DCF Intrinsic Value Model
Methodology & Framework

The DCF Intrinsic Value Model is built on a foundational belief: fair value is the present value of all future free cash flows a business will generate for its shareholders. This is the most theoretically grounded approach to equity valuation, rooted in decades of academic finance and practiced by institutional investors worldwide.

However, DCF is only as good as its inputs. Small changes in growth or discount rate assumptions can swing output by 50% or more. Our model addresses this through conservative guardrails, scenario analysis, and radical transparency — every assumption is visible, adjustable, and documented.

Core Design Principles

• DCF is for structural cash generators: The model is deliberately restricted to companies with demonstrated, recurring free cash flow. Cyclical, pre-revenue, and speculative businesses are excluded — not because they lack value, but because DCF provides false precision for them.
• Exclusion is a feature, not a bug: Showing "N/A" for companies where DCF is inappropriate is intellectually honest. Alternative valuation lenses (relative multiples, EV/EBITDA bands) are provided for excluded stocks.
• Honest framing over false confidence: The model avoids language like "conservative estimate" or "market is wrong." Instead, outputs are presented as "DCF at X% growth suggests Y" — letting users draw their own conclusions.
• Sanity bounds prevent mathematical nonsense: Uncapped DCF models routinely produce "9,000% upside" for a stock trading at $50. Hard caps on growth, upside, and IV/Price ratios keep outputs actionable.
• Reinvestor awareness: High-growth companies that reinvest heavily (suppressing current FCF) receive normalized cash flow treatment to reflect long-term earnings power.

Not every stock belongs in a DCF model. The eligibility framework ensures that only companies with the financial characteristics suited to discounted cash flow analysis receive a valuation. Every gate exists for a specific mathematical or financial reason.

Required Criteria

A stock must pass all of the following gates to receive a DCF valuation:

Why Exclude Cyclicals?

DCF assumes stable, predictable cash flows that grow at a modelable rate. Cyclical businesses fundamentally violate this assumption:

• Margins mean-revert to long-term averages — current high margins are peak-cycle, not sustainable
• Terminal value calculations become unreliable when base-year FCF is at a cyclical peak
• Minor assumption changes swing intrinsic value by 5–10× because the starting cash flow is unstable
• Commodity price dependency makes forward cash flow fundamentally unpredictable via extrapolation

Coverage Statistics

As of February 2026, the eligibility gates produce the following coverage:

The starting cash flow figure is the most critical input to any DCF model. It must accurately represent the company's sustainable cash generation capability — not a cyclically inflated or temporarily suppressed number.

Company-Type-Specific Cash Flow

Different business models generate cash through fundamentally different mechanisms. The model selects the most appropriate cash flow metric for each company type:

The Reinvestor Problem

Some of the world's most valuable companies — Amazon, Nvidia, Meta — reinvest so heavily that their reported FCF dramatically understates long-term earnings power. Using actual FCF for these companies produces absurdly low intrinsic values, which would be misleading.

Reinvestor Detection Criteria

A company is flagged as a "reinvestor" when all of the following are true:

Normalized FCF Formula

When a company is flagged as a reinvestor, the model substitutes actual FCF with a normalized figure:

The 8% target margin represents a reasonable long-term FCF margin for mature technology and consumer companies. This normalization is only applied when the normalized figure exceeds actual FCF — the model never artificially deflates cash flow.

The growth rate determines how fast projected cash flows increase over the 5-year explicit forecast period. Getting this right — or at least keeping it reasonable — is essential for meaningful output.

Multi-Metric Growth Selection

Rather than relying on a single growth metric, the model evaluates multiple growth dimensions and selects the highest, subject to caps. This "best of" approach captures the strongest signal of the company's expansion trajectory:

• Revenue Growth: 5-year CAGR → 3-year CAGR → TTM (cascade from longest available)
• EPS Growth: 5-year CAGR → 3-year CAGR → TTM
• Cash Flow Growth: FCF growth for industrials, Net Income growth for banks/insurance, FFO growth for REITs
• Floor: Minimum 8% growth rate applied universally — prevents value collapse for stable but slow-growing businesses

Size-Adjusted Growth Caps

Uncapped growth rates are the primary source of absurd DCF outputs. A $2 trillion company growing at 30% would double the size of most national economies within a decade. The model applies hard growth caps based on company size:

The Weighted Average Cost of Capital (WACC) determines the rate at which future cash flows are discounted to present value. A higher WACC reflects greater risk and produces lower intrinsic values.

CAPM-Based Estimation

WACC is derived from the Capital Asset Pricing Model (CAPM) with size-specific adjustments:

Beta (β) is sourced from publicly available market data and measures the stock's sensitivity to broad market movements. A beta of 1.0 means the stock moves in line with the market; above 1.0 implies higher volatility.

Step 1 — Blume Beta Mean-Reversion Adjustment

Raw beta estimates from price regression are noisy and systematically biased away from 1.0 due to measurement error. Before plugging beta into CAPM, the model applies the Blume (1971) mean-reversion adjustment — the same formula used by Bloomberg, Damodaran, and FactSet:

In practice this matters most for outlier stocks: a high-beta stock at β=2.0 becomes β=1.67 after adjustment — 33bp lower WACC than raw CAPM would produce. A defensive stock at β=0.50 becomes β=0.67 — preventing unrealistically low discount rates for utilities and consumer staples. At β=1.0 (market), there is no change.

Raw beta is clamped before Blume to reject data-corrupt outliers: Technology companies are capped at β=1.75 (betas above this almost always reflect short-term price dislocations, not structural risk); all other sectors are capped at β=2.25. Blume is then applied to the clamped value.

Step 2 — Size Premium

Smaller companies carry a higher equity risk premium due to liquidity constraints, less institutional coverage, and weaker access to capital markets. The model adds a size premium on top of CAPM:

Step 3 — WACC Clamp by Company Quality Tier

After CAPM + Blume + size premium, the resulting discount rate is clamped based on company type and a quality tier check. This prevents data-quality extremes (a biotech with one year of price history and raw β=2.8 getting a 19% WACC) while keeping rates in the range institutional analysts actually use. The tiers are calibrated to observed implied discount rates for each category, not just theoretically derived values:

Complete Walkthrough — Apple (AAPL, April 2026)

Walking each step for a real company shows how the components combine:

• Risk-free rate: 4.5% (10Y Treasury yield)
• Beta (raw): 1.20 → clamped at 1.75 Tech ceiling (no change) → Blume: (2/3 × 1.20) + (1/3 × 1.0) = 1.133
• Equity risk premium: 5.0% (standard for $100B+ liquid companies)
• Size premium: 0% (market cap ≫ $100B)
• Raw cost of equity: 4.5% + (1.133 × 5.0%) = 10.17%
• Platform Quality adjustment: −0.75% (qualifies: Tech sector, ~$3T MCap, ~27% FCF margin — see §11)
• Adjusted cost of equity: 9.42%
• After debt blending: WACC ≈ 8.3% (minimal net debt, near-equity structure)
• Clamp check (Platform Quality tier, 7.5–14%): 8.3% is within range → no adjustment

Without the Platform Quality adjustment, AAPL's WACC would be 9.2% — above the 8.5% floor but close enough that the entire difference flows through to intrinsic value. The 90bp gap produces a ~$36 difference in the base case per-share value (~$152 vs ~$188).

The core engine uses a two-stage DCF model: a 5-year explicit forecast period followed by a terminal value that captures all cash flows beyond year 5 in perpetuity.

Stage 1: Explicit Forecast Period (Years 1–5)

Free cash flow is projected forward for 5 years using the selected growth rate, then each year's projected FCF is discounted back to present value:

Where g is the capped growth rate from Section 4 and WACC is the discount rate from Section 5. Each year's cash flow grows at the same rate — the model does not apply growth fade during the explicit period (this is handled by the growth caps which already reflect sustainable rates).

Stage 2: Terminal Value (Gordon Growth Model)

Beyond year 5, the model assumes the company grows in perpetuity at a fixed terminal growth rate. This is computed using the Gordon Growth Model:

Terminal growth is not a single fixed number. It varies by company size and type, reflecting the economic reality that a $10B mid-cap has more structural room to outgrow GDP than a $2 trillion company that is already larger than most national economies:

Sector-level fine-tuning is applied on top of these base rates — Consumer Defensive industries receive a small downward adjustment (their volumes are inelastic, not growth-oriented); high-growth tech-adjacent verticals receive a small upward push. The final terminal growth rate is clamped hard at 1.5%–3.5%, so no company can be modeled growing faster than the long-run ceiling for US nominal GDP in perpetuity.

The terminal value is then discounted back to present value:

Enterprise Value → Equity Value → Intrinsic Value

The final intrinsic value per share follows the standard corporate finance waterfall:

Net debt is defined as total debt minus cash and cash equivalents. For companies with net cash positions (cash exceeds debt), net debt is negative, which increases equity value above enterprise value. Share counts are sourced from the most recent publicly filed quarterly report.

Point estimates create false precision. A single intrinsic value number implies a level of certainty that does not exist in financial modeling. To address this, the model produces three scenarios that bracket the most likely range of outcomes.

Scenarios are built by applying absolute percentage-point shifts to the base case inputs — not percentage-of-base haircuts. This matters because a 20% haircut on a 5% growth rate is only 1pp, whereas a 20% haircut on a 15% growth rate is 3pp. Absolute shifts ensure the spread between bear and bull is driven by the magnitude of uncertainty, not by the starting level of the base case.

Why Asymmetric Shifts?

Bear shifts are intentionally larger than bull shifts. This reflects how downside risk actually materializes in practice:

• Bear WACC +1.5pp vs Bull −1.0pp: Risk deterioration (credit stress, rising rates, execution failure) tends to be sudden and non-linear. Damodaran's implied ERP data shows equity risk premiums spike 200–400bp during market crises while compression rarely exceeds 100–150bp during recoveries.
• Bear margin −2pp vs Bull +1.5pp: Margin compression happens fast — cost inflation, competitive undercutting, and pricing power erosion are the most common FCF failure modes for high-margin businesses. Margin expansion is slower and requires demonstrated operating leverage, not just optimism.
• Terminal shift cap +0.3pp: Terminal growth of 3%+ already stretches credibility for most businesses. Allowing a large bull shift on terminal growth would make the model mechanically sensitive to a perpetuity assumption rather than to actual near-term cash flow — the opposite of what investors should be weighing.

Concrete Example — Microsoft (MSFT, Base WACC 8.0%)

With base growth of 12% (mega-cap cap applied) and base margin at TTM levels:

• Bear: 10% growth, 9.5% WACC, margin −2pp → per-share value ~$285
• Base: 12% growth, 8.0% WACC, base margin → per-share value ~$380
• Bull: 13.5% growth, 7.0% WACC, margin +1.5pp → per-share value ~$510

The $285–$510 range is wide enough to acknowledge genuine uncertainty but narrow enough to be actionable. A model that shows a $50–$5,000 range for a stock trading at $400 is not providing analysis — it is providing noise.

Even with growth caps and WACC bounds, certain edge-case combinations can produce mathematically correct but financially meaningless outputs. The model applies a final layer of sanity checks before publishing any intrinsic value.

Output Bounds

Valuation Status Classification

Based on the relationship between base case intrinsic value and market price:

Different business models require different valuation inputs. A bank's "free cash flow" is fundamentally different from a technology company's. The model adapts its inputs based on the company's classification:

For SaaS and software infrastructure companies, a standard FCFF model systematically overstates intrinsic value. The root cause is stock-based compensation (SBC) — a real economic cost to shareholders that GAAP accounting allows companies to add back into operating cash flow, creating a large structural gap between reported FCF and true distributable earnings.

The SBC Gap Problem in Detail

Consider ServiceNow (NOW) in FY2025: operating margin was roughly 14%, but reported FCF margin was ~28% — a 14 percentage-point gap driven almost entirely by ~$1.5B of SBC being added back in the cash flow statement. If a DCF model uses the 28% FCF margin as its base, it implicitly treats SBC as free — as if the company is generating 28 cents of distributable cash per dollar of revenue. In reality, those shares are diluting existing shareholders every year. The true owner earning power is closer to the 14% operating margin, not the 28% FCF margin.

Datadog (DDOG), Zscaler (ZS), and Snowflake (SNOW) exhibit the same pattern: FCF margins of 20–30% sitting 15–20pp above GAAP operating margins. Using raw FCF for these companies would produce intrinsic values 40–80% higher than the economically correct number.

Routing Criteria

A company is routed to the Owner Earnings model when all of the following are observed in its trailing-twelve-month financials:

Owner Earnings Calculation

Rather than using either FCF or net income, the model uses operating income (EBIT) as the economic base — which already deducts SBC as a compensation expense — then applies the standard DCF adjustments:

This is identical to FCFF except that EBIT (post-SBC) replaces NOPAT derived from reported cash flows. The effect is to charge SBC as a real cost rather than adding it back. D&A is added back because it is a non-cash charge that does not reduce the company's economic earning power; CapEx and working capital changes are deducted because they represent real capital consumption.

Standard CAPM systematically overstates the cost of equity for a narrow category of exceptional businesses: durable-moat, platform-scale companies where near-zero bankruptcy risk, deep ecosystem lock-in, and highly predictable recurring cash flows make the beta-implied risk premium economically inaccurate.

The Problem With Beta for Mega-Cap Platforms

AAPL, MSFT, and GOOGL each have a reported beta close to 1.0 — implying they are "average risk". But this conflates price correlation with the index with fundamental business risk. These stocks move with the market not because their businesses are risky, but because they arethe market — they constitute 15–25% of the S&P 500 by weight. Their actual business risk (probability of cash flow impairment, margin collapse, or default) is orders of magnitude lower than a typical beta-1.0 industrial company.

Damodaran's (2026) implied equity risk premium analysis shows AAPL and MSFT trade at implied discount rates of 6.5–7.5% — well below the 8.5% floor the CAPM model would assign without adjustment. The market, in aggregate, has been pricing these companies as if their WACC is 7–8%, not 8.5–10%. A model that ignores this produces bull cases below the current market price — which is methodologically incoherent.

Qualifying Criteria

Platform quality adjustment is applied when all three conditions are met simultaneously:

Qualifying Companies (April 2026)

The three-way criteria identify approximately 10 companies across the active DCF universe: Apple (AAPL), Microsoft (MSFT), Alphabet (GOOGL), Meta Platforms (META), Netflix (NFLX), Cisco (CSCO), and a small number of other large-cap platform businesses. Notably excluded:

• NVDA / AMD / AVGO (Semiconductors): Already excluded from DCF entirely by the cyclical industry gate — chip-cycle margins are not suitable for terminal value extrapolation. No overlap with PQ adjustment.
• AMZN (Consumer Cyclical / Specialty Retail): AWS's platform value is missed because Amazon is classified at the parent company level, not broken out. This is a known limitation documented in Section 13.
• TSLA (Consumer Cyclical): FCF margin below 18% threshold and classified outside Technology sector. Not eligible.

Three Changes Applied to Qualifying Companies

Before & After — Apple (AAPL, April 2026)

With AAPL trading at approximately $195 at the time of calibration:

Before the adjustment, the bull case was $217 — below the current market price of ~$195. That result is methodologically incoherent: if even the optimistic scenario sits below market price, it means the market is pricing in assumptions more aggressive than the model's bull case, which would imply the stock is universally overvalued under any reasonable WACC assumption. That conclusion conflicts with how the world's most sophisticated institutional buyers — who collectively own trillions of dollars of AAPL — are actually pricing the asset. The platform quality adjustment corrects the source: the discount rate assumptions, not the growth assumptions, were too conservative for this tier of business.

The three-scenario analysis (bear/base/bull) provides a useful range but obscures the continuous relationship between assumptions and value. The sensitivity matrix makes the model's behavior fully transparent: it shows intrinsic value across a full grid of WACC and terminal growth combinations, so users can immediately see which assumption drives value the most and whether the investment thesis holds under stress.

Grid Construction

The matrix is a 5×5 grid — 5 discount rate scenarios × 5 terminal growth scenarios — centered on the base case. Shifts are applied symmetrically:

Each of the 25 cells contains a full intrinsic value recalculation using that specific WACC and terminal growth combination, with base-case revenue growth and margin projections held constant. This isolates the discount rate / terminal growth interaction — the two parameters that drive the largest share of valuation spread in practice, and the two where honest uncertainty is highest.

How to Read the Matrix

Every cell is colour-coded relative to the current market price:

• Green (Upside): DCF value exceeds market price by more than 5% — this combination of WACC and terminal growth supports the investment case
• Blue (Fair): DCF value is within ±5% of market price — fairly valued under this assumption combination
• Red (Premium): Market price exceeds DCF value by more than 5% — the market is pricing in assumptions more optimistic than this cell

The center cell (0% shift on both axes) always equals the base case intrinsic value. Reading across a row reveals terminal growth sensitivity at a fixed WACC. Reading down a column reveals WACC sensitivity at a fixed terminal growth. Reading the diagonal from top-right to bottom-left (highest terminal growth + lowest WACC to lowest terminal growth + highest WACC) captures the full valuation range — from most optimistic to most pessimistic.

A thesis is robust when most of the upper half of the grid (lower WACC scenarios) is green — meaning the stock is cheap even at average cost-of-capital assumptions. A thesis is fragile when only the most optimistic corner (top-right: minimum WACC, maximum terminal growth) produces an upside case — meaning investors need everything to go right simultaneously on both discount rate and terminal value for the investment to work.

Technical Implementation

The matrix is computed in the Python DCF engine and stored as a JSONB blob in the dcf_snapshots.sensitivity_json table column. The frontend reads grid dimensions (rowCount and colCount) directly from the data array lengths — there is no hardcoded grid size in the frontend. The center-cell index is computed as Math.floor(rowCount / 2), ensuring the base case always stays centred regardless of future grid dimension changes. Upgrading from 5×5 to 7×7 requires only a backend change to the shift arrays; the UI adapts automatically.

Intellectual honesty demands acknowledging where the model falls short. These limitations are structural — they arise from the inherent nature of DCF modeling, not from implementation gaps.

1. Cyclical Businesses

Energy, basic materials, and marine shipping companies are excluded because DCF assumes stable, projectable cash flows. Cyclical businesses should be valued on mid-cycle earnings using EV/EBITDA bands and commodity cycle analysis. Our Relative Valuation model provides peer-based multiples for these sectors.

2. Narrative-Driven Stocks

Companies like Tesla, where valuation is heavily driven by narrative and future optionality (autonomous driving, energy, robotics), produce unreliable DCF outputs because minor assumption changes swing value by 5–10×. Showing "N/A" is the correct behavior — it signals that DCF is not the right tool, not that the stock has no value.

3. Pre-Revenue and Early-Stage Companies

Companies with less than $1 billion in revenue are excluded. These businesses are typically valued using revenue multiples, TAM analysis, or venture-style frameworks that are outside the scope of a DCF model.

4. M&A-Driven Transformations

The model extrapolates from historical financial data and cannot predict the cash flow impact of a major acquisition or divestiture. Post-M&A financials will flow through the model on the next update cycle, but there may be a lag of 1–2 quarters before the full impact is captured.

5. Macro Regime Shifts

The model does not directly incorporate macroeconomic variables (interest rates, GDP growth, inflation). A sharp rise in rates would increase WACC for all companies, but the model relies on the risk-free rate embedded in the CAPM formula rather than forward rate expectations. The scenario analysis partially hedges this limitation.

6. Currency Effects

For ADRs and companies reporting in non-USD currencies, financial figures are converted to USD using current exchange rates. This introduces currency risk that the model does not explicitly price — a weakening foreign currency could reduce USD-denominated fair value independent of business fundamentals.

The integrity of any quantitative model depends on the quality and timeliness of its inputs. All data used in the DCF model is derived from publicly available sources and regulatory filings.

Data Sources

Refresh Cadence

• Daily: Market prices, shares outstanding, and net debt are updated after market close. Intrinsic values are recomputed daily for the full eligible universe (~80 seconds runtime).
• Quarterly: Financial statement data is updated as SEC filings (10-Q, 10-K) become available — typically within 2–4 weeks of the reporting date.
• Continuous: The front-end caches intrinsic values for 4 hours (ISR) to balance data freshness with infrastructure efficiency.

The DCF model is versioned and updated on a defined cadence. Each version increment documents what changed and why, ensuring full audit trail transparency.

Planned Enhancements

• v3.1 (Planned): Amazon (AMZN) model-family override — current classification as Consumer Cyclical / Specialty Retail causes the FCFF model to miss AWS cloud value entirely. A segment-weighted approach (cloud DCF + retail EV/EBITDA) would produce materially better results.
• v3.2 (Planned): Stale-cache detection for share counts — the current sanity check compares float-shares against income-statement shares from the same cached file. If both are stale and consistently stale (pre-split), the ratio ≈ 1.0 and the check passes silently. A freshness check against the FMP lastUpdated field would catch these cases before they enter the DCF pipeline.
• v4.0 (Research): Specialized bank and insurance models — Excess Return Model for banks (using ROAE vs cost of equity spread), NAV-based approach for REITs to complement the current FFO proxy. Multi-stage DCF with explicit growth fade (high → stable → terminal) for companies in strategic transition.