Estimates & Forecast Engine
Methodology & Framework

The Estimates Engine is built on a core principle: the best forward estimate combines quantitative financial extrapolation with the collective intelligence of Wall Street analysts. Neither approach alone is sufficient — quantitative models excel at consistency and breadth but miss qualitative catalysts; analyst consensus captures forward-looking catalysts but suffers from herding and recency bias.

Design Principles

• Transparency over black-box: Every computation step is documented and inspectable. No hidden layers.
• Breadth over depth: Coverage of 3,700+ tickers with consistent methodology, rather than hand-tuned models for a few names.
• Graceful degradation: Companies with fewer data years receive lower confidence scores but still get estimates — the model adapts rather than refusing to compute.
• Internal consistency: Revenue × Net Margin = Net Income. Net Income / Shares = EPS. If any variable changes, all downstream values are recomputed.
• Blend, don't choose: Rather than picking model vs analyst, the system weights both based on measurable coverage quality.

Revenue is the foundation of every downstream estimate. The model uses a Blended CAGR approach that weights recent growth more heavily while retaining longer-term structural trends as a stabilizer.

Base Growth Rate Calculation

The base revenue growth rate is a weighted average of three historical growth windows:

This weighting gives 50% to the most recent year's trajectory (capturing current momentum), 30% to the 3-year CAGR (medium-term structural growth), and 20% to the 5-year CAGR (long-term trend).

Growth Caps by Market Cap

To prevent unrealistic extrapolation, growth rates are capped based on company size. Larger companies face harder scaling constraints — a $2T company cannot sustain 40% revenue growth.

Growth Fade for Outer Years

For FY+3 and FY+4, the model assumes growth converges toward the industry median rate. This reflects the empirical observation that above-average growth rarely persists beyond 2–3 years:

The exponential decay factor of 0.3 means ~26% of excess growth fades per year. By FY+4, the growth rate is heavily anchored to the industry median.

Quarterly Revenue Seasonality

The model also tracks quarterly revenue share patterns (Q1/Q2/Q3/Q4 splits) based on the company's historical seasonality. This is used for quarterly estimate breakdowns and for identifying seasonal patterns in the data.

After revenue is projected, the model determines the path of operating margin and net margin over the forecast horizon. Margins are the bridge between revenue and income — and they are notoriously difficult to forecast because they depend on pricing power, cost structure, mix shift, and macro conditions.

Mean-Reversion Framework

The model uses a recency-weighted 5-year moving average as the target margin. More recent years get higher weight because they reflect current cost structure and pricing dynamics:

Adaptive Reversion Speed

The speed at which current margin reverts toward the target is not uniform — it depends on whether margins are expanding, contracting, or stable:

A 3-year trend continuation factor (beta = 0.2) is also blended in, so the model respects structural margin shifts while still pulling toward the long-run average.

Operating vs Net Margin

Both operating margin and net margin follow the same mean-reversion framework independently. Net margin captures below-the-line items (interest, taxes, extraordinary items) and is used directly for EPS derivation. Operating margin is tracked separately for the Operating Income Forecast section.

EPS is the single most important metric for equity valuation because it directly connects to P/E-based price targets. The model derives EPS from first principles rather than forecasting it directly:

Share Count Projection

The model tracks the historical annual rate of change in diluted shares outstanding. This captures:

• Buybacks: Companies like AAPL consistently reduce shares by 2–4% annually, which compounds EPS growth.
• Dilution: High-growth names like TSLA or early-stage companies may increase shares through SBC or secondary offerings.

The annual share change rate is clamped to ±5% per year to prevent runaway projections from one-time events (e.g., a massive secondary offering or an accelerated buyback program).

FCF Derivation

Free Cash Flow is derived using the historical FCF-to-Net-Income conversion ratio applied to the projected net margin. This implicitly captures the company's capital intensity and working capital dynamics without requiring explicit capex forecasting.

This is the most critical differentiation of the model. Rather than choosing between quantitative extrapolation and analyst consensus, the system dynamically blends both based on coverage quality.

Why Blend?

The quantitative model uses only financial history — it cannot see M&A integrations (AVGO/VMware), drug pipelines (LLY/GLP-1), pricing power shifts (V), or regulatory changes. Analyst consensus captures these forward-looking catalysts. Blending gives the best of both worlds.

Blending Weights

Key design choices:

• Floor of 15%: Even with minimal analyst coverage, some consensus signal is preserved.
• Ceiling of 65%: The model always retains at least 35% weight, preventing pure consensus pass-through.
• FY+2 decay: Analyst estimates for Year 2 are less reliable, so the model gets more weight.
• FY+3/FY+4: Pure model projections — analyst consensus rarely extends this far.

Post-Blend Consistency

After blending EPS and Revenue separately, the model back-computes Net Income and Net Margin to maintain internal consistency. This prevents the scenario where blended EPS implies a margin that contradicts the blended revenue.

Not all estimates are equally reliable. A P&G revenue forecast should carry more weight than a biotech startup's. The confidence score quantifies this reliability as a composite 0–100 metric.

Score Composition

Interpreting the Score

Point estimates create false precision. The model produces three scenarios at FY+2 to give users a probability-weighted range of outcomes rather than a single target.

Scenario Definitions

Price Target Derivation

Each scenario's EPS at FY+2 is multiplied by the company's median historical P/E ratioto produce an implied price target. Using the median (not average) makes the calculation more robust against outlier P/E periods.

The 12-month price forecast chart visualizes these three targets as diverging paths from the current price, with the Bear/Base/Bull endpoints labeled with exact dollar values. The historical price line provides context for where the stock has been relative to where the model projects it could go.

After producing blended estimates, the model runs a cross-check comparing its output against analyst consensus. This serves as a sanity check and divergence signal.

Model vs Consensus Alignment

The model classifies the relationship with consensus into three categories:

When the gap exceeds 10%, users should investigate the cause. Common reasons include: recent M&A that hasn't been reflected in historical financials, regulatory changes, or a major product launch that analysts are pricing in but the model's historical extrapolation does not capture.

Transparency demands acknowledging where the model falls short. These limitations are structural — they arise from the model's design choices and data inputs.

1. Financial Sector

Banks, insurance companies, and REITs have fundamentally different financial structures. Net Interest Income, Combined Ratios, and FFO are not captured by the standard Revenue → Margin → EPS pipeline. These companies currently use the industrial model, which produces usable but less precise estimates. Specialized models are planned for a future version.

2. M&A-Driven Growth

The model extrapolates from historical financials and cannot predict the revenue impact of a major acquisition (e.g., AVGO/VMware, AMZN/MGM). Analyst blending partially compensates, but the model will structurally lag during integration periods. The cross-check section flags these divergences.

3. Cyclical Industries

Companies in energy, mining, and materials have margins that swing dramatically with commodity prices. Mean-reversion may miss cycle turns. These companies receive lower confidence scores by design, signaling users to rely more on scenario ranges than point estimates.

4. Newly Public Companies

Tickers with fewer than 3 years of annual financial data are excluded from the estimates universe. This currently affects approximately 430 tickers. The requirement ensures minimum data quality for meaningful margin and growth projections.

5. Macro Regime Shifts

The model does not incorporate macroeconomic variables (interest rates, GDP growth, inflation) directly. A rising-rate environment that compresses multiples or a recession that slashes revenue growth will not be captured until it flows through actual financial results. The scenario analysis partially hedges this by providing Bear case estimates.

The estimates engine is versioned and updated on a defined cadence. Each version increment documents what changed and why.

Update Cadence

• Daily: Estimates are recomputed daily for the full universe (~80 seconds runtime). This captures new quarterly filings, analyst revisions, and price changes.
• Weekly: Earnings surprise data (beat/miss history) is refreshed every Saturday.
• Model revisions: Major methodology changes are versioned and documented. Users always see the current model version on the dashboard.

Coverage Statistics