth trends may indicate market maturity.
• Asset-Light Model: Companies that generate high revenue with low asset requirements (e.g., software, services).

The moat score is reported as context but no longer moves the rate or terminal growth directly — durable pricing power measured from the filings (gross-margin stability + real revenue growth) is what earns a higher terminal g, and business risk is priced from earnings fundamentals.

🔧 Special Adjustments:

• CapEx Split (self-builders): When capital expenditures run ≥10% of revenue, we split CapEx into maintenance (capped at depreciation) and growth CapEx, and perpetuate owner earnings — relieving only the buildout the business funds from its own cash (externally financed buildout stays a real cash charge). This handles strategic investment phases (e.g., a cloud-infrastructure buildout) explicitly, with the evidence shown in the output, instead of silently rewriting CapEx.
• Financials Lane: Banks are valued on earnings — FCF is structurally meaningless for lenders, since loan capital shows up as negative cash flow. P&C insurers are valued on a residual-income model: normalized ROE on ex-AOCI tangible book, an accident-year combined-ratio window (10–12 years, catastrophe losses included), and an adverse-development gate that governs regardless of how cheap the print looks. Life insurers print an honest NOT-APPLICABLE rather than a forced number. All lanes use the same bond-anchored discount rate.
• Foreign Filers (IFRS, 20-F/40-F): Annual filers are valued on their audited fiscal year and labeled as such (no quarterly XBRL exists for them), and fundamentals stay in the filer's reporting currency with an explicit currency flag — no silent FX conversion.
• Stock Split Detection: Automatically detects and applies stock splits from 8-K filings to ensure accurate share counts.

🧭 Reading a Print: Trust, Range & the Hype Meter:

• Data-Quality Gate: Every extraction runs through a quality gate (stale vintages, sentinel zeros, missing CapEx, mixed reporting bases). A record that fails is refused — the page shows the reason instead of a number, and no valuation or comparison is produced for a gate-blocked name.
• Trust Tier (A–D): Every print — including recalculations — carries the engine's own confidence: A = trust direction and rough magnitude, B = direction only, C = artifact-prone, D = do not use. Tiers come from the engine's own self-flags, and each reason names the fix that would unlock a higher tier.
• Value Range: A low–central–high range from real engine re-runs at the corners: growth-estimator disagreement, rate uncertainty scaled by the rate's own reliability grade, and the base-policy alternatives. A range wider than 2.5× the low prints a WIDE_RANGE flag — look at the uncertainty and the floor before acting.
• Hype Meter: Compares the growth the market price implies (reverse-DCF of the same model) with the growth the filings support. Bands: UNDER_FLOOR (price below even the no-growth floor), PRICE_UNDER_ASKS, FULLY_PRICED (gap ≤3pp), PRICED_FOR_MORE (3–8pp), HYPE_ZONE (>8pp), EXTREME. An exit signal fires only on a valid instrument (trust A/B, converged solver, uncontaminated base). Price is interrogated here — never an input to the valuation.

📚 Academic Foundation:

Our methodology borrows from established academic finance research:

• Earnings-Based Risk: Ball & Brown (1968), Beaver et al. (1970), Dichev & Tang (2009)
• Earnings Quality: Sloan (1996), Dechow & Dichev (2002) — applied as an accruals check (cash flow vs. net income) in the durability reads
• Growth Model Selection: AIC/BIC criteria (Akaike 1974, Schwarz 1978)
• Retired: the Altman (1968) Z-score screen and the CAPM cost-of-equity floor (the Damodaran implied S&P 500 ERP, ~4.23% as of Jan 2026, survives in reference diagnostics only) — the deployed rate is bond-anchored from business fundamentals, with no beta or market-premium input