The MonkScore™ Methodology
How the MonkScore™ signal is built and tested
One score. Five pillars. Twenty-five years of evidence.
MonkScore™ distills 149 academically-grounded factors into a single 0–100 rating — built to be trusted, and tested to be doubted.
Luck dominates the short term. Something else shows up over decades.
Markets are noisy. In any given quarter a great company can fall 30% for reasons that have nothing to do with its fundamentals, and a weak one can rise 40% on a short squeeze. Over months, luck dominates — which is why most stock-picking is built on hope.
Across thousands of companies and over multi-year horizons, a different picture emerges. Certain combinations of fundamental characteristics — the things you can measure in the filings — have preceded sustained outperformance more often than not. Not every time, and not in every period, but consistently enough across decades and across markets to take seriously.
This is not new. It runs from Graham and Dodd through the modern factor literature. What is new is what technology now makes possible: scoring every public company on all of it, every quarter, in close to real time. MonkScore is our answer to one question — after seven decades of research on what makes stocks outperform, what does the evidence look like distilled into a single number per company?
A single number for a question with thousands of moving parts.
MonkScore is a fundamental equity-selection signal: one 0–100 score, assigned point-in-time to a global universe, built as the geometric mean of five separately-validated pillars. The higher the score, the stronger the historical edge — and the score is informative across its whole range, not just at the extremes.
The reason for compressing everything into one number is not laziness; it's bandwidth. No investor can hold the joint distribution of profitability, growth, capital discipline, distress risk, and valuation across tens of thousands of companies in view at once — and the academic apparatus that studies these traits has itself advanced by isolating them one at a time. A score that reads all five together, on every company, every quarter, is measuring something the market assembles only imperfectly.
Five questions every serious investor asks — answered systematically.
Growth — is the company genuinely expanding?
Not how fast it grew, but whether the growth is real, broad, and durable.
Growth is the trickiest pillar in factor research, and it's worth being precise about what MonkScore does with it. Decades of evidence have shown that past growth on its own is a poor predictor of future returns — the “glamour stocks” picked out by a single hot trailing number tend to disappoint. The Growth pillar is built around that finding, not against it. What actually predicts returns is the quality, persistence, and breadth of expansion: growth that shows up across multiple parts of the business, holds up over time, and is corroborated by forward expectations. A single strong trailing metric triggers nothing on its own. The pillar measures the shape of growth across a business, not its magnitude at any one point.
Profitability
Does the business model generate real economic returns?
The level of economic return — the spread of return on invested capital over its cost — evaluated across several layers of the income statement and cross-checked against cash generation, so reported earnings actually convert to cash. The mechanism is compounding: a business earning 20% on capital builds wealth far faster than one earning 5%.
Novy-Marx · Fama-French RMW · Asness-Frazzini-Pedersen
Quality
How well is the company managed?
Capital-allocation discipline: the trajectory of returns on capital, earnings quality through accruals, balance-sheet resilience, and reinvestment discipline. Together these describe a company that's profitable today and likely to stay that way.
Sloan · Piotroski
Market Conviction
Has the market priced in confidence — and do the fundamentals justify it?
The market's valuation premium, read directly across several independent measures that must agree. A sustained premium on a sound business is treated as the market's own vote of confidence — but it only lifts the score when the other four pillars corroborate it, so an expensive stock with weak fundamentals is pulled down, not rescued by its price.
Jegadeesh-Titman · Asness-Frazzini-Pedersen
Safety
Are there warning signs?
Unlike the other four, Safety flags weakness rather than rewarding strength — the absence of three established failure modes: valuation-distress dynamics, fundamental deceleration, and accruals-based earnings-quality deterioration. A high score means none of these patterns is showing. It does not mean the company is risk-free. Nothing is.
Campbell-Hilscher-Szilagyi · Sloan
Combined by geometric mean
Not an average. A geometric mean punishes weakness in any single pillar far harder than an arithmetic one would — so a company cannot buy a high MonkScore by acing one dimension while failing another. The top scores go to companies strong across all five at the same time.
The result is only as honest as the rules that produced it.
Point-in-time scoring
Every score uses only the financial data filed and knowable on that date. Ratios are ranked against a trailing 180-day window, so a late or restated filing can never leak the future into a past score.
Peer-relative ranking
Every ratio is ranked against the company's natural comparison set — industry within region first, widening outward only as a fallback. A Japanese regional bank is judged against its peers, not against a US software company.
The measurability gate
Where a company lacks the data to be scored honestly on a pillar, MonkScore declines to guess rather than impute a flattering value. Coverage is 98.9% at the most recent date.
Geometric-mean aggregation
Weakness in any one pillar pulls the whole score down. This conditions a rich valuation on sound fundamentals — an expensive stock with broken fundamentals is sunk by those pillars, not rescued by its price.
Built with discipline, not a blind sweep. MonkScore reached its current form through roughly a dozen named versions. Every candidate factor was sourced from the academic literature first and tested second — across decades, regions, and market regimes — and more than forty constructions were discarded on held-out evidence. The direction of travel was consistently toward fewer free parameters, not more.
The proof is the shape, not a single number.
Annualized return by MonkScore decile · full global universe · 2000–2026
D1 − D10 = +27.9pp
Equal-weighted decile sorts, rebalanced quarterly, across 436,693 company-quarters. The implementable, long-only top decile nets roughly 29% per year after costs against a 30.3% gross — the cost drag is under a point and a half.
Top-scored stocks didn't just go up — they beat the average stock in 90% of quarters, for 25 years.
Over any five-year hold, a top-decile name beat the universe median 69% of the time; a bottom-decile name only 32% — and the gap widens with horizon.
Stocks scoring 90+ doubled within five years 1.6× as often as the average stock — 1 in 2 versus 1 in 3.
A signal that strengthens with time: its rank information coefficient rises from one month to five years — the opposite of a fading anomaly, which decays toward zero.
It holds when the ground shifts. The spread is positive and significant across three very different market regimes — the dot-com unwind, the financial crisis, and the pandemic-and-rate-shock years — and it is wider in the worst market quarters than in calm ones. Re-run on a wholesale change of universe — global ex-North-America, a different 84 quarters, a quarter of it emerging markets — the staircase holds and, if anything, steepens to +33.3pp.
And it holds on data the model never saw. The architecture was selected and frozen on the dense 2015–2024 record, then applied unchanged to the fourteen years of earlier history ingested only afterward — a genuine out-of-sample holdout. There the spread is +24.8pp, about 82% of its in-sample magnitude, with the identical 90% hit rate. An overfit signature does not survive on data that was absent when the model was built.
“Isn't this just value? Or quality? Or momentum?”
MonkScore beats its strongest single component three to one — +27.9pp versus the +9.8pp the best classic factor recovers.
Residual alpha that survives spanning against the academic factor zoo, on two independent universes (t > 9). It is not a relabeling.
A pattern no single factor reproduces: an information coefficient that rises with horizon, and a crisis-defensive 2008 where the nearest factor lost ground.
MonkScore is not new physics. It is a disciplined combination of signals the literature already validates — joined by a geometric mean that punishes weakness in any pillar, and gated by a filter that refuses to score what it cannot measure. Its value is precisely that it is irreducible: not to any one external factor, and not to any one of its own five pillars.
The one limitation we own — and why it makes the result stronger, not weaker.
Survivorship
The backtest universe is assembled from companies trading today, with their history back-computed — so firms that delisted, failed, or were acquired before the present are absent. This biases absolute return levels upward: the disasters that never entered the sample could never drag it down.
It is the reason this page never headlines a single absolute return. Every figure here is a gap — top versus bottom, score versus the average stock — and on the spread, the bias runs in our disfavor. The failed companies that are missing would have clustered in the low-scoring deciles, so their absence narrows the gap we report. The true top-minus-bottom spread is, if anything, wider than +27.9pp, not narrower.
The bias we cannot remove with this data, we name, bound, and lean against. The survivorship-free validation that would settle it — a re-run on institutional-grade, point-in-time data that includes the delisted and the bankrupt — is the natural next step, and the test this signal most wants to face. A genuine signal predicts it will hold or strengthen there. An overfit one does not survive it.
Where this model could underperform — in plain sight.
Sustained value rotations are the headwind
In a prolonged value rally — the kind that dominated 2000–2007 — that anti-value tilt becomes a drag, and MonkScore's forward edge compresses. We don't hide that. In value-led regimes in-sample, the edge roughly halves; it has stayed positive and significant rather than inverting, and the long book still beat the universe most of the time — but a real, bounded lag against broad benchmarks in those regimes is something to size for, not a possibility to wave away.
The structure provides partial protection, not full. Because MonkScore is a composite, it cannot reach the top on Conviction alone — a name also has to clear Profitability, Quality, and Safety, the pillars resting on the deepest and longest-standing evidence. The spread survives the removal of any single pillar, which is what makes the composite structurally less exposed than a lone factor to the “factor winter” that can sideline a single premium for a decade. Partial is not full. A broad, prolonged repricing of fundamental quality would still hurt it. We publish this so every reader can decide how to weight MonkScore in their own process.
What MonkScore isn't designed to do.
Credible, not replicable.
What we disclose
- ✓The five pillars and what each one measures
- ✓The academic literature each pillar rests on
- ✓The geometric-mean combination and what it conditions
- ✓The measurability gate and the peer-ranking ladder
- ✓The point-in-time discipline
- ✓The full empirical results — stated with their limitations
What we withhold
- —The identities of the 149 individual ratios
- —The weights within and across the pillars
- —The specific construction of the Safety failure-mode detectors
Why the edge should persist.
Published predictors tend to fade once the market learns them. Three features place MonkScore on the durable side of that line. Where the edge lives: it concentrates in mid- and small-capitalization names — exactly the corners arbitrage capital cannot cheaply reach, which is where surviving mispricing is supposed to live. It has never been published: the known anomalies it draws on have already paid the post-publication decay tax; the particular integration that turns them into a monotone +27.9pp sort has not. And the complexity is a moat: to arbitrage it away, the market would have to price the joint interaction of five dimensions, across thousands of names, every quarter — the conjunction no linear combination of the underlying legs reproduces.
There's a fourth tell, from the other end of the distribution. The bottom decile has a stable fingerprint — low profitability, weak growth, cheapness without fundamental support: cheap because it deserves to be. That failure profile barely moves across eras or market directions, while the winning profile rotates with the style regime. A signal anchored in the durable features of failure is harder to arbitrage away than one riding a rotating set of winning traits — the same insight that underwrites the long tradition of avoidance-based investing.
None of this is a guarantee, and we don't present it as one. It is a structural argument that the inefficiency MonkScore harvests is the durable, hard-to-arbitrage kind rather than the crowded, quick-to-decay kind.
A signal that worked yesterday may not work tomorrow.
So we keep testing it. As new quarterly cohorts arrive, we revalidate the spread and the hit rate. We periodically re-run the regime analysis to check whether the edge is weakening in particular market environments. Any factor we propose adding to — or removing from — the composite has to clear a pre-registered validation protocol before it goes in, so a change is never a hunch dressed as a finding.
When we make changes, we document them. When we find something that contradicts an earlier claim, we publish the correction. This page is maintained as the research evolves — the same posture of showing our work, caveats and limitations included, that the rest of it tries to hold.
See the method, or put it to work.
For advisors, RIAs & allocators
Open a conversation about the full research — the complete results, the robustness battery, the adversarial self-test, and survivorship-free validation.
For serious investors
See MonkScore live on any stock, with the full five-pillar breakdown. 15 days of full access, no card required.
Start your free trialImportant disclosures
All statistics on this page are derived from historical backtesting of survivors-only data; delisted, acquired, and bankrupt companies are excluded, which inflates absolute return levels. For this reason MonkStreet reports gaps — top decile versus bottom, score versus the average stock — rather than absolute return levels quoted in isolation. The benchmark is the equal-weight investable universe (the “average stock”), not a published market index. Return figures are gross or net of costs as individually labeled. Statistical significance is reported using Newey-West standard errors. The statistical significance of a historical signal does not guarantee that the signal will persist in future market environments. Past performance is not indicative of future results.
MonkScore™ is a quantitative research tool. It is not investment advice, a recommendation to buy or sell any security, or a solicitation to engage in any investment strategy, and it does not account for any individual's financial situation, risk tolerance, tax position, or time horizon. MonkStreet is not a registered investment adviser. MonkScore is computed from fundamental financial data provided by third-party vendors; while we take reasonable care over accuracy, we cannot guarantee that all inputs are error-free, and errors in source data will propagate to scores. Investors should consult a licensed investment adviser before making investment decisions, particularly for significant allocations of capital. MonkStreet and its affiliates accept no liability for investment decisions made on the basis of MonkScore or any information presented here. For questions about methodology or to report an error, contact methodology@monk.st.