Relative Valuation Model
Methodology & Framework

The Relative Valuation Model is built on a foundational premise: a company's intrinsic worth, while ultimately unknowable with precision, can be meaningfully approximated by studying how the market prices similar businesses. This is not a momentum overlay, a technical signal, or a sentiment gauge. It is a fundamental, bottom-up valuation framework that treats market multiples as the collective judgment of thousands of institutional investors, sell-side analysts, and market participants.

Why Relative Valuation?

Discounted cash flow (DCF) models are theoretically elegant but suffer from acute sensitivity to terminal growth rate and discount rate assumptions — small changes in these inputs produce vastly different outputs. Relative valuation sidesteps this fragility by asking a different question: given how the market prices comparable businesses, what should this company trade at?

The model combines this relative approach with absolute quality metrics, historical context, and stochastic simulation to produce a blended fair value estimate that is more robust than any single methodology alone.

Core Principles

1. Multi-method convergence. No single valuation multiple is authoritative. The model blends 4–10 methodologies, weighted by data quality and relevance, to reduce single-method bias.
2. Quality deserves a premium. A company with 48% ROIC should not trade at the same multiple as a 12% ROIC peer. The model applies a justified quality premium calibrated on historical back-tests.
3. Margins revert. Peak earnings create valuation traps. The model detects above-trend margins and applies cyclicality haircuts to prevent overvaluation at cycle peaks.
4. Confidence matters. A fair value estimate without a confidence score is a guess. Every output includes a 0–100 confidence assessment based on data depth, peer fit, and model agreement.
5. Transparency over opacity. Every weight, every adjustment, and every assumption is visible to the user. There are no black boxes.

The model operates on a proprietary data pipeline that ingests, normalizes, and validates financial data from multiple institutional-grade sources. Data integrity is the foundation of every valuation output — no model can compensate for contaminated inputs.

Primary Data Sources

Data Processing Pipeline

1. Ingestion: Raw XBRL filings parsed within 24 hours of SEC publication. Market data ingested at end-of-day.
2. Normalization: Financial line items mapped to a standardized 180+ metric schema. Handles US-GAAP presentation differences across filers (e.g., "Revenue" vs "Net Revenue" vs "Total Revenue").
3. Validation: Automated cross-statement reconciliation (e.g., Net Income on IS = Net Income on CF). Anomaly detection flags outlier values >3σ from historical.
4. Derived Metrics: 90+ computed ratios (ROIC, FCF yield, Piotroski F-Score, etc.) calculated from normalized fundamentals.
5. Screening Engine: Pre-computed screening tables updated daily with current multiples, growth rates, and quality scores for the full US equity universe.

The quality of a relative valuation is entirely dependent on the quality of the peer comparison set. An inappropriate peer universe renders all subsequent analysis meaningless. The model uses a hierarchical, four-level industry classification system to construct the most relevant peer group for each company.

Four-Level Industry Taxonomy

Peer Selection Rules

• Minimum viable peer set: Require ≥10 companies at L2 level for valid industry median. If fewer, fall back to broader classification.
• Market cap tiering: Mega-cap ($200B+), large-cap ($10B–$200B), mid-cap ($2B–$10B), small-cap ($300M–$2B), micro-cap (<$300M). Cross-tier comparisons apply haircuts.
• Business model filtering: Companies classified as banks, REITs, insurance, utilities, or biotech use specialized multiple sets (P/FFO for REITs, P/TBV for banks).
• Outlier exclusion: Peers with negative earnings or multiples >3× the industry median are excluded from median calculations but retained for scatter visualization.

Scatter Plot Visualization

The peer comparison scatter plot maps each company's P/E ratio (y-axis) against 5-year EPS CAGR (x-axis). Bubble size represents market capitalization. A regression line (quality curve) shows the expected relationship between growth and valuation, with R² measuring goodness-of-fit. Companies below the quality curve are potentially undervalued relative to their growth profile.

Not all valuation multiples are created equal. A trailing P/E ratio is meaningless for a pre-revenue biotech, and EV/EBITDA is inappropriate for a bank. The model dynamically selects and prioritizes multiples based on the target company's business model, data availability, and sector relevance.

Available Multiples

Dynamic Weight Assignment

Weights are assigned algorithmically based on three criteria:

1. Tier classification — Primary multiples receive 2–4× the weight of secondary multiples.
2. Data availability — Multiples with <10 peers contributing data receive reduced weight. Multiples with no historical data receive zero weight.
3. Business model relevance — Banks receive zero weight on EV/EBITDA; REITs receive maximum weight on P/FFO and P/AFFO; standard companies receive maximum weight on P/E and EV/EBITDA.

Weights are normalized to sum to 100%. The resulting blend is the raw fair value before quality and cyclicality adjustments.

Each contributing multiple produces an implied fair value — the price at which the target company would trade if it were valued at the industry median multiple. The blended fair value is the weighted average of all implied values.

Implied Value Calculation

Three-Source Fair Value

Each multiple can produce up to three implied values:

• Industry-implied: Using current industry median multiple.
• Historical-implied: Using the company's own 7-year average multiple.
• Forward-implied: Using consensus forward estimates and forward multiples (where available).

The model prioritizes industry-implied values (most weight) but incorporates historical and forward perspectives to reduce recency bias and consensus herding risk.

Weighted Blend Formula

The industry median multiple treats all companies as equal. But a company with vastly superior profitability, faster growth, and a fortress balance sheet deserves a premium. The Quality Adjustment Framework quantifies this justified premium (or discount) based on three fundamental pillars.

Three-Pillar Scoring

Composite Premium Calculation

One of the most dangerous valuation traps is buying a cyclical company at peak earnings. The P/E looks cheap because earnings are temporarily inflated — when margins normalize, both earnings and the multiple contract simultaneously (the "double whammy"). The model's cyclicality framework detects this risk.

Margin Z-Score

The primary detection mechanism is the Margin Z-Score — a standardized measure of how far the current net margin deviates from its 7-year historical average.

Cycle Position Labels

• Trough: Margins >1σ below average. Recovery potential. Multiple expansion likely.
• Expansion: Margins improving but below peak. Normal business conditions.
• Peak: Margins significantly above average. Mean-reversion risk is elevated.
• Contraction: Margins declining from peak. Multiple compression may follow.

Normalized Fair Value

When cyclicality is detected (Z-Score >+1.0σ), the model computes a normalized fair value — what the company would be worth if margins reverted to their 7-year average. This is displayed as the "downside scenario" in the Peak Risk Analysis section.

A single-point fair value estimate creates false precision. The Monte Carlo engine generates a probability distribution of fair values by running 10,000 simulations with randomized inputs, producing percentile-based confidence intervals that acknowledge the inherent uncertainty in valuation.

Simulation Parameters

Output Percentiles

Conviction Zone Mapping

The current stock price's position within the Monte Carlo distribution determines the conviction zone:

• Strong Buy: Price below P10 (deep value territory — extremely rare)
• Buy: Price between P10 and P25
• Mild: Price between P25 and P50
• Neutral: Price between P50 and P75
• Sell: Price above P75

Every fair value estimate should be accompanied by a confidence score. A model that says "fair value is $200, confidence 92" communicates something fundamentally different from "fair value is $200, confidence 38." The confidence score (0–100) measures the reliability of the model's inputs and internal consistency, not the probability that the stock will reach fair value.

Six-Component Breakdown

The Signal Strength Score (0.00–1.00) is the model's single most actionable output. It synthesizes valuation attractiveness, model reliability, risk factors, and Monte Carlo positioning into one number. This is not a buy/sell recommendation — it is a quantitative assessment of the current setup's favorability.

Four-Component Weighted Formula

Signal Labels

• STRONG SIGNAL (≥ 0.75): Multiple indicators converge favorably. Historically associated with positive 90-day outcomes.
• WATCH (0.55–0.74): Mixed signals. Some indicators favorable, others neutral or negative. Monitor for improvements.
• NEUTRAL (< 0.55): Insufficient evidence of favorable setup. Model does not identify attractive risk/reward.

The Action Zone framework translates the quantitative model output into concrete, price-level-based zones. These zones represent model-derived price thresholds where the risk/reward profile changes meaningfully — they are not buy/sell targets.

Zone Construction

• Primary Zone: Price at or below which the model calculates ≥15% margin of safety to blended fair value, AND the price remains above the P10 downside scenario. This is the zone where the model identifies the most favorable risk/reward.
• Cautious Zone: Price range where upside exists (+8% to +15%) but margin of safety is below institutional thresholds. Additional confirming factors (DCF agreement, improving margins) would strengthen the case.
• Current Price Zone: Displays live market price with its risk/reward assessment. If MOS <8%, the model considers the setup unfavorable at current levels.

Threshold Conditions

Beyond price levels, the model tracks four quantitative conditions that must all be satisfied for the model to register a "favorable" signal:

1. Price ≤ Primary Zone: Market price at or below the model's primary threshold.
2. Confidence ≥ 75: Sufficient data quality and model agreement.
3. Margin Z-Score ≤ +1.0σ: Margins not dangerously elevated.
4. P50 Fair Value > Price: Monte Carlo median confirms upside.

The industry median approach assumes all companies deserve the same multiple. The quality-adjusted approach applies a blanket premium. The regression approach takes this further: it builds a cross-sectional regression model that predicts a company's "expected" multiple based on its specific fundamental characteristics.

Regression Specification

If a company's actual multiple is significantly below its regression-predicted multiple, it may be undervalued relative to its fundamental characteristics. The regression provides signals like "Cheap" (trading below expected), "Fair" (near expected), or "Rich" (trading above expected).

The Peak Earnings Risk framework is the model's guard against the classic value trap: buying a stock that looks cheap on peak earnings, only to see both earnings and multiple contract. It answers the question: "What if current profitability is unsustainably high?"

Three Outputs

The sensitivity matrix explores how fair value changes across different combinations of two key drivers:EPS growth rate (rows) and P/E multiple (columns). This reveals the model's sensitivity to its most impactful assumptions.

Matrix Construction

• P/E scenarios: Typically 5 columns spanning from −4× to +4× around the current trailing P/E.
• EPS growth scenarios: 4 rows: Bear Case (50% of base), Conservative (75% of base), Base Case (current consensus), and Bull Case (133% of base).
• Fair value at each cell: Computed as Forward EPS (adjusted for growth scenario) × P/E scenario.

The center cell (base case growth × current P/E) represents the status quo. Cells highlighted in green indicate fair values above current price (upside); red cells indicate fair values below current price (downside).

A stock can be undervalued and still be a poor fit for a specific portfolio. The Portfolio Fit module analyzes how the stock would interact with common portfolio archetypes, providing hypothetical sizing and correlation context.

Outputs

• Suggested Role: Core Compounder, Growth Satellite, Deep Value Recovery, Income Anchor, or Speculative Position — based on market cap tier, growth profile, dividend characteristics, and quality score.
• Correlation Analysis: 90-day rolling correlation to SPY and QQQ (or sector-specific ETFs). High correlation (0.80+) suggests limited diversification benefit.
• Hypothetical Position Size: Model-suggested allocation range based on market cap, conviction level, and portfolio archetype. Includes maximum allocation threshold.
• Entry Urgency: 1–5 scale reflecting how time-sensitive the current setup is. Based on signal score trend, approaching catalysts, and proximity to support levels.

The Signal History table records every model signal generated for a given ticker, along with the subsequent forward returns. This creates an auditable track record that allows users to assess the model's historical accuracy for each specific stock.

What Gets Recorded

• Signal Date: Date the model produced the signal.
• Signal Label: STRONG SIGNAL, WATCH, or NEUTRAL.
• Signal Score: The composite score at the time of signal generation.
• Price at Signal: Stock price when the signal was generated.
• Forward Returns: Actual stock return at 30, 60, 90, 180, and 365 days after the signal.
• Outcome: Whether the signal was followed by a positive or negative return (evaluated at 90 days).

Track Record Metrics

• Win Rate (90d): Percentage of signals that produced positive 90-day forward returns.
• Average Return: Mean 90-day forward return across all signals.
• Total Signals: Number of signals generated for this ticker.