Fair value as a distribution, not a number

Every stock in our universe is valued by running a Monte Carlo simulation across many plausible paths. Each path randomizes the inputs that real analysts argue about — revenue growth, operating margin, the target multiple the market should reward — within bounded distributions calibrated from the company's recent history and its peer cohort. The output is not a point estimate. It's a probability distribution over plausible fair values.

From that distribution we publish three numbers prominently. Bear case is the downside fair value — outcomes meaningfully below the central estimate. Base case is the median fair value — our best single estimate. Bull case is the upside fair value — outcomes meaningfully above the central estimate.

The width of the bear-to-bull band is itself information. A tight band signals high model confidence; a wide band signals that small changes to assumptions move the answer materially. A wide band is not a flaw — it's the model being honest about uncertainty. Two stocks can both have the same base-case fair value with very different bear-to-bull widths; treat them differently when sizing positions.

Comparing apples to apples

The most-debated input in any valuation is the target multiple — what should this stock trade at, in price-to-earnings or enterprise-value-to-EBITDA terms? We don't pull that number from thin air. We derive it from how the stock's peer cohort is actually priced today.

For every stock, we identify a cohort of comparable companies within the same sector and roughly similar in size, margin profile, and growth posture. We then look at the relationship between each peer's trading multiple and its Quality Score (more on Quality in § 04). That relationship — how the cohort prices quality — gives us a peer-implied multiple for the subject company, which we apply to its forward earnings power to derive a peer-anchored fair value.

Two refinements matter.

The subject is excluded from its own peer benchmark. Including the stock in the regression of its own peers would pull the trend toward the subject's current multiple — a name trading cheap would bias its own "predicted" multiple toward cheap, dampening the dislocation signal. We compare each stock against its peers only.

Fit quality and cohort size become confidence signals exposed to users. On every research page and every screener row, we publish two metrics alongside the fair-value distribution: how tightly the peer cohort actually agrees on the quality-vs-multiple relationship, and how many peers were in the cohort. When the peer fit is loose or the cohort is small, the model is telling you "I see this stock, but the comparables don't agree with each other — trust this signal less." When the math doesn't support a confident answer, we refuse to score the name rather than make one up. A blank where another tool would put a number is a feature, not a gap.

One framework doesn't fit all

EV/EBITDA is the right multiple for most operating businesses. It's not the right multiple for everything. Most retail tools blanket-apply one framework to all stocks and produce nonsense fair values on banks (where leverage is the business model and EBITDA is meaningless) and on REITs (where depreciation is largely non-economic and Funds From Operations matters more).

We make the framework match the asset class. The defaults that ship today:

The discipline is not "we use Monte Carlo." It's "we use the right method for the right asset class, then we Monte Carlo the inputs to that method." The Monte Carlo wrapper is constant; the underlying multiple framework is sector-aware. When the engine encounters a name where the framework doesn't apply — banks, REITs, and other specialty structures — it deliberately abstains rather than produce a number we don't believe.

This is the rare case where saying "no" is doing the user a favor. A retail tool that produces a fair value for every ticker, including banks and REITs and BDCs, is one that's quietly producing wrong numbers on a meaningful subset of its coverage. We'd rather have a smaller universe valued well than a larger one valued badly.

The Quality Score

Quality Score is a 0–100 composite that summarizes the fundamental health of a company. It blends measures of profitability, efficiency, financial strength, and earnings consistency into a single comparable number across our coverage universe. It serves two distinct jobs in our pipeline.

As the peer-benchmark input. The peer relationship described in § 02 plots peer multiple against peer Quality Score and reads off the cohort's pricing of quality. Higher-quality peers trade at higher multiples; the relationship is the basis of the peer-implied fair value.

As a gate inside the screener. Each screener strategy has Quality Score thresholds the candidate must clear, tuned to the strategy's intent. Quality + Value requires strong quality and a margin of safety. Quality Momentum requires acceptable quality alongside trend strength. The Value Trap warning fires specifically when low quality combines with deteriorating fundamentals — the "looks cheap, but it's distressed" failure mode that pure value screens consistently fall into.

Importantly, we publish the Quality Score itself on every research page and every screener row. We don't hide it behind a paid tier. A user looking at a screener signal sees the underlying Quality Score, the peer cohort, the fit quality — the math that produced the signal is visible. You don't have to take our word for the rank; you can see why the name appears where it does.

What feeds the model

Three categories of data flow into every valuation. Pricing, fundamentals, and universe membership — each refreshed on its own cadence, each anchored to a documented institutional source.

Pricing data. Daily closing prices, adjusted close, volume, and intraday quotes via Financial Modeling Prep's pricing endpoints. End-of-day data is canonical for the model; intraday quotes drive the daily refit on top of the weekly snapshot.

Fundamental data. Quarterly and annual financial statements (income, balance sheet, cash flow), key ratios, and analyst estimates via the same Financial Modeling Prep pipeline. We use trailing-twelve-month figures for current state and next-twelve-month consensus blends for forward inputs — the "NTM Blend" label that appears on the Valuation tab signals this is the basis being used.

Universe and constituent membership. The S&P 500, 400, and 600 constituent lists. We restrict our covered universe to these indices — roughly 1,500 names — because they're the universe where institutional-quality data exists and where the peer-cohort math has enough density to be reliable. Stocks outside the S&P universe are deliberately not covered; we'd rather have no opinion than a thinly-supported one.

The weekly snapshot is the canonical fair-value distribution per stock. The daily refit refreshes peer multiples and current-price inputs against the same distribution shape, so the screener can surface intraday-relevant dislocations without re-running the full simulation every day.

Every run, verifiable

Every valuation we publish carries a cryptographic hash — a fingerprint that uniquely identifies the inputs that produced the result. Two runs against the same underlying data produce the same hash. A change in input — a fresh earnings print, a peer cohort revision, an updated forward consensus — produces a different hash, and the change is recorded.

What this earns us in practice: when we publish a fair-value range, that statement is anchored to a specific, deterministic computation against a specific data snapshot. The number isn't a vibe; it's the reproducible output of a documented model applied to documented inputs. If we got it wrong, the wrongness will be visible in retrospect rather than hidden by post-hoc rationalization. That visibility is the discipline.

The hash is exposed on every research page in a small mono caption. It's there for the sophisticated user who wants to track changes across runs or compare a current view to a prior snapshot. Casual users will never notice it; that's fine. Its job is to keep us honest.

What this model deliberately doesn't tell you

Sophisticated users distrust tools that pretend to know everything. Here's what we deliberately don't capture, and where you should go instead.

Management quality and capital allocation. The Monte Carlo doesn't know whether the CEO is shrewd or asleep. A name with great fundamentals run badly will eventually disappoint the model; a name with mediocre fundamentals run brilliantly will eventually surprise it. Where to go instead: the Earnings Call AI tab on every ticker, which surfaces what management actually said across many quarters of conference calls. Themes, risks, and capital-allocation framing in management's own words.

Regime shifts and macro inflections. The peer relationship and forward consensus inputs are calibrated to the recent past. A genuine regime change — a war, a credit crisis, a sectoral disruption — will produce visibly stale signals before the inputs catch up. Where to go instead: the Global Pulse editorial coverage on /insights, which is human-curated and macro-aware. The Pulse tells you what's changing now; the model tells you what's priced now.

Idiosyncratic events. Pending litigation, FDA decisions, M&A overhang, accounting controversies — the model is blind to one-off events that don't yet show up in financial statements or peer multiples. Where to go instead: the News tab on every ticker aggregates current coverage from Reuters, Barron's, Seeking Alpha, and a dozen other sources. The model gives you the central tendency; the news gives you the tail risks.

The right entry price. Fair value is not a buy signal. A stock trading meaningfully below base-case fair value may continue trading there for months. The Valuation tab translates the distribution into an Entry Zone (where margin of safety is favorable), a Take-Profit range, and a Stop Loss — but timing the entry is the trader's job, not the model's. Use the distribution to size the opportunity; use your own discipline to time the trade.

Stocks outside our universe. The S&P 500/400/600 universe is roughly 1,500 names. The remaining US-listed tickers are out of scope. International listings, ADRs, and OTC stocks are not covered. We'd rather cover 1,500 names well than ten thousand names thinly.