The Engine Behind CirclFi

13 Valuation Models.
One Unified Signal.

Each model attacks valuation from a fundamentally different angle — intrinsic, relative, scenario, option-based, and ensemble. Together, they produce a confidence-weighted consensus that no single model can achieve alone. All models run daily on 6,162+ US equities using SEC EDGAR financial data, FRED macroeconomic indicators, and GDELT news sentiment.

1. Bayesian DCF2. Earnings Power Value3. Markov DDM4. Dynamic NAV5. EROIC Spread6. ML-RIV7. First Chicago8. PWERM9. Regime Cross-Sectional10. Sentiment SOTP11. CUCE Ensemble12. FTNN Topology13. RCMH-DCF
MODEL 1 Intrinsic

Bayesian DCF

Core Formula EV = Σ FCFₜ/(1+WACC)ᵗ + TV/(1+WACC)ⁿ

Projects free cash flow over a 10-year horizon using 10,000 Monte Carlo simulations with a deterministic seed for reproducibility. Each simulation path draws from a Bayesian posterior distribution of growth rates (from Python MCMC training or macro-conditioned heuristics) and applies Merton (1976) jump-diffusion — modeling sudden FCF shocks with λ=0.10 intensity and -15% mean jump. Growth, WACC, and terminal rates are all stochastic. The 3-year weighted FCF base (50%/30%/20% recency weighting) smooths one-time events. WACC includes Hull-White stochastic rate drift (±30bps) for rate-sensitive sectors. Enterprise value converts to equity via a full institutional bridge: EV − lease-adjusted debt + cash − minority interest − preferred stock.

Key Innovation

10,000 Monte Carlo + Merton jump-diffusion

Data Inputs

OCF, CapEx, Beta, FRED rates, VIX, MCMC posteriors

Best For

Stable FCF companies

MODEL 2 Intrinsic

Earnings Power Value

Core Formula EPV = NOPAT / Kₑ

Implements Bruce Greenwald's complete three-part framework from Columbia Business School: (1) Reproduction Asset Value (RAV) — the cost to rebuild all tangible and intangible assets from scratch with sector-specific replacement multipliers (PP&E at 1.1×, R&D IP at 3-5× annual spend, brand equity from SGA × brand multiplier), (2) Earnings Power Value — cyclically normalized EBIT across all available fiscal years, after-tax, adjusted for sector-specific maintenance CapEx ratios (90% for capital-heavy, 30% for R&D-heavy, 60% default) and stripped of restructuring charges, goodwill impairment, and FX gains/losses, capitalized at the unlevered cost of equity (Kₑ), and (3) Franchise Value — the spread between EPV equity and RAV equity, classifying companies as Wide Moat, Narrow Moat, No Moat, or Value Destroyer. Adds a bounded growth premium (capped at 30%) when ROIC exceeds WACC, using McKinsey-style incremental value computation.

Key Innovation

Greenwald Franchise Value Trilogy + RAV

Data Inputs

Multi-year EBIT, D&A, CapEx, SGA, R&D, PP&E, Goodwill

Best For

Mature earners, Moat companies

MODEL 3 Intrinsic

Markov DDM

Core Formula V = Σ Dₜ × Pᵢⱼ / (1+Kₑ)ᵗ

Extends the classic Gordon Growth Model by modeling dividend growth as a Markov chain with discrete states representing growth regimes (high growth, stable, cut, freeze). Transition probabilities are estimated empirically from multi-year DPS history or loaded from Python-trained transition matrices. Critically, uses total shareholder yield (dividends + share buybacks) rather than dividends alone — expanding coverage to non-dividend payers like AAPL, GOOG, and META that return capital via repurchases. The buyback yield is computed as shares repurchased ÷ market cap. A 10-year forward simulation applies regime-specific growth rates with stochastic transitions, then discounts at the cost of equity derived from CAPM with VIX-adjusted equity risk premium.

Key Innovation

Regime-switching + total shareholder yield

Data Inputs

DPS history, shares repurchased, payout ratio, beta

Best For

Dividend payers, Buyback-heavy (AAPL/GOOG)

MODEL 4 Asset-Based

Dynamic NAV

Core Formula NAV = Σ(Assets × Recovery%) − Liabilities

Computes net asset value by marking each balance sheet item to fair value using sector-specific recovery rates calibrated for stress scenarios. PP&E and real estate assets use higher recovery rates (0.70-0.90) than intangibles (0.30-0.50) and goodwill (marked to zero under distress). Cash is taken at face value. Inventory recovery varies by sector (retail ~60%, manufacturing ~50%, perishable ~30%). Critically, deducts hidden liabilities that standard NAV misses: operating lease obligations (ASC 842), underfunded pension deficits (obligation − plan assets), off-balance-sheet VIE liabilities, litigation reserves, and asset retirement obligations. The final equity bridge strips minority interest and preferred stock. Especially powerful for REITs (real estate at appraised value), banks (loan book at marked value), and resource companies (reserves at commodity-adjusted value).

Key Innovation

Sector-specific recovery rates + hidden liabilities

Data Inputs

All balance sheet items, lease liabilities, pension, preferred

Best For

REITs, Banks, Insurance, Resource companies

MODEL 5 Intrinsic

EROIC Spread

Core Formula EV = IC + Σ(ROIC−WACC)×IC/(1+WACC)ᵗ

Implements the McKinsey & Company economic profit framework. Capitalizes R&D expenditure as an intangible asset (with industry-specific amortization: 5 years for pharma/tech, 2 years for utilities, 3.5 years default) and adds it to invested capital alongside capitalized software, lease-adjusted debt, pension deficits, and equity method investments. Goodwill is adjusted downward by 50% of excess when goodwill exceeds 30% of total assets (serial acquirer penalty). Computes NOPLAT from normalized EBIT (stripped of restructuring, impairment, and FX items) × (1 − effective tax rate from EDGAR). The EROIC spread (ROIC − WACC) is projected over a Competitive Advantage Period (7-15 years based on historical ROIC consistency), with a linear fade rate of 10% per year toward WACC. Invested capital grows each year by reinvestment = NOPLAT × reinvestment rate (g/ROIC per McKinsey). Classifies companies as Strong Value Creator, Value Creator, Neutral, or Value Destroyer.

Key Innovation

McKinsey Value framework + CAP fade + R&D capitalization

Data Inputs

NOPLAT, invested capital, R&D, leases, pension, goodwill

Best For

Moat companies, High-ROIC businesses

MODEL 6 Intrinsic

ML-RIV

Core Formula V = BV + Σ(ROE−Kₑ)×BVₜ/(1+Kₑ)ᵗ

Machine learning-enhanced Residual Income Valuation that decomposes ROE into five DuPont components: net profit margin, asset turnover, financial leverage, tax burden, and interest burden. Models the persistence of each factor independently using multi-year financial data — high-persistence factors (like recurring margins) are projected forward more aggressively than mean-reverting ones (like leverage spikes). Computes abnormal earnings as the spread between ROE and cost of equity (Kₑ), applied to growing book value. Includes clean surplus adjustments for OCI violations (FX translation, pension adjustments, AFS securities) that bypass the income statement. Especially effective for financial institutions where book value is a meaningful anchor and earnings quality varies significantly across cycles.

Key Innovation

5-factor DuPont ROE decomposition + persistence modeling

Data Inputs

Book value, ROE components, clean surplus, multi-year financials

Best For

Banks, Financials, Insurance

MODEL 7 Scenario

First Chicago

Core Formula V = P₁×V₁ + P₂×V₂ + P₃×V₃

Constructs three independent valuation scenarios — bull (expansion), base (steady state), and bear (recession/disruption) — each with its own revenue growth trajectory, margin assumptions, and terminal multiple. Probability weights are dynamically adjusted based on: (1) the company's current position in its business cycle (using trailing revenue growth momentum), (2) sector cyclicality characteristics (biotech gets higher bull/bear spread than utilities), and (3) macro regime indicators (VIX level, yield curve slope, credit spreads). Each scenario runs a full mini-DCF with scenario-specific WACC. The final value is the probability-weighted average across all three outcomes. Particularly valuable for high-uncertainty situations where a single-point DCF gives false precision — early-stage growth, biotech with binary drug trial outcomes, and cyclicals at inflection points.

Key Innovation

Cyclicality-adjusted probability weights + sector benchmarks

Data Inputs

Revenue growth, margins, FCF, analyst estimates, VIX

Best For

Growth stocks, Biotech, Cyclicals, Turnarounds

MODEL 8 Option-Based

PWERM

Core Formula Equity = Call(V, D, σ, T)

Probability-Weighted Expected Return Method using Merton's structural credit model, which treats equity as a European call option on the firm's total assets with the debt face value as the strike price. Runs 5,000 Monte Carlo scenarios to simulate asset value paths with stochastic volatility and jump-diffusion dynamics. In each scenario, equity value = max(0, Asset Value − Total Debt), capturing the asymmetric payoff structure of equity. The option-theoretic framework naturally handles distressed companies (where equity has significant optionality even when assets < debt) and M&A targets (where the probability of a premium bid can be embedded in scenario weights). Particularly powerful for highly leveraged firms where traditional DCF breaks down because small changes in asset value create large swings in equity value.

Key Innovation

Merton structural model + stochastic volatility

Data Inputs

Asset value, total debt, volatility, risk-free rate

Best For

Distressed companies, M&A targets, High-leverage

MODEL 9 Relative

Regime Cross-Sectional

Core Formula V = Peer_Multiple × Metric × Regime_Adj

Identifies the current macroeconomic regime using six indicators: VIX level, yield curve slope, high-yield credit spread, GDP growth rate, unemployment trend, and inflation expectations. Classifies into one of four regimes (expansion, late-cycle, contraction, recovery), each with historically calibrated multiple ranges. Within the identified regime, selects sector-appropriate valuation multiples (EV/EBITDA for industrials, P/B for banks, P/FFO for REITs) and applies them to the company's normalized metrics. The PEG ratio is adjusted for the regime's growth premium or discount — growth stocks get lower PEG premiums in contraction regimes where growth is scarce and valuable. Cross-references against sector peers using 6-dimensional similarity (size, profitability, growth, leverage, volatility, dividend yield) rather than simple industry classification.

Key Innovation

6-indicator macro regime + sector-adjusted multiples

Data Inputs

P/E, EV/EBITDA, P/B, peers, VIX, yield curve, spreads

Best For

All sectors with sufficient peers

MODEL 10 Hybrid

Sentiment SOTP

Core Formula V = Σ(Segmentᵢ × Multipleᵢ × Sentiment_Adj)

Sum-of-the-parts valuation using actual business segment data from EDGAR filings (revenue and operating income by segment), with each segment valued at its industry-appropriate multiple. Overlays a four-layer sentiment analysis from GDELT news data: (1) overall company tone score, (2) industry-level sentiment, (3) segment-specific news volume, and (4) sentiment momentum (improving vs deteriorating). Sentiment adjusts segment multiples by ±15% — positive sentiment expands multiples, negative sentiment compresses them. The sentiment signal is asymmetric: negative news impacts multiples more strongly than positive news (behavioral finance loss aversion). Particularly effective for conglomerates like Berkshire Hathaway, General Electric, and Alphabet, where the whole-company multiple misses segment-level value creation or destruction.

Key Innovation

EDGAR segment data + GDELT news sentiment

Data Inputs

EDGAR segments, GDELT tone scores, sector multiples

Best For

Conglomerates, Diversified, Multi-segment

MODEL 11 Ensemble

CUCE Ensemble

Core Formula V = Σ wᵢVᵢ, where wᵢ ∝ 1/(σᵢ² × Cᵢⱼ)

The meta-model that combines all other model outputs into a single consensus fair value. Uses Correlation-Unbiased Certainty-Equivalent weighting: each model's weight is proportional to the inverse of its variance (high-confidence models get more weight) but adjusted for cross-model correlation (models that agree with each other have their combined weight dampened to prevent double-counting of similar signals). Outlier models (those more than 2 standard deviations from the median) have their weights reduced automatically. The ensemble produces a final fair value that is demonstrably more stable and accurate than any individual model — reducing model-specific bias while preserving signal from high-conviction estimates. The confidence score reflects the dispersion of the underlying models: tight agreement → high confidence, wide dispersion → lower confidence.

Key Innovation

Correlation-adjusted inverse-variance weighting

Data Inputs

All 12 model outputs, confidences, cross-correlations

Best For

All sectors (meta-model)

MODEL 12 Relative

FTNN Topology

Core Formula V = Σ KNN_j × w_j, where w ∝ K(d_ij)

Financial Topology Neural Network that finds the most financially similar companies using a 6-dimensional Gaussian kernel similarity function. The six dimensions are: (1) log market cap (size), (2) EBITDA margin (profitability), (3) revenue growth rate (momentum), (4) debt-to-equity ratio (leverage), (5) beta (volatility), and (6) dividend yield (income). Rather than grouping by SIC code or GICS sector (which puts Apple in "Consumer Electronics" alongside Samsung), FTNN finds peers based on actual financial DNA. Companies with similar financial profiles get higher kernel weights, and their market-implied valuations (EV/EBITDA, P/E) are used to derive fair value. The Gaussian kernel naturally handles non-linear relationships and gives exponentially more weight to closer matches. Particularly effective for companies that don't fit neatly into traditional sectors.

Key Innovation

6D Gaussian kernel financial similarity

Data Inputs

Size, margins, growth, leverage, volatility, yield

Best For

Tech, Healthcare, Consumer, any sector

MODEL 13 Intrinsic

RCMH-DCF

Core Formula V = Σ Pᵣ × DCF_r (r ∈ {expansion, slowdown, recession, recovery})

Regime-Conditioned Macro-Hedged DCF that runs four complete, independent DCF valuations in parallel — one for each macroeconomic regime (expansion, slowdown, recession, recovery). Each regime uses different discount rates, growth assumptions, and terminal values calibrated from FRED macroeconomic data: the expansion DCF uses lower spreads and higher growth; the recession DCF uses stressed WACC and negative growth assumptions. Regime probabilities are estimated from current macro indicators (VIX level, yield curve slope, GDP growth, high-yield credit spread). The Hull-White stochastic rate model adds interest rate drift (±30bps annual volatility) within each regime. The final valuation is the probability-weighted average across all four regimes. Especially valuable for rate-sensitive sectors (utilities, REITs, banks) where the interest rate environment fundamentally changes the cost of capital and asset values.

Key Innovation

4-regime parallel DCF + FRED macro integration

Data Inputs

FCF, FRED macro series, VIX, yield curve, GDP, HY spreads

Best For

Rate-sensitive, Utilities, REITs, Cyclicals

Data Pipeline & Sources

CirclFi ingests data from four institutional-grade sources daily:

SEC EDGAR

700+ XBRL tags mapped across US GAAP, IFRS, Banking, Insurance, REIT, and Utility sector templates. Multi-year 10-K and 10-Q filings.

FRED (Federal Reserve)

Risk-free rates, VIX, yield curve slope, GDP growth, high-yield credit spreads, and inflation expectations. Drive WACC and regime classification.

Market Data

Real-time pricing, beta, market cap, dividend yield, short interest, and sector peer multiples for relative and hybrid models.

GDELT News Sentiment

Global news tone scores, volume, and sentiment momentum at the company and industry level for behavioral-adjusted multiples.

Want the Full Mathematical Details?

The CirclFi Methodology Whitepaper includes exact formulas, derivations, and implementation details for all 13 models. Included free with every membership.

Access Terminal + Free Whitepaper →