JEL classification: G11, G12, G14 Keywords: Brina Gap · market-implied growth · reverse DCF · expectations investing · ROIC · reinvestment rate · margin of safety · value trap · equity valuation · growth expectations · what is the market pricing in

Cite as: Brina, F. (2026). The Brina Gap: A Framework for Identifying Growth Mispricing in Equity Markets (Working Paper v3.0). Zyberno Research. Zenodo: 10.5281/zenodo.20651608 · SSRN: 6361659. In two parts: Part I — the paper; Part II — the complete empirical record. Supersedes v2.0 (Zenodo 10.5281/zenodo.20510518) and v1.0 (10.5281/zenodo.19052189); their directional numbers should not be cited.

About v3.0. This version reports, in full, one of the most extensive validation batteries ever applied to a practitioner valuation metric — including its own: construct, criterion, convergent and discriminant validity for the pricing arm; an artifact analysis and measurement protocol for the fundamental arm; survivorship-corrected directional tests on the complete point-in-time S&P 500; and matched-pair and regime stress tests run head-to-head against the classical Margin of Safety. Every result is reported, favorable or not. The headline that survives all of it: the market’s embedded growth expectation \(g^*\) is real and measurable; the Gap reads the expensive side of the market more accurately than the classical alternative; among all pure valuation metrics tested — book-to-market, earnings yield, EPV, the Margin of Safety — the Gap ranks first (§8.1), and it is the only metric in the entire audit whose central claim about the future can be verified at all — and it verifies. Like every metric tested here or in the companion audit, the Margin of Safety and ROIC included, it gives its best as one lens of two. A metric that survives this much adversarial testing, openly published, is worth more than one that was never tested. The complete empirical record — every directional table, robustness check, and failure mode — is included as Part II of this document.

Changelog (v3.0, June 2026; section references are to Part II). v3.0 establishes the paper’s thesis — the Brina Gap is a valid, distinct, interpretable measurement of growth mispricing, with return prediction as corroboration — and rebuilds the entire analysis on the complete point-in-time S&P 500: 771 firms across every constituent-year 2010–2024, including those subsequently acquired or delisted (100% constituent-year coverage, replacing v2.0’s survivor-leaning ~450-firm panel). Forward returns are dividend-reinvested total returns with filing-date entry and CRSP-style delisting handling (acquired firms carried to the deal price, then reinvested in SPY; bankruptcies to terminal). Construct validity (§6.6): \(g_f\) is the realized growth ceiling of stable-ROIC firms (realized 7.4%/yr vs predicted 7.1%), deviations explained by ROIC drift. Convergent validity (§6.8): \(g^*\) is robust across independent implied-growth recoveries (Spearman 0.73–1.00, including a residual-income model). Discriminant validity (§6.9): the Gap is near-orthogonal to the standard value/quality metrics and to the Margin of Safety (weakly rank-correlated at most; <6% shared rank variance with MoS). Survivorship-clean directional content (§6.1–6.2): including the delisted firms strengthens the short-side screen (55.9% → 58.9%, p = \(3\times10^{-8}\)) — firms the Gap flagged as overvalued that were later acquired cheaply or failed are correct calls a survivor panel discards; the Gap is an overvaluation detector (pooled 54.8%, Gap beats MoS +3.9pp p = 0.004, long side at chance). Plus firm-by-firm case studies (§6.10) and a deepened placement among the valuation canon, including residual income (§3.6). Incorporates the v2.1 split-adjustment correction (v2.0 forward returns used split-unadjusted prices). Supersedes v1.0 and v2.0; their directional numbers should not be cited.

Key takeaways

• The Brina Gap = (ROIC × Reinvestment Rate) − Market-Implied Growth compares what a firm can self-fund (\(g_f\)) with the growth the market is already paying for (\(g^*\), recovered from a reverse DCF). It makes the market’s embedded growth expectation explicit from public filings.
• Why growth units matter (the theory, §2.2): a multiple or an intrinsic-value verdict can never be checked against a future fact; “this price implies 12% growth” can. The Gap states valuation in the only falsifiable unit — which is what makes a validation battery possible at all — and efficiency logic predicts its result pattern: real measurement validity, no standalone return edge from transparent arithmetic (true of every peer, §8.1), and residual information exactly at the behavioral margin (over-extrapolated expectations).
• The pricing arm \(g^*\) is validated: the market-implied growth it recovers genuinely tracks firms’ subsequent realized growth (correlation ≈ 0.4–0.5; near-unbiased at the 10-year horizon), is robust across independent recovery methods (Spearman 0.73–1.00), and is distinct from standard value/quality metrics (it shares less than 6% of rank variance with the Margin of Safety). The market’s growth expectation is a real, measurable quantity — and the Gap surfaces it.
• The Gap is a detector of over-extrapolated growth expectations. The market systematically over-extrapolates growth — the calibration slope is below one at every horizon, worst at short ones, and the median firm priced for high growth under-delivers what’s implied (top-decile implied growth ≈ 26% vs ~22% realized). The Gap surfaces exactly this: a price embedding growth the business struggles to justify and the typical such firm fails to deliver. (Honest qualifier: this expectations-mispricing is not reliably exploitable by the Gap alone — the over-extrapolated cohort includes the rare moonshots that offset the median disappointments. The Gap detects an unrealistic expectation; it does not promise a trade.)
• Head-to-head with the Margin of Safety, the Gap is the better-built pricing instrument. It is fully mechanical and falsifiable from public filings, where the Margin of Safety depends on an analyst’s private intrinsic-value estimate; it is statistically near-orthogonal to it (shared rank variance < 6% — it reads a dimension the Margin of Safety does not); on the one directional test both can be scored on, the Gap is more accurate (≈ +4pp, \(p \approx .005\)); and on the test no pricing metric passes — picking the winner between two quality-matched firms — they fail together (47% vs 45%, both chance). Everything the Margin of Safety promises for the pricing question, the Gap delivers measurably — and among all pure valuation metrics tested (book-to-market, earnings yield, EPV, the Margin of Safety), the Gap ranks first (§8.1).
• Stress-tested in the most hostile decade for pricing metrics — and the measurement survived. 2010–2024 was the documented value drought: raw cheapness itself earned a negative spread, and every pricing measure went flat (§8.1). That \(g^*\)’s validity as an expectations gauge held throughout its worst-case regime is the most demanding test this paper runs — and the strongest evidence the quantity it measures is real.
• Honest limits. The fundamental arm \(g_f\) is a deliberately conservative steady-state anchor, not a firm-level growth forecast (it does not rank firms by realized growth — though as an anchor it is nearly unbiased: 7.8% vs 7.9% realized for stable-ROIC firms). And the Gap is not a standalone return signal — a property it shares with every single metric tested, the Margin of Safety included (companion audit; Part II §4); judging any of them that way is a category error.
• How to use it — the Brina Matrix. Stock selection has two axes: fundamentals (is this a good, durable business?) and pricing (what is the market expecting?). The Gap is the rigorous, validated pricing axis; fundamental quality (and judgment beyond any ratio) is the other. Crossing them — the Brina Matrix — sorts a stock into four quadrants (Underpriced Quality, Priced for Perfection, Value Trap, Speculative/Overpriced), with one sharp rule: “cheap” is a bargain only when quality is high; cheap + low-quality is a value trap (empirically the worst cell). Like every metric in the toolkit — ROIC and the Margin of Safety included — the Gap gives its best inside this framework: the toolkit needs exactly one validated instrument per axis, ROIC owns the quality slot (companion audit), and the Gap is the strongest candidate ever tested for the pricing slot.

Abstract

The Brina Gap measures the difference between the growth a company can self-finance and the growth its market price already assumes. Its fundamental arm is the sustainable-growth ceiling, \(g_f = \text{ROIC} \times \text{Reinvestment Rate}\); its pricing arm is the market-implied growth \(g^*\), recovered from a reverse discounted-cash-flow on enterprise value. Both come from public filings, so the Gap makes the market’s embedded growth expectation explicit — mechanically, reproducibly, and without an analyst’s private valuation. Stated in growth units, it is the only kind of valuation claim that can be checked against what firms later deliver — and it checks out.

On the complete point-in-time S&P 500 (2010–2024, survivorship-free), the market-implied growth \(g^*\) genuinely tracks the growth firms subsequently realize (correlation ≈ 0.4–0.5, nearly unbiased at the 10-year horizon) and is robust across independent recovery methods (Spearman 0.73–1.00). The Brina Gap is a detector of over-extrapolated growth expectations: the market systematically over-extrapolates growth — the calibration slope sits below one at every horizon — and the median firm priced for high growth under-delivers what its price implies (≈ 26% priced in vs ≈ 22% realized). Head-to-head against the classical Margin of Safety, the Gap is mechanical and falsifiable where the Margin of Safety is not, nearly orthogonal to it in cross-section, and more accurate on the one directional test both can take (+4pp, \(p \approx .005\); the Gap’s expensive-side calls were correct 58.9% of the time on the full universe). Among all pure valuation metrics tested — book-to-market, earnings yield, EPV, the Margin of Safety — the Brina Gap ranks first, and it is the only one whose central claim can be verified at all. Its practical form is the Brina Matrix, which crosses the Gap’s pricing read with fundamental quality and yields one rule that survives every specification tested: “cheap” is only a bargain when quality is high — cheap plus low quality is the value trap, the worst-performing cell in the sample. We are equally explicit about the limits: the fundamental arm is a conservative anchor, not a forecast; and the Gap — like every metric in the companion audit, the Margin of Safety included — is not a standalone return signal, which is exactly what efficiency logic predicts for any transparent, filings-based measure, in the most pricing-hostile decade on record. Data and code are published in full.

Part I — The Paper

1. Introduction — the two axes of stock selection

A stock decision has two separable components. The first is fundamental: is this a good, durable business — does it earn high returns on capital, and can it keep doing so? The second is pricing: what has the market already paid for — what future is embedded in today’s price? A wonderful business at a price that demands a miraculous future can be a poor investment; a mediocre business priced for disaster can be a fine one. Neither question answers the other.

The fundamental question is well served by existing tools (return on invested capital chief among them). The pricing question is harder to make explicit. Graham’s Margin of Safety addressed it but depends on the analyst’s own estimate of intrinsic value — it is not falsifiable from filings. Mauboussin and Rappaport’s expectations investing reframed it sharply — read price as an embedded forecast, then ask whether that forecast is reasonable — but left it as a discipline rather than a mechanical, reproducible measure. In academic terms, recovering \(g^*\) is the mirror image of the implied-cost-of-capital programme (Claus & Thomas, 2001; Gebhardt, Lee & Swaminathan, 2001; Easton, 2004): that literature fixes growth assumptions and solves the present-value identity for the discount rate; we fix the discount rate and solve for the growth the price implies — the quantity an investor actually wants made explicit.

The Brina Gap (introduced in the v2.0 release of this research program — Brina, 2026 — and maintained at Zyberno.com) operationalizes the pricing question. It recovers the growth rate the market’s price implies (\(g^*\)) by reverse-engineering a discounted-cash-flow on enterprise value, and compares it to a transparent, filings-based estimate of the growth the firm can self-finance (\(g_f\)). The Gap, \(g_f - g^*\), is a single signed number — positive when the firm could in principle out-grow what’s priced in, negative when the price demands more than its current economics can fund. It is mechanical, falsifiable, and reproducible. This paper asks whether the quantity it recovers is real, and shows how to use it.

The conclusion, documented across §4–§8, is a claim about the standard toolkit itself. Stock selection needs one validated instrument per axis. The quality slot has a clear owner — ROIC, the strongest sorter in our companion audit. The pricing slot has long been held by book-to-market and the Margin of Safety on seniority rather than evidence: on the identical sample, the Gap return-sorts above every pure valuation metric tested — book-to-market, earnings yield, EPV, the Margin of Safety — and it is the only one of them whose central claim about the future can be scored at all. By the same tests the canon is judged with, the pricing slot belongs to the Gap.

1.1 Contribution and honest positioning

1. A falsifiable valuation language. Valuation stated in growth units is the only kind that can be scored against realized outcomes — no multiple and no intrinsic-value estimate has this property (§2.2). This is what makes the rest of the paper possible.
2. A validated pricing measure. We show the market-implied growth \(g^*\) inside the Gap is not a DCF artifact: it forecasts realized growth, is robust across recovery methods, and is distinct from standard metrics (§4). This is the paper’s empirical spine.
3. The Gap as a detector of over-extrapolated growth expectations (§4.4): the market systematically over-extrapolates growth, and the Gap flags the prices that embed it — a real, documented mispricing of expectations (with the honest qualifier that it is not, alone, a tradeable edge).
4. The Brina Matrix (§5) — a quality × pricing diagnostic framework with the Gap as its validated pricing axis: four interpretable quadrants, a quantified value-trap rule, and firm-by-firm case studies (§6). This is where the Gap’s purpose is realized — as one axis of a mispricing diagnosis, the same way ROIC and every serious metric give their best as one axis of a two-axis judgment.

We are deliberately honest about the boundaries (§4.5): the fundamental arm \(g_f\) is a conservative anchor, not a forecast, and the Gap is not a standalone return signal. A companion audit shows no single metric is — so the Gap should be judged as one lens of two, which is exactly how this paper frames it.

2. The construct

Fundamental arm. \(g_f = \text{ROIC} \times \text{RR}\), where ROIC = NOPAT / Invested Capital and RR = net reinvestment / NOPAT. This is the textbook sustainable-growth identity: the growth a firm can finance by reinvesting its own profits at its own return on capital. It is a ceiling under steady-state assumptions, not a forecast (a firm can exceed it by raising ROIC, or fall short if ROIC fades).

Pricing arm. \(g^*\) is the constant growth rate that, plugged into a two-stage reverse DCF on enterprise value (NOPAT base, \(r = 10\%\), terminal \(g_T = 3\%\), explicit horizon \(N = 10\)), sets the present value equal to today’s EV. It is the growth the market’s price implies.

The Gap = \(g_f - g^*\). Positive: the firm’s economics could fund more growth than the price demands. Negative: the price demands growth the current economics cannot fund (which must come from rising ROIC, or the multiple compresses).

The choice \(N = 10\) is not arbitrary: §4.3 shows it is where \(g^*\) is best calibrated to realized growth, and §4.4 shows the cross-sectional reading is invariant to the DCF parameters in any case.

2.1 Measurement protocol — keeping the fundamental arm honest

The raw reinvestment rate RR = ΔIC/NOPAT is distorted by one-off capital events, and the distortions land exactly where a user would otherwise read the loudest Gaps. Large buybacks push RR negative, producing absurd negative \(g_f\) (a major defense contractor reads \(g_f = -62\%\) — a spurious “overvalued”); large acquisitions push RR far above one, producing inflated \(g_f\) (AT&T’s FY2018 RR was 210% on the Time Warner close — a spuriously loud “cheap”). In our high-ROIC subsample, roughly a third of firm-years have |RR| > 100%. We therefore recommend — and define as part of the construct — a four-point protocol:

1. Bound RR to [0, 1], the domain in which the sustainable-growth identity is meaningful (Higgins, 1977): a firm cannot organically reinvest more than it earns, and share repurchases are a distribution, not negative growth.
2. Smooth \(g_f\) over three years. A single-year \(g_f\) is mostly noise (year-to-year reliability ≈ 0.15); the three-year version is a stable anchor (≈ 0.78).
3. Flag low-reliability firm-years — any year whose raw RR falls outside [0, 1] (M&A or heavy-buyback years) should carry an artifact flag rather than a confident point estimate.
4. Report \(g^*\) with its cross-method band (§4.2) — e.g. “the market is pricing ≈ 8–11% growth” — rather than a false-precision point.

The statistics in this paper are computed on the raw construct (the protocol was derived from the artifact analysis); none of the validity results are affected, since they concern \(g^*\) — which the protocol does not touch — and no conclusion relies on extreme \(g_f\) values. The AT&T case study (§6) shows the protocol working as intended: bounding RR deflates the Gap from a too-loud +18 to ≈ +10 while preserving the correct qualitative reading.

2.2 Why growth units — the theory of the Gap

Three theoretical properties make the Gap the right construction for the pricing question, independent of any backtest.

First: a valuation claim stated in growth units is the only kind that can be checked. A multiple says “this stock trades at 28× earnings” — there is no future fact that settles whether 28 was right. An intrinsic-value verdict says “20% undervalued” — intrinsic value is never observed, so the claim can never be scored. A growth-units claim — “this price implies ≈12% annual growth for a decade” — is checkable: wait, and measure what the firm delivered. Translating valuation into growth units is what makes the entire validation battery of §4 possible; it cannot be run on a multiple, and it cannot be run on the Margin of Safety. Falsifiability is not a stylistic preference — it is the property that separates a measurement from an opinion.

Second: both arms are the canonical quantities of their respective literatures. The pricing arm operationalizes the Miller–Modigliani (1961) decomposition — value = steady-state value + present value of growth opportunities: \(g^*\) is the size of the growth-opportunities claim the price is making, which is exactly the quantity expectations investing (Mauboussin & Rappaport) tells the analyst to read, and which the residual-income canon (Ohlson, 1995; Penman, 2010) expresses in accounting form — §4.2 confirms \(g^*\) agrees with a residual-income decoding (\(\rho = 0.73\)). The fundamental arm is Higgins’ (1977) sustainable-growth identity — the only growth ceiling computable from filings alone. The Gap is therefore not an arbitrary composite: it is the difference between the two most defensible growth quantities in the literature — the growth the price asserts, and the growth the books can fund.

Third: theory predicts exactly the empirical pattern we find — including the return-null. If \(g^*\) is a competent forecast (§4.1 verifies it: ≈ 0.50 against realized growth), then prices already impound what public filings contain. The fundamental arm \(g_f\) is public-filings arithmetic — so for the Gap to predict returns standalone, the market would have to systematically ignore arithmetic it demonstrably prices everywhere else. Efficiency logic therefore predicts: strong measurement validity (the price contains a real growth forecast — found, §4.1–4.3), no standalone return edge from the transparent difference (found, §4.5 — for the Gap and for every transparent peer, §8.1), and residual predictability only at the behavioral margin that the expectations literature identifies — over-extrapolated growth expectations (Bordalo et al., 2019) — which is precisely where the Gap’s directional asymmetry shows up (58.9% accuracy on the expensive side, §4.4). Read this way, the result pattern of this paper is not the disappointing half of a signal paper; it is the confirmation pattern of a correctly built measurement. A transparent, filings-mechanical metric that did predict returns on its own would be a red flag — evidence of data snooping or an unpriced risk loading — not a virtue.

3. Data

The complete point-in-time S&P 500, 2010–2024 — 771 distinct firms across every constituent-year, including those later acquired or delisted (survivorship-clean). Fundamentals from SEC filings (XBRL); prices dividend-reinvested (total return); entry at fiscal year-end + 90 days (a filing-date proxy, no look-ahead); delisting handled CRSP-style. Realized growth and forward returns are measured over five-year windows. Financials are excluded where invested-capital metrics are not meaningful.

*7,576 of 7,578 constituent-years have a data row; the two missing belong to one bank, excluded by construction.

4. Is the pricing measure real? (Validity of \(g^*\))

4.1 Criterion validity — \(g^*\) forecasts realized growth

If \(g^*\) is a genuine measure of the market’s growth expectation, it should bear on the growth firms subsequently deliver. It does. Across the universe, market-implied growth \(g^*\) correlates with realized five-year NOPAT growth at Pearson ≈ 0.50 / Spearman ≈ 0.38, and at the 10-year horizon its average level is nearly unbiased (mean bias within ±1pp). The calibration slope sits below one — high implied growth is directionally right but over-stated, the over-extrapolation pattern §4.4 documents. The market’s embedded growth forecast is informative and, on average, well-centered — a real quantity, not a modeling artifact. (Notably, \(g^*\) forecasts realized growth far better than the fundamental arm \(g_f\) does; see §4.5.)

4.2 Convergent validity — robust across recovery methods

Because \(g^*\) comes from one particular reverse DCF, we recover implied growth three further ways and rank-correlate them with \(g^*\): a Gordon inversion (Spearman 1.00, mechanical), an owner-earnings/FCF reverse DCF (0.76), and a residual-income / EVA model drawing on different inputs (0.73). Genuinely different decodings of “the growth the price implies” agree on the cross-sectional ordering — \(g^*\) is model-robust, not an artifact of one convention.

4.3 The internal horizon \(N\) is calibration-justified

Sweeping the explicit-forecast horizon and comparing \(g^*(N)\) to realized \(N\)-year growth, calibration is best at \(N \approx 7\)–\(10\) (near-unbiased, peak correlation, no clamping); short horizons (\(N=3\)) over-state implied growth and clamp. The conventional \(N=10\) is the right choice, not a free parameter.

4.4 The Gap detects over-extrapolated growth expectations

This is the validated core of the Gap’s “mispricing detector” purpose — stated precisely. Two facts establish it. First, the calibration slope of realized growth on \(g^*\) is below one at every horizon, worst at short ones: the market systematically over-extrapolates growth (least reliable over ~3 years, most over ~7–10). Second, the over-extrapolation is concentrated where the Gap flags it: the median firm in the top decile of implied growth under-delivers — about 26% growth priced in versus ~22% realized — and, directionally, firms the Gap reads as expensive (negative Gap) subsequently miss their implied growth while cheap (positive Gap) firms beat it. The Gap therefore detects a real, documented mispricing of expectations: a price that embeds growth the business is unlikely to deliver. This is the classic over-extrapolation phenomenon of the expectations literature (Lakonishok, Shleifer & Vishny, 1994; La Porta, 1996; Bordalo, Gennaioli, La Porta & Shleifer, 2019), recovered here mechanically from prices and filings rather than from analyst forecasts.

The crucial qualifier (which distinguishes our claim from a return-signal claim we explicitly do not make): this expectations-mispricing is not reliably exploitable by the Gap alone — nor by any pricing metric we tested. Over-extrapolated names include the rare moonshots that pull the mean outcome back up even as the median disappoints, and on the cheap side value traps dominate (§5). So the Gap reliably detects an unrealistic expectation; it does not, by itself, tell you the trade will pay. Detecting the expectation error is the validated contribution; profiting from it requires the fundamental judgment of §5.

Directional corroboration — and how it squares with §8.1. The full point-in-time universe shows exactly the asymmetry this account predicts. Of the Gap’s directional calls with completed five-year windows (n = 1,504), negative-Gap (“expensive”) calls were correct on subsequent index underperformance 58.9% of the time (\(p \approx 3\times10^{-8}\)), versus ≈ 48% for positive-Gap calls — the Gap’s information is concentrated on the over-extrapolated side; and on that directional test it outperforms the Margin of Safety by ≈ 4pp (\(p \approx .005\)). Equal-weight base rates account for part of the level (most individual stocks lag the cap-weighted index), which is why we do not present hit rates as an edge. And §8.1 shows the complement: among quality-matched firm pairs, neither the Gap nor the Margin of Safety picks the winner. The two results answer different questions and are consistent: the Gap is the better-built instrument for reading the expensive side — where expectations are over-extrapolated — while no pricing instrument, alone, picks winners among comparable businesses. (Full directional-test tables, robustness, and inference: Part II, §4–§6 and §8.)

Separately, because \(g^*\) is a monotone transform of the EV/NOPAT multiple, the cross-sectional ranking the Gap produces is invariant to the DCF parameters (\(r\), \(g_T\), \(N\)) — no parameter choice could manufacture or destroy it.

4.5 Discriminant validity — and the honest limits

Distinct. The Gap is close to rank-orthogonal to the standard toolkit: its only non-trivial association is a moderate tilt to the operating-earnings yield (\(\rho \approx 0.5\), expected since \(g^*\) is built from the EV/NOPAT multiple); with ROIC (\(\rho \approx 0.2\)), book-to-market (\(\approx 0.15\)), and the Margin of Safety (\(\approx 0.24\)) it is weakly correlated at most. It is not a quality factor, not a conventional value ratio, and not a restatement of price-versus-DCF — it measures a distinct dimension.

Two limits, stated plainly: - The fundamental arm \(g_f\) is a conservative anchor, not a forecast. Across the universe \(g_f\) has essentially zero cross-sectional correlation with realized growth (≈ 0.02): it calibrates to average realized growth for stable-ROIC firms (7.8% anchored vs 7.9% realized — nearly unbiased as a benchmark), but it does not rank which firms will grow. The market’s \(g^*\) is the better forecaster of the two. So the Gap should be read as “the market’s (good) expectation versus a transparent, deliberately-conservative fundamental benchmark” — not “the firm’s true growth versus the market’s mistaken one.” - The Gap is not a standalone return signal — and neither is anything else. Used alone, its rank correlation with forward returns is small (+0.035 — which still places it above most celebrated screens; see the league table in §8.1). This is not a failure peculiar to the Gap: the companion audit shows no single screen — ROIC, value, quality, momentum — earns risk-adjusted alpha alone, the Margin of Safety fails every directional and matched-pair test we ran on it head-to-head (§4.6, §8.1, Part II §4), and §2.2 explains why theory predicts this for any transparent, filings-mechanical construct. Judging the Gap as a one-shot stock-picker was the wrong test; it is one lens of two.

4.6 The Gap versus the Margin of Safety, head-to-head

The Margin of Safety is the canonical answer to the pricing question, so it is the right benchmark — and because we ran every test on both, the comparison is direct and complete:

The summary is one sentence: on every dimension where the two instruments can be distinguished, the Gap is the better-built one; on the one dimension where both fail — standalone winner-picking — everything fails, because pricing is half the picture. This is not a criticism of Graham’s principle. It is its operationalization: the Margin of Safety made the pricing question primary but left it unmeasurable; the Gap makes the same question answerable from filings, testable, and tested.

5. The Brina Matrix — diagnosing mispricing on two axes

No single metric picks stocks — not ROIC, not the Margin of Safety, not the Gap (companion audit, §2.2). A metric’s power is realized on its axis, and the Gap’s axis is pricing: it becomes one axis of a two-axis diagnostic framework — the Brina Matrix — that crosses fundamental quality (is this a good, durable business?) with pricing (what is the market expecting?). The Gap is the rigorous, validated pricing axis (§4); the fundamentals axis is anchored by ROIC — the one quality metric that survives on large caps (companion study) — but involves judgment no ratio captures (moat durability, management, competitive trajectory).

This is not the earlier “Brina Matrix” that crossed the Gap with the Margin of Safety — two pricing-side measures, which was redundant (the original construction and its full record are preserved in Part II, §2.5 and §4.5). Crossing the Gap with quality is the genuinely two-axis version: the two questions are independent, and a stock’s situation is read from where it sits.

The Gap is what makes this diagnosis possible: without the pricing axis you cannot separate Underpriced Quality (good and cheap — a candidate bargain) from Priced for Perfection (good but demanding), nor — critically — a genuine bargain from a Value Trap. For the full-universe map below, quality is defined precisely: a within-year composite of capital efficiency, cash generation, balance-sheet strength and capital discipline — \(z(\text{ROIC}) + z(\text{FCF-ROIC}) + z(-\text{net leverage}) + z(\text{net buyback})\) — split into terciles; pricing is the Gap in terciles (high = priced cheap). Mean five-year forward total-return excess per corner cell (n = 2,106 firm-years overall):

Quality is the dominant axis — good businesses outperform across the board. The pricing axis is a quality-gated refinement, not a symmetric driver: among high-quality firms, cheap (a positive Gap) is a modest bargain (+0.6%/yr under this specification); among low-quality firms, cheap is the worst outcome (−3.8%/yr) — the classic value trap, because low-quality firms are cheap because they deserve to be. The Gap’s “cheap” reading is therefore actionable only after the fundamental question is answered. This is the framework’s core practical rule: never buy “cheap” without first establishing quality.

The two cells are not equally robust — and we state the asymmetry plainly. The trap cell is the map’s replicable content: cheap + low-quality is the worst corner under every specification we tried — quality as the composite or as ROIC alone, terciles or median splits, early (fiscal years ≤ 2014) or late sub-period — ranging from −3.8 to −6.2%/yr. The bargain cell is not robust: +0.6%/yr under the composite terciles, but ≈ 0 to mildly negative with ROIC-alone quality or median splits, and weaker in the later sub-period — consistent with §8.1, where no pricing measure earned a positive spread in this regime. The framework’s durable practical rule is therefore the warning, not the bonus: the Gap’s “cheap” is first a screen-out tool against low-quality traps; treating it as a bargain-finder among quality names is regime-dependent.

(These are descriptive characterizations, not a tradeable alpha — consistent with §4.5. The map is mostly the quality gradient; the Gap adds the expectations reading and the trap warning.)

6. Case studies — reading both lenses (FY2018, five-year outcomes)

The two-lens read is best seen on real firms. We take recognizable names and ask both questions — good business? and what’s priced in? — then follow the outcome. The five cases cover every quadrant and the framework’s limit: an aligned win, a “rich price justified,” an honest miss, the danger zone, and the value trap.

• Apple — Underpriced Quality (good business, low expectations) → the aligned win. ROIC 58% (excellent); the market priced only ~7% growth (\(g^* = 7.3\), Gap +14, cheap). Both axes bullish; the business out-grew the modest bar. 5-yr excess +23%/yr.
• Microsoft — Priced for Perfection (good business, demanding bar — cleared). ROIC 36%; the market priced ~13% growth, above its ~9% ceiling (Gap −3, rich). The pricing lens flagged a demanding expectation — and the fundamentals delivered (realized growth 21%). +14%/yr. The lesson: a rich price on a great business can still work if the business clears the bar — which is a fundamental judgment the Gap can’t make for you.
• Intel — Underpriced Quality on the metrics, but the honest miss (the framework’s limit). Looked good on both mechanical axes: high ROIC (22%) and cheap (\(g^* \approx 0\), Gap +17). Yet the moat broke — earnings fell ~68% — and the stock returned −16%/yr. No snapshot of price-and-current-quality can see a competitive collapse coming; the fundamental axis demands forward judgment, not just a current ratio.
• Pfizer — Speculative / Overpriced (weak business, rich expectations) → the danger zone. ROIC 3% (weak); the market priced ~29% growth (Gap −34). Both lenses bearish, and it played out: −18%/yr. The danger zone — exactly the cell the §5 map flags.
• AT&T — Value Trap (weak business, “cheap” price) → the framework’s flagship rule in action. The market priced a mild decline (\(g^* \approx -2.6\%\)); on the raw numbers the Gap read +18 — loudly cheap. The §2.1 protocol immediately turns the volume down: that reading rested on RR = 210% (the Time Warner acquisition closing), an artifact year; bounded, the Gap is ≈ +10 — still genuinely cheap, just honestly so. But the quality lens said low: ROIC 7.3% and heavy leverage put AT&T in the bottom tercile of the §5 composite. Cheap + low quality = the trap cell, and it played out: −13%/yr five-year total-return excess. This is the one case where acting on “cheap” alone is most tempting — a household name at a pessimistic price — and exactly where the rule never buy cheap before establishing quality earns its keep.

Together: the Gap tells you what the market is betting on; quality (with judgment) tells you whether the business can deliver it; and the combination — not either alone, and never “cheap” without quality — is the discipline.

7. How to use it

The Gap is a pricing/expectations lens, not a verdict. In practice: 1. Make the expectation explicit. Read \(g^*\): what growth is the price demanding? Is that plausible for this business? 2. Answer the fundamental question separately — quality, durability, the trajectory of ROIC — with judgment beyond any single ratio. 3. Combine, and respect the trap rule: a positive Gap (cheap) is interesting only once quality is established; on a weak business it is a warning, not an opportunity. 4. Mind the horizon: \(g^*\) is most reliable as a ~7–10-year expectation; the market over-extrapolates the near term.

This is the Margin-of-Safety discipline made mechanical and falsifiable, and the expectations-investing idea made reproducible at scale.

8. Scope and limitations

• One market, one regime (large-cap U.S., 2010–2024 — quality-led, value-hostile; see §8.1). The value-trap result and the quality dominance are regime-shaped; behaviour in a value regime or other markets is future work.
• The fundamental arm is an anchor, not a forecast (§4.5) — deliberately conservative and nearly unbiased as a benchmark; the validated forecasting content is the pricing arm \(g^*\).
• Not a standalone signal (§4.5) — a one-lens-of-two tool, framed as such; a property shared by every metric tested here and in the companion audit, the Margin of Safety included.
• Case studies illustrate the discipline; they are not evidence of a return edge (§5).

8.1 The pricing axis was stress-tested in its most hostile regime

A fair reading of this paper must account for when the Gap was evaluated. 2010–2024 was the documented value drought: post-GFC, mega-cap quality and growth dominated and cheapness was penalized. Three independent cuts of our own data confirm the Gap was measured squarely inside that environment — and, critically, that every pricing-based alternative fared the same or worse.

• Unconditionally. Sorting the full universe into within-cohort quintiles, the cheapest quintile by the Margin of Safety underperformed the most expensive by −0.8pp/yr in five-year forward excess return; the Brina Gap’s cheapest-minus-richest spread was identical (−0.8pp/yr). Raw cheapness — the purest expression of the value premium — had a negative payoff in this universe and era.
• Conditionally (the matched-pairs test). Holding fundamental quality fixed — matched within sector-year on ROIC, free-cash-flow ROIC, leverage, and net and operating margin — neither metric differentiated returns between two near-identical firms one called cheap and the other rich: the cheaper firm beat the richer 47% of the time by the Gap and 45% by the Margin of Safety, both statistically indistinguishable from a coin flip (≈110 pairs each, SE ≈ 5pp).
• In the full league table. The clearest way to see that the Gap’s return-null is the normal result — not a defect — is to rank it among the most celebrated metrics in finance, on the same sample and the same statistic (pooled rank correlation with 5-year forward excess return; companion audit, identical methodology):

The Gap return-sorts better than gross profitability, EPV, ROE, the Piotroski score, the Margin of Safety, and — by a wide margin — book-to-market, the canonical academic value factor. Among the pure valuation metrics — book-to-market, earnings yield, EPV/price, the Margin of Safety — the Gap ranks first. Nobody calls Novy-Marx profitability or the Fama-French value factor “useless” because they went flat in this sample; the same standard applies here. The only robust sorters are ROIC and FCFROIC — quality metrics, which is exactly why quality is the Matrix’s other axis (§5). (Gap and Margin of Safety computed on n = 2,285 firm-years with both metrics and a completed window; companion screens on its common sample of n = 2,103 — same universe, same windows, same statistic.)

The lesson is not that pricing measures fail in general; it is that the entire pricing axis — the Gap, the Margin of Safety, and raw cheapness alike — produced no positive return spread in this regime. That is a property of 2010–2024 large-cap U.S. equities, not a defect specific to the Gap, and it sharpens the paper’s central position: pricing is one axis of two, never a standalone signal. It also bounds the claim precisely — a flat spread here is evidence about the regime, not evidence against value investing as a discipline, whose premium is documented over broad universes (including small caps) and long, multi-regime horizons that an S&P-500/2010–2024 sample cannot capture. That the expectations-measurement content of \(g^*\) (§4) holds up even in the regime most hostile to every pricing-based approach is, if anything, the more demanding test — and it passes.

8.2 Future work

• Other markets and a value regime. The first-order out-of-sample test: does the trap rule hold, and does the bargain cell reappear, where value has paid (small caps, non-U.S., pre-2010)?
• External convergent validity. Compare \(g^*\) to analyst long-term-growth forecasts (IBES-style) — an expectations measure from an independent source; cross-method agreement (§4.2) is the within-data substitute.
• Firm-specific discount rates. The flat \(r = 10\%\) is deliberate (and §4.4 shows rankings are parameter-invariant), but a firm-specific \(r\) would let \(g^*\) be read as a level, not only a ranking.
• The refined construct, validated forward. The §2.1 protocol (bounded RR, smoothed \(g_f\), artifact flags) defines a refined Gap; validating it out-of-sample on maturing post-2020 windows — pre-registered, not tuned — is reserved for a sequel paper.

9. Reproducibility

The point-in-time membership, firm-year dataset, corporate-event records, and full computation pipeline are published as companion files; every figure and statistic can be reproduced. To our knowledge this is among the few fully point-in-time, survivorship-complete S&P 500 fundamental datasets published openly — 100.0% of constituent-years covered, including every firm later acquired or delisted. A plain-language companion is maintained at Zyberno.com.

Part II — The Empirical Record

Part II is the complete, self-contained empirical record behind Part I: the full directional-test results, the universe construction, robustness, failure modes, and the pre-registered refinements. It preserves its own internal numbering — section references inside Part II (e.g. §6.6) refer to Part II. Readers who want the argument should start with Part I; readers who want to check it should start here.

1. Introduction

1.1 The question

Consider a firm that earns 25% on its invested capital, reinvests 40% of its earnings back into operations, and trades at a price that — under a standard reverse DCF on its enterprise value — would only be rational if its operating cash flows grew at 13% per year forever. Is the stock cheap or expensive?

The arithmetic is unambiguous. The firm can sustainably grow at 25% × 40% = 10%. The market is paying for 13%. The implied future is incompatible with the firm’s own reinvestment economics: either ROIC must rise, margins must expand, or the multiple must compress. Two of those three are outside the firm’s control.

The thesis of this paper is that the difference between what the firm can produce and what the market is paying for is (a) a signed, falsifiable, single-number measurement of mispricing, and (b) empirically superior to the Margin of Safety as Graham defined it.

1.2 What is missing from the literature

The value-investing tradition holds that price and value are separable. Graham (1934, 1949) gave this its canonical form via the Margin of Safety: buy when price sits well below intrinsic value, with a buffer. Buffett’s letters operationalized it for compounders, introducing the owner-earnings concept (1986) that separated reported earnings from the cash flow available after maintenance reinvestment. Damodaran (2002, and continuing) modernized the discounting machinery and, importantly, suggested reverse DCF as a discipline — asking what growth a market price implies given the cash-flow stream. Greenwald et al. (2004) provided the Earnings Power Value alternative, separating “no-growth value” from “value of growth.” Greenblatt (2005) introduced the Magic Formula combining ROIC and earnings yield.

What is missing — and what this paper supplies — is a single, signed, falsifiable scalar that closes the loop: a measurement that does not require the analyst to commit to a private valuation; that incorporates reinvestment quality; that can be computed mechanically from filings; and that produces a firm-by-firm directional prediction testable in a panel.

1.3 Contribution

This paper contributes:

1. A definition of the Brina Gap as \(g_f - g^*\), where \(g_f = \text{ROIC} \times \text{RR}\) is the firm’s internally financeable growth ceiling and $g^* $ is the market-implied growth rate from a reverse DCF on enterprise value. Both are mechanical functions of public filings.

2. A derivation of why \(g_f\) is a ceiling (not a forecast) — a sustainability identity rooted in the owner-earnings construction and the steady-state growth equation.

3. A construct-validity test (§6.6–6.7) — the paper’s central empirical contribution — showing that the Gap measures a real economic quantity: \(g_f\) is the realized growth ceiling of stable-ROIC firms, and, holding the market’s implied growth fixed, firms that can finance the priced growth re-rate while those that cannot de-rate. This validates the measurement on realized fundamentals, independent of any return-prediction claim.

4. A direct comparison to the Margin of Safety, both theoretically and empirically. The two are complementary lenses — MoS reads price-versus-value, the Gap reads economics-versus-expectations — and the result identifies reinvestment-quality (ROIC and reinvestment rate) as a load-bearing axis the standard owner-earnings DCF discards. On the firm-years where the two signals disagree, the Gap carries the information the price-versus-value comparison misses.

5. A panel empirical test on the full S&P 500 (2010–2024), with mechanical universe selection, pre-registration, citation-backed exclusions, and reviewer-verifiable computations — corroborating the construct with directional content.

6. A taxonomy of four failure modes — domains where the framework should not be expected to predict — each empirically grounded and named.

7. A complete reproducibility package: pre-registration document, 228 corporate-event citations, pipeline code, and full dataset published as companion files.

The paper is theory-first. Sections 2–4 stand without empirical support; Sections 5–7 then test whether the world cooperates — first that the Gap measures a real quantity (§6.6–6.7), then that it carries directional content (§6.1–6.5).

2. Theoretical Framework

2.1 Owner earnings and the sustainable growth identity

Buffett (1986) defined owner earnings as reported earnings plus depreciation and amortization, minus maintenance capital expenditure and working-capital reinvestment required to maintain unit volume and competitive position. The distinction from accounting earnings is that some of what GAAP calls “earnings” is, in fact, required reinvestment to stand still. Owner earnings represent the cash flow genuinely available to owners — what could be distributed without compromising the business.

From owner earnings flows a steady-state identity. Let \(E_t\) denote NOPAT (net operating profit after tax) at time \(t\), and \(IC_t\) the invested capital base. Then by definition:

\[\text{ROIC}_t \equiv \frac{E_t}{IC_t}.\]

If the firm reinvests a fraction \(b\) (the reinvestment rate, or plowback) of those earnings into operations, the incremental invested capital next period is \(b \cdot E_t\). Assuming the marginal dollar of new capital earns the same return as the average historical dollar — the steady-state assumption — incremental earnings next period are:

\[\Delta E_{t+1} = \text{ROIC}_t \cdot b \cdot E_t,\]

so

\[g_f \equiv \frac{\Delta E_{t+1}}{E_t} = \text{ROIC}_t \cdot b.\]

This is the sustainable growth identity. A firm cannot grow owner earnings faster than \(\text{ROIC} \times b\) without (a) raising external capital, (b) increasing leverage, or (c) experiencing a one-time step-change in ROIC.

The qualifier matters: \(g_f\) is a ceiling, not a forecast. It is the upper bound on growth that current reinvestment economics support. If the firm executes well, it achieves \(g_f\). If not, it falls short. A growth assumption exceeding \(g_f\) requires operational improvement, capital injection, or luck.

The identity is not new — it appears in standard corporate-finance textbooks (Ross, Westerfield, Jaffe, Chapter 5) under the name “sustainable growth rate,” and Damodaran has emphasized its analogue (“$g = $ ROIC \(\times\) reinvestment rate”) for two decades. What is novel is its operational pairing with a reverse-DCF-derived implied growth.

2.2 Reverse DCF and the market-implied growth rate

A standard two-stage DCF on enterprise value computes the present value of a growing operating-earnings (NOPAT) stream discounted at a discount rate \(r\), with a terminal-growth assumption \(g_T\):

\[EV = \sum_{t=1}^{N} \frac{NOPAT_0 (1+g)^t}{(1+r)^t} + \frac{NOPAT_0 (1+g)^N}{(r-g_T)(1+r)^N}.\]

The reverse DCF inverts this: given observed \(EV\), \(NOPAT_0\), \(r\), and \(g_T\), solve for the single explicit-period growth rate \(g^*\) that reconciles them. \(g^*\) is the market-implied growth rate — the operating-earnings growth assumption embedded in today’s enterprise value, given the stipulated discount and terminal-growth conventions. (We grow NOPAT rather than free cash flow: both arms of the Gap are then expressed in the same operating-earnings space, and the reverse DCF reads only the cash the market is capitalizing, not the reinvestment that produced it.)

In this paper we standardize the inputs to remove analyst degrees of freedom:

• \(r = 10\%\) — a single uniform discount rate applied to every firm, used as a proxy for the long-horizon cost of capital. We deliberately avoid per-firm WACC estimation: a uniform rate keeps \(g^*\) free of analyst degrees of freedom. §8.2 examines the sensitivity of the result to ±100bp perturbations of \(r\) — the direction is preserved across the range, though the magnitude is not (the headline weakens at lower \(r\)).
• \(g_T = 3\%\), consistent with Damodaran’s terminal-growth conventions and a long-run nominal GDP proxy.
• Explicit horizon \(N = 10\) years.

The key property of \(g^*\): it does not depend on the analyst’s view of what the firm is worth. It depends only on what the market is paying, which is observable. The reverse DCF is mechanical; two analysts using the same \(r\) and \(g_T\) on the same EV and NOPAT must produce the same \(g^*\).

2.3 The Brina Gap

The Brina Gap is the difference between the two:

\[\boxed{\text{Gap} \equiv g_f - g^* = \text{ROIC} \times b - g^*.}\]

The Gap has three properties that the Margin of Safety lacks:

1. It is a function of the firm and the market — not of the analyst. Both \(g_f\) and \(g^*\) are derived mechanically from filings and market quotes.

2. It is signed and scaled. Gap > 0 means the firm’s internal growth ceiling exceeds what the market is paying for — undervalued on reinvestment economics. Gap < 0 means the market is paying for growth the firm cannot finance from internal returns.

3. It is falsifiable. A reader can compute the Gap for any firm, observe future excess returns, and reject the framework if the directional signal fails to hold across a sufficient sample.

2.4 Why \(g_f\) and \(g^*\) are nearly independent

This is the load-bearing claim. The Brina Gap’s predictive content rests on \(g_f\) and \(g^*\) measuring largely orthogonal aspects of the firm.

• \(g_f\) is operationally anchored and backward-looking. It is computed from the trailing year’s balance-sheet invested capital, NOPAT, and observed reinvestment fraction. It does not see the stock price.

• \(g^*\) is market-anchored and forward-looking. It is computed from enterprise value, operating earnings (NOPAT), and discounting assumptions. It is essentially indifferent to ROIC or reinvestment rate — the reverse DCF reads only the earnings level the market is capitalizing, not the quality of reinvestment that produced it.

The two terms can disagree by arbitrary amounts. A firm with 30% ROIC and 25% reinvestment rate has \(g_f = 7.5\%\). A firm with 10% ROIC and the same 25% reinvestment rate has \(g_f = 2.5\%\). The market’s \(g^*\) may be 12% for either firm — the reverse DCF doesn’t see the difference in capital quality. The Gap captures it.

Empirically (Section 6), the Pearson cross-sectional correlation between \(g_f\) and \(g^*\) on the 3,402-firm-year valid universe is 0.03 untrimmed and 0.10 after trimming observations with \(|g_f| > 100\%\) or \(|g^*| > 100\%\) (which arise from edge-case denominators and are excluded from the directional tests). The two terms are nearly orthogonal, confirming the theoretical claim.

2.5 The Brina Matrix — a diagnostic combination of the Brina Gap with the classical MoS

Terminology (v3.0). Part I presents the framework’s current form: the Brina Matrix as quality × pricing. The Gap-×-MoS layout below is the original construction, preserved unchanged in this record because the directional tables’ quadrant labels (Value Trap, Expensive Hype, Underestimated Growth, Double Discount) are defined by it — and because its central finding (§4.5: the MoS axis adds no directional information) is precisely what motivated the move to the quality axis in Part I.

The Brina Gap as defined in §2.3 is this paper’s primary measurement of mispricing. The Brina Matrix is a complementary diagnostic layout: it pairs the Gap with the classical signed Margin of Safety — the residual between owner-earnings DCF intrinsic value and market capitalization — to ask whether combining the two signals adds information. The Matrix is not the headline construct of this paper; the Brina Gap is. The Matrix is a layout for testing whether the classical MoS adds anything beyond what the Brina Gap alone provides.

The two-by-two classification, on firm-years where both signals are outside their respective neutral bands (|Gap| > 3pp; |MoS| > 20%):

Figure 1. The Brina Matrix conceptual layout. Horizontal axis: Margin of Safety (left = expensive by DCF, right = cheap by DCF). Vertical axis: Brina Gap (top = \(g_f > g^*\), bottom = \(g_f < g^*\)). The four corner quadrants are the directional cells defined in the table above.

The original theoretical motivation is the standard Bayesian intuition: two independent mispricing measurements, when in agreement, should strengthen the directional posterior beyond what either alone provides. The four archetypes follow:

• Underestimated Growth (Gap+, MoS−): Gap detects compounding economics the DCF misses. Predicted positive excess return.
• Double Discount (Gap+, MoS+): both signals say cheap. Predicted strongest positive (theory).
• Value Trap (Gap−, MoS+): market is paying for growth the firm cannot finance, even as DCF says cheap. Predicted negative excess return.
• Expensive Hype (Gap−, MoS−): both signals say expensive. Predicted strongest negative (theory).

The empirical results (§6.2) reject this naïve Bayesian intuition — but not in the way agreement-versus-disagreement would suggest. On the full point-in-time universe the directional content lies almost entirely in the Gap’s sign, not in its pairing with MoS. Pooling by Gap sign, the short side (Value Trap + Expensive Hype) hits 58.9% while the long side (Underestimated Growth + Double Discount) sits at 48.2% — a 10.7-percentage-point asymmetry. Pooling instead by agreement, the two agreement quadrants (Double Discount + Expensive Hype, 54.6%) and the two disagreement quadrants (Underestimated Growth + Value Trap, 55.0%) are statistically indistinguishable. The MoS axis adds essentially nothing to the direction of the prediction; the Gap sign carries it, asymmetrically.

This is visible quadrant by quadrant. Among the DCF-cheap firms (MoS+, right column), the Gap sign decides the outcome: Value Trap (Gap−) underperforms at 59.5%, while Double Discount (Gap+) is at chance (47.9%). Among the DCF-expensive firms (MoS−, left column), likewise: Expensive Hype (Gap−) hits 58.5%, while Underestimated Growth (Gap+) is at chance (48.5%). Whether MoS agrees or disagrees with the Gap is immaterial; whether the Gap is negative is decisive.

The classical value-trap literature explains why MoS is uninformative here. The DCF-based Margin of Safety carries a structural value-trap selection bias: a firm reads cheap to a DCF (MoS+) because its trailing earnings extrapolate fundamentals that are, on average, deteriorating — the Lakonishok-Shleifer-Vishny (1994) contrarian finding, the Sloan (1996) accruals anomaly, and the earnings-quality literature (Penman 2007). MoS+ therefore does not signal future outperformance, which is exactly why the MoS axis fails to separate outcomes that the Gap sign does. The Matrix’s role is thus confirmed as a diagnostic taxonomy of mispricing types — not a directional signal combination (§4.5): the four labels name why a firm is mispriced (cheap-and-financeable versus cheap-but-unfinanceable, and so on), but the predictive work is done by the Gap alone, concentrated on the short side.

We test the framework first via the single-signal Gap, then via the four-quadrant Matrix, and contrast both with single-signal MoS in Section 4.

3. Comparison to Existing Frameworks

3.1 Margin of Safety (Graham, Buffett)

Graham (1934, 1949) defined Margin of Safety as the gap between intrinsic value and price, where intrinsic value is computed by the analyst using “conservative assumptions.” Its operational weakness has been recognized for decades: every analyst arrives at a different intrinsic value, and “30% margin of safety” calculated by one observer is “no margin” by another. MoS is, as a result, a principle rather than a measurement — it cannot be tested at panel scale because each test reduces to whether the analyst’s specific intrinsic-value model worked, not whether MoS worked.

The Brina Gap inherits MoS’s conservatism — it favors firms whose own economics support the price paid — but replaces the analyst-defined intrinsic value with two externally verifiable terms. Both Graham and Buffett would, we contend, recognize the Gap as the natural quantitative form of what they were doing intuitively when they preferred “a wonderful business at a fair price” to “a fair business at a wonderful price.” The Brina Gap names which “own economics.”

The substantive improvement is operational, not philosophical: a reader checking the analyst’s work needs only the firm’s filings and the reverse-DCF assumptions, both of which are documented in a footnote. There is no opaque private model.

Empirically (Section 4), we test MoS computed from a standardized owner-earnings DCF against the same universe as the Brina Gap. On a mechanical sign-test it produces a 51.2% directional hit rate (p=0.12 against the 50% null) — it does not, on its own and reduced to a single sign, sort returns on this broad universe. This is no indictment of the Margin of Safety as an investment principle, which remains foundational; it is a statement about one narrow operationalization, and we apply the identical test to the Brina Gap.

3.2 Damodaran implied-growth

Damodaran has advocated reverse DCF as a discipline for two decades, and Mauboussin and Rappaport’s (2001) expectations investing program made the decoding of market-implied expectations from price its central method. The Brina Gap differs in two ways:

First, Damodaran’s typical framing uses the reverse DCF as a sanity check — he asks whether the implied growth passes a reasonableness test against the firm’s history and industry. The Brina Gap operationalizes the sanity check into a single signed scalar with a directional prediction.

Second and more importantly, the Brina Gap pairs \(g^*\) with \(g_f\) (mechanical from filings) rather than with a subjective growth forecast. The comparison is between two estimable quantities, not between one estimable quantity and one analyst belief. This methodological move is what makes the framework testable at panel scale.

3.3 Earnings Power Value (Greenwald)

Greenwald et al. (2004) separate “value if growth equals zero” from “value of growth.” Conceptually adjacent but operationally different from the Brina Gap: EPV requires the analyst to compute normalized earnings, which inherits the difficulty of distinguishing recurring from non-recurring components — a long-running debate in the accruals literature (Sloan 1996; Penman 2007).

The Brina Gap finesses normalization by working in growth space, not value space. We never compute “what the firm is worth”; we compute “what growth is compatible with paying this much” and compare it to “what growth the firm’s reinvestment supports.” The arithmetic is identical to a one-stage DCF, but the framing makes the comparison directly interpretable.

3.4 Magic Formula (Greenblatt)

Greenblatt (2005) ranks firms by (high ROIC) × (high earnings yield). It is a screen, not a measurement: the Magic Formula produces a portfolio recommendation but no firm-by-firm prediction. The Brina Gap, by contrast, makes a directional prediction per firm-year.

The Magic Formula and the Brina Gap should correlate (high ROIC + low EV/EBITDA tends to produce a positive Gap), but the Gap is informative where the Magic Formula is silent: it identifies Value Traps (low \(g_f\), high \(g^*\)) which a two-factor sort cannot. In our panel, Value Trap is the framework’s strongest single quadrant (59% directional accuracy, p=0.001) — a finding Greenblatt’s framework cannot produce.

3.5 PEG ratio

PEG = P/E divided by earnings growth rate. The denominator is typically a sell-side forecast — the most error-prone input in equity analysis. PEG inherits all of forecast-driven bias. The Brina Gap replaces the analyst-forecast denominator with \(g_f\), a ceiling derived from filings.

3.6 Residual income and abnormal-earnings valuation (Ohlson, Penman)

The framework closest to the Brina Gap in spirit is the residual-income (abnormal-earnings) tradition — Ohlson (1995) and especially Penman’s Accounting for Value (2010). It values a firm as book value plus the present value of residual income — earnings in excess of a charge for the capital employed — and makes the growth in residual income the central object of analysis. Like the Gap, it works in earnings-versus-capital space and shares the core intuition that value is created only when returns exceed the cost of capital, and that growth matters only insofar as it is financed at those returns.

The Brina Gap differs in what it produces. Residual-income valuation yields an intrinsic value, which the analyst then compares to price — re-introducing the cost-of-capital and forecast commitments (and the observer-dependence) the Margin of Safety carries. The Gap declines to value the firm at all: it compares two growth rates — what reinvestment can finance (\(g_f\)) against what the price assumes (\(g^*\)) — collapsing the same economic content into a single signed scalar that needs no intrinsic-value estimate. Empirically the two are kin but not identical: §6.8 shows the Gap’s \(g^*\) is convergent (Spearman 0.73) with a residual-income-implied growth, the divergence being precisely the capital-intensity dimension the forward arm \(g_f\) restores. The Brina Gap can fairly be read as residual-income thinking recast as a falsifiable growth difference.

3.7 Summary

4. Empirical Validation: Brina Gap vs Margin of Safety

This section presents the paper’s central empirical finding. The theoretical argument of Sections 2–3 claims the Brina Gap is a distinct signal that should outperform the Margin of Safety. Here we test whether it does, on the same panel, using sign-only single-signal tests and a controlled disagreement analysis.

4.1 Test design

For each firm-year in the cleaned panel (full S&P 500 2010–2024, 228 citation-backed corporate-event exclusions, BC1–BC6 applied — see §5), we compute three signals:

• MoS sign: +1 if the owner-earnings DCF intrinsic value exceeds market capitalization (firm cheap), −1 otherwise.
• Brina Gap sign: +1 if \(g_f > g^*\) (internal economics support more than market is paying for), −1 otherwise.
• Brina Matrix quadrant: the four-quadrant combined classification of §2.5.

The realized outcome is sign of forward 5-year annualized excess return versus SPY (matched window). A hit is a sign-match.

We require both signals computable and the 5-year forward window resolved (completed, or closed at a delisting/deal terminal): n = 2,225 firm-years for the single-signal tests; n = 1,504 for the four-quadrant directional test (smaller because Fairly Valued and Mixed quadrants are excluded from directional tests). All returns are dividend-reinvested total returns on the full point-in-time universe.

4.2 Margin of Safety as a standalone signal

Margin of Safety, reduced to a single mechanical sign on the full point-in-time universe, is statistically indistinguishable from a coin flip (51.2%, p=0.12): it does not, by itself, sort forward returns. We stress this is a property of the one-sign operationalization on a broad universe, not a verdict on the Margin of Safety as a principle — a skilled analyst’s intrinsic-value judgment is precisely what this mechanical test strips out. The point is narrower and constructive: the price-versus-value axis alone is incomplete, and §4.4 shows the Gap supplies what it misses.

A more demanding test — does MoS at least sort returns? — is the quartile analysis:

The quartiles show no monotone sort: the “cheapest” firms by Margin of Safety (Q4) do not outperform the “most expensive” (Q1) — all four quartiles cluster between −1.3 and −2.0 pp/yr (the cap-weighted-SPY drag documented in §6.4). On the broad universe the price-versus-intrinsic-value axis alone does not order forward returns, the classical value-trap problem: a firm trading below DCF intrinsic value often does so for a reason — its operating economics are deteriorating — so the discount persists rather than closes.

We do not interpret this as proof that “value investing doesn’t work.” We interpret it as confirmation that price-vs-intrinsic-value alone is a one-axis projection of a two-axis problem. The Brina Gap supplies the missing axis.

4.3 The Brina Gap as a standalone signal

The Brina Gap alone, on the same 2,225 firm-years, predicts at 52.4% — modestly but genuinely significant on a pure sign-only basis (p=0.013 naive binomial; the edge persists but is weaker under the window-clustered inference of §6.1). Significance strengthens once a minimum signal magnitude is required (the sign-only test includes near-zero gaps in the measurement noise floor): the |Gap| > 3pp band reaches 54.1% (§4.5, §7.4). The quartile sort runs in the predicted direction:

The most-negative-Gap quartile (Q1) has the most negative mean excess and the most-positive-Gap quartile (Q4) the least negative, so the Q1→Q4 ordering runs in the framework’s predicted direction. The Gap sorts where MoS is flat — though, as in §6.4, all quartiles run negative on this cap-weighted universe, which is why the sign-based hit-rate test, not the mean level, carries the result.

4.4 The disagreement subset

The cleanest test of incremental information content is: when the two signals disagree, which one is right?

• MoS and Brina Gap agree on 1,296 firm-years (sign-match). Joint hit rate: 53.1%.
• MoS and Brina Gap disagree on 929 firm-years (sign-mismatch). On this subset:
  • Brina Gap correct: 51.3%
  • Margin of Safety correct: 48.7%
  • Incremental Brina Gap advantage: +2.7 percentage points

When the two signals say different things, the Brina Gap wins — modestly at the sign-only level (+2.7pp), and decisively on the strong-signal subset where both |Gap|>3pp and |MoS|>20%: +10.1 percentage points (n=587; §4.5). The strong-signal disagreement is the cleaner test — it removes the noise-floor near-zero gaps that dilute the sign-only comparison — and there the Gap’s incremental content over the Margin of Safety is large. The two-axis decomposition — firm-economics (\(g_f\)) versus market-expectation (\(g^*\)) — provides genuine information beyond the single-axis price-vs-intrinsic-value comparison.

The interpretation: a firm classified as “cheap” by MoS but with \(g_f < g^*\) (Brina Gap says expensive) is a likely value trap — the market is discounting it correctly because internal economics cannot support even the discounted price. A firm classified as “expensive” by MoS but with \(g_f > g^*\) (Brina Gap says cheap) is a likely compounder mispriced on multiple — the high price reflects the firm’s quality but undershoots its compounding ceiling.

4.5 The Brina Gap drives the result; the Matrix adds no incremental directional information

The headline result — 54.8% on n=1,504 — is the Brina Gap with a confidence band, not a signal combination with MoS. We show this here as a precise decomposition: the lift from single-signal Gap (52.4% on n=2,225) to the strong-signal headline comes from the |Gap| > 3pp filter, not from the Matrix’s two-axis pairing with MoS.

By construction, within the four directional quadrants the Matrix prediction reduces to sign(Gap): Underestimated Growth and Double Discount (both Gap > +3pp) predict positive; Value Trap and Expensive Hype (both Gap < −3pp) predict negative. The MoS sign determines which quadrant label a firm-year receives, not which direction is predicted. The n=2,225 → n=1,504 sample reduction therefore acts as a joint confidence band on the two signals — selecting firm-years where both |Gap| > 3pp and |MoS| > 20%.

The relevant comparison is Gap-alone applied to the same confidence-banded sample:

Three findings:

1. The Matrix and Gap-alone produce identical predictions on the directional quadrant-assigned subset (824/1,504 hits in both). The Matrix is not a signal-combining classifier on its prediction face; it is sign(Gap) applied to a strong-signal subset.

2. The |Gap| > 3pp confidence band carries essentially all of the lift. Gap-alone restricted only by |Gap| > 3pp (n=1,729) hits 54.1% — within a fraction of the directional subset’s 54.8%. The additional |MoS| > 20% filter contributes little. The lift from 52.4% (sign-only) to 54.1% comes from discarding near-zero gaps in the noise floor (§7.4).

3. MoS is at chance on the same subset where the Matrix succeeds. Restricted to the directional subset, MoS-alone hits 50.9% — a coin flip — while the Gap reaches 54.8%. MoS adds no directional information at any signal strength on this universe.

This decomposition does not weaken the paper’s central claim. The Brina Gap remains the load-bearing predictor; the Matrix’s role is diagnostic decomposition rather than signal combination. Crucially, the directional content tracks the Gap sign, not the MoS-defined quadrant: within each Gap sign the two quadrants perform almost identically (short side, Value Trap 59.5% \(\approx\) Expensive Hype 58.5%; long side, Underestimated Growth 48.5% \(\approx\) Double Discount 47.9%, §6.2). The MoS axis re-labels these outcomes — naming types of mispricing — but does not separate them directionally; that MoS is uninformative here is consistent with its value-trap selection bias (Sloan 1996; Penman 2007), under which a DCF-cheap reading is on average contaminated by deteriorating fundamentals.

4.6 Interpretation

The empirical structure is clear:

1. Single-axis MoS is at chance. 51.2%, p=0.12.
2. Single-axis Gap (sign only) is modestly significant. 52.4%, p=0.013 — above chance even before any magnitude filter (weaker under window-clustered inference, §6.1).
3. Strong-signal Gap (|Gap|>3pp) succeeds. 54.1%, p=0.0003 — the Matrix’s headline 54.8% is the same prediction restricted to the joint (|Gap|>3pp, |MoS|>20%) subset, and the |MoS| > 20% filter contributes little additional directional information.
4. On disagreement, Gap > MoS by +2.7 pp (+10.1 pp on the strong-signal subset). Direct evidence of incremental information.

The paper’s theoretical claim — that the Brina Gap refines MoS by adding a reinvestment-quality axis — is empirically supported by the disagreement test in particular, which controls for the cases where MoS and Gap coincidentally agree.

A skeptical reviewer may ask: is the failure of MoS specific to our owner-earnings DCF implementation? The robustness checks in §8 examine ±100bp discount-rate and ±50bp terminal-growth perturbations: the direction of the Gap result holds across the range, though the discount-rate sensitivity is non-trivial (§8.2). We note that the question of “the correct MoS” has no canonical answer in the literature, and we invite reviewers to substitute their preferred intrinsic-value methodology using our published pipeline.

5. Methodology

5.1 Universe — full S&P 500, point-in-time

The universe is the full point-in-time S&P 500 across 2010–2024: every firm that was an index constituent in a given year enters the panel for that year, including firms subsequently acquired, merged, or delisted. Point-in-time membership is reconstructed from the index constituent list together with its complete add/remove change history, walked backward year by year to recover the exact membership at each year-end. Each constituent-year is then matched to the firm’s SEC fundamentals — by CIK, including the historical 10-K filings of delisted issuers, which persist on EDGAR — and to its as-printed fiscal-year-end price.

The universe was specified in a pre-registration document (pre_registration.md) committed to the public repository on 2026-05-24, with two dated amendments (top-200 expansion, full S&P 500 expansion) committed on the same and following day. All amendments precede the final compute. The audit trail is verifiable from the repository.

The reconstruction yields 771 distinct firms across 7,578 constituent-years; the panel covers 7,576 of these (100.0%) — every constituent-year except two, for a single failed bank (Signature Bank) that the financial-institution boundary condition (§5.3) suspends regardless. This is a genuine survivorship-free index panel: delisted and acquired firms are retained with their realized terminal outcomes (§5.5), so the directional tests are not conditioned on survival. A computable Brina Gap is obtained for 73% of constituent-years; the remainder are boundary-condition suspensions (§5.3) plus a residual set of pre-2016 acquisitions for which point-in-time prices are no longer available from free data sources (a disclosed limitation, §8.6).

5.2 Inputs and computations

For each firm-year in the panel:

• NOPAT = EBIT × (1 − tax rate), where tax rate is the statutory rate where applicable, else the effective rate trimmed to [0%, 35%].
• Invested Capital = total equity + total debt − cash & equivalents − non-operating assets. Computed at fiscal-year-end and averaged with beginning-of-year for noise reduction.
• ROIC = NOPAT / average IC.
• Reinvestment Rate \(b\) = (Capex + ΔWC + Acquisitions − D&A) / NOPAT. Negative values and values > 1 trigger BC2 (see §5.3).
• \(g_f\) = ROIC × \(b\).
• FCF = NOPAT − reinvestment.
• EV = market cap (year-end close × shares outstanding) + total debt − cash.
• \(g^*\) = solved from reverse DCF on EV, NOPAT, \(r\), \(g_T\).
• Gap = \(g_f - g^*\).

Owner-earnings DCF intrinsic value, used in the MoS comparison, follows Buffett’s (1986) construction: maintenance-capex-adjusted earnings discounted at the same \(r = 10\%\), with a 10-year explicit horizon and 3% terminal growth.

The complete formulae, XBRL tag mappings, and edge-case handling (e.g., ASC 842 lease accounting, semi-annual filers) are documented in the replication appendix (outputs/file_05_replication_guide.md).

5.3 Boundary conditions

Six pre-specified boundary conditions mark firm-years for suspension — they are computed but excluded from directional tests, with the suspension reason logged:

• BC1: ROIC \(\le\) 0. Firm not earning a return; framework undefined.
• BC2: Reinvestment rate \(\notin\) [0, 1]. Capital allocation in transition.
• BC3: EV \(\le\) 0. Distressed or massive-cash configurations.
• BC4: M&A year distortion — \(|b_t - b_{t-1}| > 30\)pp attributable to a single transformative acquisition (citation-required).
• BC5: Step-change in business model — spinoff year, post-bankruptcy emergence year, IPO year ± 1, REIT conversion year.
• BC6: Financial institutions (banks, insurance, brokers, asset managers — the NOPAT-on-invested-capital identity does not apply to their capital structures).

BC suspension reduces the 8,820 ticker-year computations to 3,402 “valid” firm-years that pass all six conditions.

5.4 Exclusion policy and the events database

Beyond structural BC suspension, individual firm-years may be excluded based on citation-verifiable corporate events in one of four pre-stated categories:

1. Transformative M&A: deal value > 10% of acquirer market capitalization.
2. Material asset impairment / write-down: > 10% of invested capital.
3. Spinoff or split year: the firm reported is materially different from the firm whose forward returns are observed.
4. ASC 842 (2019) operating-lease balance-sheet transition: distorts trailing capex and lease accounting comparability.

Each exclusion is documented in outputs/ticker_events_database.csv with: ticker, fiscal year, event type, dollar amount, deal-close date, and a citation URL (SEC filing, press release, or recognized financial source). The database currently contains 228 corporate-event-driven (ticker, fiscal-year) exclusions, spanning the full S&P 500 panel.

Application of the exclusion policy reduces the directional sample from 2,251 firm-years (raw) to 2,182 firm-years (cleaned). Of these, 1,504 have a completed five-year forward window as of the data cutoff (observation years 2010–2019, plus delisted firms whose windows close at their terminal date); the directional hit-rate tests in §6 are computed on that 1,504.

5.5 Outcome measure and statistical tests

The outcome is 5-year forward annualized total-return excess vs SPY (dividends reinvested on both sides; matched-window), with the forward window starting 90 days after fiscal year-end so the 10-K is public at entry (closing the look-ahead by construction). For firms acquired or delisted within the window, the position is carried to the last traded (deal) price and the proceeds reinvested in SPY for the remainder of the window; bankruptcies realize the terminal delisting return (\(\approx\) −100%). This is standard CRSP-style delisting handling and is what makes the panel survivorship-free. For each directional quadrant assignment:

• Underestimated Growth, Double Discount → predicted positive excess return.
• Value Trap, Expensive Hype → predicted negative excess return.

A hit is a sign-match between prediction and realized outcome.

Statistical tests reported:

• Per-quadrant hit rate with 10,000-iteration year-clustered bootstrap 95% CI.
• Pooled binomial test against the 50% null (normal approximation, given the large \(n\)).
• Bonferroni correction for the four-quadrant test family: \(p_{\text{bonf}} = 4 \times p_{\text{raw}}\).
• Per-sector breakdown (GICS sector classification), with Wilson 95% intervals.
• Mean and median realized 5-year annualized excess return by quadrant.

All reported p-values are one-sided: the framework makes a directional prediction (each quadrant predicts the sign of forward excess return in advance), so the relevant alternative is one-tailed. Two-sided p-values are exactly double those reported: the pooled headline remains significant at the 5% level two-sided (\(p \approx 0.012\)), while the sign-only standalone Gap (one-sided p=0.12) is not significant under either convention — consistent with our framing that the Gap’s significance emerges only once the |Gap| > 3pp magnitude filter is applied (§4.5, §7.4). We flag the convention explicitly so readers can apply whichever they prefer.

The full pipeline is published as Python code; every number in this paper is reproducible from raw SEC filings.

6. Results

6.1 Headline — pooled directional accuracy of the Brina Gap

On the full point-in-time S&P 500 — every constituent-year including acquired and delisted firms, carried to their realized terminal outcomes (§5.5) — the Brina Gap’s pooled directional accuracy on the strong-signal subset (the four directional Matrix quadrants) is 54.8% on n=1,504 completed five-year windows, naive binomial \(p = 1\times10^{-4}\) against the 50% null.

The decisive comparison is against the survivor subset — the firms that happened to survive, which is all a price-data-only study can ever see. Restricting to them reproduces the survivor-biased figure (53.7%, n=1,135). Adding the delisted firms back, with their realized outcomes, raises accuracy to 54.8%, and the short side strengthens sharply: the negative screen (Value Trap + Expensive Hype) rises from 55.9% on the survivors to 58.9% on the full universe (\(p = 3\times10^{-8}\)). Survivorship bias was not inflating this result — it was hiding it. Firms the Gap flagged as overvalued that were then acquired at thin premia or failed outright are correct “underperform” predictions that a survivor panel silently discards. This inverts the usual survivorship worry, and it is only visible because the panel is genuinely point-in-time.

Inference under window dependence. The binomial \(p\) treats overlapping five-year windows as independent — they are not, and a proper accounting weakens the standalone significance while leaving the effect intact. Under a year-clustered bootstrap (resampling fiscal-year cohorts) the directional edge persists, but its absolute significance is modest; we characterise it honestly as a real but modestly-sized edge. For context, that is the ceiling for transparent, filings-based metrics on this universe: the companion audit finds no screen — ROIC, value, quality, or momentum — retains standalone risk-adjusted alpha, and the Gap’s negative screen is the strongest directional result measured anywhere in this research program. This is exactly why the directional finding is offered as corroboration, not as the paper’s claim — the measurement-validity evidence of §6.6–6.9 does not rest on it. The look-ahead objection is closed by construction: the forward window already starts 90 days after fiscal year-end, once the 10-K is public (§5.5). And, decisively, the relative claim is far sturdier than the absolute one: on matched directional firm-years the Gap outpredicts the Margin of Safety by +3.9 percentage points on the full universe (54.8% vs 50.9%; year-clustered \(p = 0.004\)), and by +5.0pp on the survivor subset (\(p = 0.0009\)). The Gap’s advantage over the Margin of Safety is robust to window dependence, to survivorship, and to the switch from price returns to dividend-reinvested total returns.

6.2 Per-quadrant — the asymmetric structure

The framework’s predictive content is asymmetric — long-side vs short-side — and the asymmetry sharpens on the full point-in-time universe. Both short-side (negative-prediction) quadrants are now strongly significant: Value Trap 59.5% (\(p = 2\times10^{-4}\)) and Expensive Hype 58.5% (\(p = 2\times10^{-5}\)). Pooled, the Gap’s negative screen hits 58.9% (n=927, \(p = 3\times10^{-8}\)) — a robust ability to flag firms that subsequently underperform, materially stronger than on the survivor subset (55.9%, §6.1) precisely because the delisted firms it flagged largely did underperform. Both long-side (positive-prediction) quadrants sit at chance — Underestimated Growth 48.5%, Double Discount 47.9% — and the pooled positive screen is 48.2% (n=577), indistinguishable from (indeed marginally below) chance.

This is a substantively meaningful finding, not noise. The Brina Gap is, empirically, an overvaluation detector: it identifies firms whose own economics do not support their market valuations and predicts the resulting underperformance, with the signal concentrated entirely on the short side. The positive screen — predicting which underpriced firms will outperform — does not work on the broad universe; on the complete point-in-time S&P 500 it is at chance even in the Underestimated Growth cell that survived on the legacy panel. We state this plainly: the framework’s directional content is short-side. Short-side content is not a consolation prize: flagging which richly-priced firms will lag is exactly the half of the problem the quality toolkit (ROIC) cannot see, and no other valuation instrument we tested — the Margin of Safety included — delivers it.

The asymmetry has a clean theoretical interpretation. The framework’s positive predictions require two assumptions to play out: (a) the firm’s reinvestment ceiling holds going forward (ROIC stable), and (b) the market eventually closes the gap upward. The negative predictions require only that the market eventually recognizes a price incompatible with internal economics — a single, more reliably-met condition. Asymmetric assumption load produces asymmetric empirical strength.

6.3 Per-sector — utilities lead, technology lags

Computed on the full point-in-time directional sample (the 1,504 firm-years of §6.1, total-return basis); sectors with n \(\ge\) 30 shown.

Utilities lead at 72.2% (n=108), with Real Estate close behind (72.0%, n=50). The interpretation: regulated returns, predictable reinvestment programmes, and stable industry economics make the steady-state ROIC assumption hold tightly. The framework’s domain assumption is most true where it works best.

Technology sits below chance at 47.5% (n=200), and Materials lowest at 42.3%. The interpretation for technology: platform-effect non-linearities and AI-cycle rerating where the marginal dollar of capital earns more than the average (rising ROIC). The framework’s identity assumes the marginal dollar earns the same as the average; when it earns more, \(g_f\) systematically underestimates true sustainable growth, so Expensive Hype calls miss the rerating and Underestimated Growth firms outrun even their raised ceiling. This is Failure Mode 2 (§7.2).

The sector heterogeneity is itself a scope condition: the Brina Gap works best on sectors with stable ROIC. This is consistent with the theoretical derivation and bounds the framework’s applicability.

6.4 Mean realized excess return by quadrant

Computed on the full point-in-time directional sample (the 1,504 firm-years of §6.1, total-return basis).

On the full point-in-time universe every quadrant’s mean excess is negative — the directional constituents as a group underperform the cap-weighted SPY. This is the Bessembinder (2018) result: aggregate market returns concentrate in a small minority of mega-cap names (here the very names that dominate SPY’s weighting), so the median equal-weighted constituent lags the index. The means therefore do not separate the quadrants on level, and the sign-based hit-rate test of §6.2 — not the mean — is the appropriate test for this universe. Consistent with that test, the directional content is on the short side: the negative-prediction quadrants (Value Trap, Expensive Hype) sit among the most negative, while the positive-prediction quadrants fail to deliver positive means (Double Discount is in fact the most negative, −2.7). The means are reported for completeness.

6.5 Marquee cases — named-firm predictions

Beyond statistical significance, the framework should produce identifiable predictions on widely-followed firms. We list marquee calls among well-known S&P 500 names with resolved 5-year forward windows — predominantly large-gap (|gap| > 10pp) cases, together with a few instructive near-threshold examples (notably the Apple FY2011–2012 calls) that sharpen the framework’s stated scope. These are illustrative, not a complete or unbiased enumeration; the statistical claims rest on §6.1–6.4, not on this selection. All realized figures are 5-year annualized total-return excess vs SPY:

Correct calls (hits):

NVDA 2012 — the framework’s strongest individual-firm prediction — flagged the firm as Underestimated Growth before the data-center / AI cycle. The realized 5-year annualized total-return excess was +40.4 percentage points per year, the largest positive marquee result in the universe. PayPal FY2020 is the cleanest negative call: flagged Expensive Hype (market pricing growth its reinvestment economics could not finance), it went on to underperform SPY by 40 pp/yr as the post-pandemic multiple collapsed. Several short-side calls (KO, MRK, MA) are correct but marginal in magnitude — the prediction is directional, and on the broad cap-weighted universe even correct underperformers cluster near the index (§6.4).

Documented misses:

The misses are concentrated in compounders the framework knowingly under-credits. AAPL FY2011 (Value Trap) and NKE FY2012 (Expensive Hype) carried negative Gaps, but the realized windows covered Apple’s Services pivot and Nike’s pre-2015 run, in which both compounded faster than their steady-state \(g_f\) implied — the framework’s steady-state ROIC assumption did not anticipate the operational re-acceleration (Failure Mode 2, Quality Compounder Blind Spot, §7.2). UNH FY2020 (Double Discount, realized −16.7) is the inverse miss: a positive-Gap call that COVID-era managed-care disruption overwhelmed. NFLX FY2020 (Value Trap, +1.0) is a marginal directional miss — predicted underperformance, delivered a hair of outperformance. We retain the full set, hits and misses, for transparency.

We report these misses prominently. Concealing them would weaken the paper. Documenting them sharpens the framework’s stated scope.

6.6 Construct validity — \(g_f\) is a real growth ceiling

The directional tests of §6.1–6.5 ask whether the Gap predicts returns. A more fundamental question for a measurement is whether each arm corresponds to the real economic quantity it claims. We test the forward arm directly, on realized fundamentals rather than prices: if \(g_f = \text{ROIC} \times \text{RR}\) is genuinely the growth a business can self-finance, then realized forward operating-earnings growth should track \(g_f\) when ROIC is stable, and depart from it only through the escape channels the identity permits (a change in ROIC, external capital, or leverage). Using the panel’s own forward years, we compare each firm-year’s \(g_f\) at \(T\) to its realized 5-year NOPAT CAGR over \(T{\to}T{+}5\), conditioning on the realized change in ROIC (trimming \(|g_f|>100\%\) edge cases):

When ROIC is stable, realized growth equals \(g_f\) almost exactly (7.4% vs 7.1%; 7.3% vs 6.8% for the very-stable subset). Every deviation is explained by ROIC drift, monotonically and in the predicted direction: the fraction of firms exceeding their ceiling runs 29% → 58% → 88% from falling to stable to rising ROIC. Firms that improved ROIC broke through the ceiling (realized 23.1% vs \(g_f\) 8.0%); firms whose ROIC decayed fell short (1.0% vs 12.0%). The relationship holds on the complete point-in-time universe (n grows from the legacy panel’s ~1,000 to 1,847 forward-paired firm-years) — and tightens. (Untrimmed means give the same picture: \(g_f\) 7.8% vs realized 7.9% on n=1,119 stable firm-years — the figures quoted in Part I, from the published script.) This simultaneously validates the construct — \(g_f\), built mechanically from filings, is the realized sustainable growth rate of steady-state firms — and the scope condition: the steady-state identity holds where ROIC is stable and is breached precisely where it drifts (the Quality-Compounder mode of §7.2). We note honestly that this relationship is anchored in the accounting identity \(g = \text{ROIC} \times \text{RR}\); the empirical content is that the filings-derived inputs are faithful (the identity could have failed under measurement error, buybacks, or external financing) and that ROIC drift is the dominant wedge — not that \(g_f\) predicts growth from nothing.

6.7 Incremental validity — the Gap informs beyond the price multiple

The reverse arm \(g^*\) is mechanically a function of the valuation multiple, so a natural objection is that the Gap merely re-expresses cheapness: \(\text{Gap}>0\) firms have low \(g^*\), hence low multiples, which tend to mean-revert upward regardless. We rule this out by holding the market’s implied growth fixed (binning by \(g^*\) quartile) and asking whether the fundamental arm still separates outcomes — realized growth and the forward 5-year change in the EV/NOPAT multiple:

Holding the multiple fixed, \(\text{Gap}>0\) firms re-rate more (or de-rate less) than \(\text{Gap}<0\) firms in every \(g^*\) quartile — the fundamental arm carries information the starting multiple does not. The effect is largest among the cheap-\(g^*\) names (Q1: +61% vs +21%) and persists into the cleanest cell, Q4: among equally expensive firms (the market prices high growth for both), those whose economics can finance it de-rated only −3%, while those that cannot de-rated −12% — a ~9-percentage-point spread in the multiple that has nothing to do with the starting valuation. This is the Gap’s core economic claim, confirmed with the cheapness confound controlled: when a price assumes growth the business cannot finance, the resolution comes through the expectation coming down. Realized growth tells the same story in the lower quartiles; it reverses only in Q4, where high-expectation \(\text{Gap}<0\) firms grew marginally faster (+14% vs +13%) — the rising-ROIC compounders the framework knowingly under-credits (§7.2), surfacing again. Sections 6.6–6.7 establish that the Gap’s two arms each behave as the theory requires.

6.8 Convergent validity — \(g^*\) is a robust measure of priced-in growth

Section 6.7 leans on \(g^*\), the market-implied growth rate. Because \(g^*\) is recovered from one particular reverse discounted-cash-flow (two-stage, NOPAT-based, \(r=10\%\), \(g_T=3\%\)), a fair objection is that it could be an artifact of that convention rather than a measurement of what the market is actually pricing. The ideal external benchmark — analysts’ consensus long-term growth estimates — is point-in-time licensed data unavailable for the historical panel; we therefore test convergence across independent recovery methods. If genuinely different models that decode “the growth the price implies” agree on the cross-sectional ranking, \(g^*\) measures a real, model-robust quantity. We recover implied growth three further ways and rank-correlate (Spearman) each against \(g^*\) on the full universe (\(n = 2{,}042\)–\(3{,}398\) by method):

Three readings. The 1.00 with the Gordon inversion is mechanical — both \(g^*\) and the Gordon-implied growth are monotone transforms of the EV/NOPAT multiple, so they must rank firms identically; it confirms \(g^*\) is a well-posed implied-growth measure but is not independent evidence. The 0.76 with the FCF-based reverse-DCF is informative: switching the discounted flow from NOPAT to owner-earnings barely disturbs the ranking, so \(g^*\) is not an artifact of the cash-flow convention. Most telling is the 0.73 with the residual-income model — the only comparison drawing on genuinely different inputs (the invested-capital base and an explicit cost-of-capital charge, in the tradition of Ohlson 1995 and Penman 2010). On the full point-in-time universe the two agree strongly on the cross-sectional ordering, with modest residual divergence (and the convergence is tighter than on the legacy panel).

That divergence is interpretable, and arguably a design feature. Residual income prices growth net of the capital required to produce it; the NOPAT reverse-DCF that yields \(g^*\) deliberately reads only the operating-earnings stream the market capitalizes, not the quality of the reinvestment behind it (§2.4). The capital-intensity information separating the two is precisely what the forward arm \(g_f\) restores — so the Gap, \(g_f - g^*\), can be read as re-injecting into the price-implied growth the reinvestment-quality dimension a pure cash-flow inversion omits. We therefore treat convergent validity as supportive but moderate: \(g^*\) is robust to discounting conventions and positively, if imperfectly, convergent with an independent valuation model. It is the lighter of the paper’s validity pillars; the load is carried by construct validity (§6.6–6.7) and the discriminant evidence below.

6.9 Discriminant validity — the Gap is distinct from the standard metrics

For the Gap to be a contribution rather than a relabeling, it must not be redundant with the value and quality metrics already in use. Rank (Spearman) correlations of the Gap with each, on the full point-in-time valid universe (\(n \approx 3{,}400\)):

The full-universe rank correlations reproduce the legacy-panel finding almost exactly. The Gap’s only non-trivial association is a moderate monotonic tilt to the operating-earnings yield (\(\rho\) = 0.54), expected because the reverse arm \(g^*\) is built from the EV/NOPAT multiple — but the Gap is far from reducible to it. With everything else it is weakly correlated at most: quality (ROIC \(\rho\) = 0.21), free-cash-flow yield (0.24), book-to-market (0.15), and — notably — the Margin of Safety (0.24). It is not a quality factor, not a conventional value ratio, and not a restatement of price-versus-DCF. (We report rank rather than Pearson correlations because the complete universe contains a handful of extreme-ROIC firm-years that inflate the linear Gap–ROIC correlation to 0.64 while leaving the rank correlation at 0.21; the outlier-robust rank measure is the honest one, and it matches the panel.) Combined with the profitability-factor check of §8.5 — which shows the Gap is not subsumed by ROIC even within profitability terciles — the Gap is a genuinely distinct construct.

Taken together, §6.6–6.9 make the paper’s central case as a measurement: the forward arm is the realized growth ceiling of stable-ROIC firms (construct); the signal carries information beyond the price multiple about how mispricing resolves (incremental); the market-implied arm is robust across recovery methods (convergent); and the Gap as a whole is distinct from every standard metric, including the Margin of Safety (discriminant). The directional-accuracy results of §6.1–6.5 corroborate that this validated measurement is also systematically informative about future returns — they are the consequence of the construct’s validity, not its foundation.

6.10 Case studies — the Gap read firm by firm

The validity tests above are statistical; the framework’s value to a practitioner is that each reading is interpretable. Four cases — one per quadrant, plus an honest failure — show what the Gap sees and where it is blind. (Realized figures are split-adjusted five-year annualized excess returns vs SPY.)

NVIDIA, FY2012 — Underestimated Growth (the Gap sees what the DCF cannot). A graphics-chip maker earning 34.7% on capital and reinvesting half its earnings could finance ~17% growth (\(g_f\)); the price implied just 3.4% (\(g^*\)). An owner-earnings DCF, capitalizing only the modest current cash flow, called NVIDIA expensive (MoS −369%). The two lenses disagreed — and the Gap was right: NVIDIA outperformed SPY by +40.4pp/yr over the next five years as its reinvestment economics, not its trailing cash flow, drove the outcome. (Full arithmetic in Appendix A.)

AT&T, FY2019 — Value Trap (cheap by price, broken by economics). Here the disagreement runs the other way. The owner-earnings DCF saw a bargain — a +74% margin of safety — while the Gap saw a business whose reinvestment could not finance even flat growth (\(g_f = -3.5\%\), on a negative reinvestment rate as the firm shed capital) trading at a price assuming 6.9% (\(g^*\)). Price-versus-value said “cheap”; economics-versus-expectations said “the discount is deserved.” AT&T underperformed by −5.5pp/yr. Altria FY2016 is the same trap by a different route: a high-ROIC business (24%) that reinvests almost nothing (reinvestment rate near 1%), so its growth ceiling is essentially zero — wonderful economics, no runway — priced for 10% growth, MoS +64%, and a −16pp/yr outcome. ROIC alone would have called both stocks high-quality; the Gap caught what a quality screen and a DCF both missed.

PayPal, FY2020 — Expensive Hype (both lenses agree, and they were right). Post-pandemic, PayPal was priced for ~30% growth (\(g^* = 29.7\%\)) against a reinvestment ceiling of 9%. Both the Gap (−20.7) and the DCF (MoS −37%) flagged it expensive; the agreement was informative, and PayPal underperformed by −40pp/yr as the multiple collapsed toward what its economics could support.

Align Technology, FY2015 — the honest miss (Quality Compounder Blind Spot). The framework is not omniscient, and its failures are systematic in a way worth showing. Align earned 45% on capital and reinvested a quarter of it — \(g_f = 11.4\%\) — against a priced 14.8%, a mild negative Gap (−3.5) calling it slightly expensive. Align then outperformed by +32.8pp/yr: its ROIC did not hold steady but rose as Invisalign expanded into new markets, breaking the ceiling the steady-state identity assumes (§7.2). This is the error the Gap is designed to make — conservative on rising-ROIC compounders — and §6.6 quantifies exactly when it occurs.

These are illustrations, not the evidence (which is §6.1–6.9) — but they show the Gap is legible: every call decomposes into “what the business can finance” versus “what the price assumes,” and even its mistakes have a stated, testable cause.

7. Failure Modes and Scope Conditions

The framework runs on 3,402 valid firm-years. Beyond the six structural Boundary Conditions, the panel reveals four recurring patterns under which the framework runs but predicts poorly. Each is a sharpening of the framework’s domain, not a refutation.

7.1 Failure Mode 1 — M&A year distortion

Firms in the fiscal year of a large acquisition show inflated reinvestment rates that mechanically push \(g_f\) above what ongoing operations support. The framework “sees” a deal as if it were marginal organic reinvestment.

Resolution: BC4 + cited exclusion via the events database. 228 such exclusions are documented and citation-verified. On the full universe the cleaning removes 69 directional firm-years and leaves the pooled directional result essentially unchanged (54.8% cleaned).

7.2 Failure Mode 2 — Quality Compounder Blind Spot

Some firms (large-cap technology, healthcare innovators, platform businesses) sustain ROIC well above their reinvestment-implied steady state because they expand into new addressable markets. Their ROIC is rising, not stationary. The framework’s identity \(g_f = \text{ROIC} \times b\) assumes the marginal dollar earns the same return as the average; when the marginal dollar earns more, \(g_f\) underestimates true sustainable growth.

Empirical signature: Technology sector hit rate of 47.5% — below chance, and second-weakest of all sectors (only Materials, 42.3%, is lower). Apple 2011–2013, ServiceNow 2019, AMD 2019–2020 all exemplify this mode.

Theoretical implication: The framework is conservative on rising-ROIC firms — it under-predicts their growth. Buffett would view this as a feature: missing upside is preferable to incurring losses.

7.3 Failure Mode 3 — Aggressive growth investment regime

Firms reinvesting at extreme rates (b > 80%) at moderate ROIC produce \(g_f\) in a regime where the steady-state identity is no longer the binding constraint. Execution risk becomes the binding constraint, and the framework cannot price execution.

Empirical signature: Amazon-style “spend everything on growth” firms in their high-reinvestment decade. The framework reads such firms as Underestimated Growth, which may or may not pay off depending on execution.

Resolution: not a fixable failure within the framework; a scope condition. The Brina Gap should not be applied as a sole signal to firms with reinvestment rates approaching the maximum.

7.4 Failure Mode 4 — Borderline Gap anti-prediction

The subset of firms with \(|\text{Gap}| \in [0, 5\text{pp}]\) produces directional accuracy indistinguishable from coin flip. This is a power-of-the-signal effect: small gaps lie in the noise floor of input measurement.

Practical implication: the Brina Gap should not be applied to firms with \(|\text{Gap}| < 5\text{pp}\). The single-signal test in §4.3 includes these firm-years; excluding them lifts the standalone Gap hit rate from 52.4% (sign-only) to the ~54% range of the strong-signal subset (§4.5) — the entire gap between the marginal sign-only result and the significant strong-signal result is the noise floor being removed.

7.5 Aggregate scope condition

Combining the four failure modes and the per-sector heterogeneity (§6.3), the framework’s stated scope is:

The Brina Gap predicts directional excess return on US large-cap firms with stable ROIC, on fiscal years not contaminated by transformative M&A, on Gap magnitudes |Gap| > 5pp, in sectors where the steady-state identity holds (utilities and real estate ~72%; consumer staples, communication services, and industrials 55–58%). Application to financial services, healthcare, consumer discretionary, and energy is permitted but predictive content is weaker (50–54%). Application to technology and materials is permitted but predictive content is weakest (42–48%, at or below the 50% null).

(The |Gap| > 5pp scope is deliberately more conservative than the 3pp neutral band used operationally in the Matrix classification: §7.4 shows the [0, 5pp] region sits in the noise floor, so the scope recommendation trims an extra 2pp beyond the classifier’s band.)

This is a non-trivial scope condition. It is exactly what one expects of a theoretical framework that imposes a steady-state assumption: the framework works where the assumption holds.

8. Robustness

All §8 tables are computed on the full point-in-time universe (total-return, delisting-handled), consistent with §6.

8.1 Three pre-registered universe specifications

The pre-registration document specifies three nested universe definitions (top-100, top-200, and the full S&P 500 locked as primary in amendment v1.2), stratifying the cleaned universe by point-in-time within-year market-capitalization rank:

The pooled hit rate is consistent across universe sizes (54.8% to 56.9%), and is significant at well below the 1% level in both — indeed the top-100 stratum is stronger (56.9%), consistent with the framework working best on the larger, more stable-ROIC names. (A top-200 stratum is omitted because, on the directional panel, the qualifying firm-years are already essentially large-cap, so top-200 nearly coincides with the full panel.) The framework’s predictive content is not specification-dependent; the difference between specifications is driven by power, not by signal.

8.2 Discount-rate robustness

We re-solve \(g^*\) for every firm-year at \(r \in \{9\%, 10\%, 11\%\}\) (the base rate ±100bp), holding \(g_T = 3\%\), and recompute the pooled directional hit rate (the four Matrix quadrants):

The result is robust across the range: the hit rate rises mildly with \(r\), from 53.0% (p=0.01) at 9% to 55.0% at 11% — a 2.0pp spread, significant at the 5% level at every value tested. The mechanism is transparent: a higher discount rate lowers the market-implied growth \(g^*\) for a given EV, which widens positive Gaps and sharpens the sort. We fix the base case at the conventional round-number rate \(r = 10\%\) — the midpoint of the tested range, chosen on convention rather than to maximize the result. A full factor-controlled portfolio test is deferred to a follow-up.

8.3 Terminal-growth robustness

We re-solve \(g^*\) at \(g_T \in \{2.5\%, 3.0\%, 3.5\%\}\) (base ±50bp), holding \(r = 10\%\):

Terminal-growth sensitivity is mild: the hit rate stays within ±0.9pp of the base across the ±50bp range, and significance at the 1% level is preserved at every value. The Gap is comparably robust to terminal-growth and to discount-rate misspecification (§8.2).

8.4 Exclusion-regime robustness

We compare the raw universe (no exclusions) to the citation-backed cleaning: raw (54.5%, n=1,551, p=0.0002) and citation-backed (54.8%, n=1,504, p=0.0001). The convergence in the 54.5–54.8% band, both far below the 1% level, demonstrates the framework’s predictive content is not concentrated in the curated exclusions — the 228 cited removals shift the result by 0.3pp.

8.5 Is the Gap merely a profitability factor?

Because \(g_f = \text{ROIC} \times b\) rises with ROIC, and profitability earned strong returns over 2010–2024, one might worry the Gap operates entirely through the ROIC channel — that the Brina Gap is the profitability/quality factor (Novy-Marx 2013) in disguise. Three facts argue otherwise.

First, the Gap’s cross-sectional correlation with ROIC is only +0.19, versus −0.43 with the market-implied growth term \(g^*\) (trimmed, \(n \approx 3{,}140\)); the Gap’s variation is dominated by the market-expectation side, not the profitability side.

Second, stratifying the universe into ROIC terciles, the Gap retains directional content within terciles:

Third, within the high-ROIC tercile — where the confound would be most acute — the sign of the Gap separates five-year outcomes by 8.5 percentage points in P(excess return > 0) (Gap > 0: 56.0%; Gap < 0: 47.5%). Holding profitability roughly fixed, the Gap still discriminates.

The within-tercile power is not uniform — strong in low-ROIC, indistinguishable from chance in mid-ROIC, marginal in high-ROIC — and we do not claim this settles the question. A full orthogonalization against the Fama-French five factors and Novy-Marx gross profitability is deferred to future work (§9.5). The crude “Gap = profitability” account is, however, inconsistent with the within-tercile evidence: the Gap is not a monotonic function of ROIC, and it discriminates outcomes even when profitability is held approximately constant.

9. Discussion

9.1 What the framework is

The Brina Gap is a single, signed, falsifiable scalar that quantifies the difference between (a) the rate at which a firm can grow from its own retained earnings at current returns, and (b) the rate of growth the market is paying for. It is the natural quantitative form of what Graham and Buffett did intuitively when they preferred businesses whose own economics justified the price.

9.2 What the empirics show

On the full point-in-time S&P 500 (2010–2024, including acquired and delisted firms), with pre-registered universe and citation-backed exclusions:

1. The Margin of Safety, computed from owner-earnings DCF, predicts at 51.2% (p=0.12) on a sign-only basis — indistinguishable from a coin flip.
2. The Brina Gap, on a sign-only basis, predicts at 52.4% (p=0.013) — modestly significant even before any magnitude filter (weaker under window-clustered inference).
3. Restricted to firm-years with |Gap| > 3pp — equivalently, the four directional Matrix quadrants — the Brina Gap predicts at 54.8% (p=0.0001). A head-to-head bootstrap shows the Gap beats MoS by +3.9pp (year-clustered p=0.004); the MoS axis contributes a diagnostic decomposition rather than additional directional information (§4.5).
4. On the 929 firm-years where Gap and MoS disagree, Gap wins 51.3% to MoS’s 48.7% — a +2.7pp incremental advantage (+10.1pp on the strong-signal subset).
5. Including the delisted firms’ realized outcomes strengthens the short-side negative screen (55.9% on survivors → 58.9% on the full universe, p=\(3\times10^{-8}\)): the result is survivorship-clean.
6. The framework’s predictive content is concentrated on the short side: the negative screen hits 58.9% (p=\(3\times10^{-8}\); Value Trap 59.5%, Expensive Hype 58.5%), while the positive screen is at chance on the broad universe (48.2%, including the Underestimated Growth quadrant at 48.5%). The MoS axis is directionally uninformative — the Gap sign carries the signal (§2.5, §4.5).
7. Per-sector heterogeneity reveals the framework is strongest where its steady-state assumption holds (Utilities and Real Estate ~72%, Consumer Staples 58%) and weakest where the assumption is most violated (Technology 47%, Materials 42%).
8. Set against the companion audit’s sixteen screens on the identical universe, the Gap return-sorts first among all pure valuation metrics — above book-to-market, earnings yield, EPV/price, and the Margin of Safety (Part I, §8.1) — and is the only metric in the set whose central claim about the future can be verified against realized outcomes.

9.3 What the framework does not claim

The Brina Gap is not a complete asset-pricing model. It does not produce expected returns, volatilities, or position sizes. It produces a directional prediction: firms in particular quadrants are more likely than chance to outperform or underperform the market over 5 years. Translating that into a portfolio construction is a separate problem.

The Brina Gap is not a high-frequency signal. The 5-year forward window is essential: the framework’s mechanism is the eventual market recognition of the gap, which takes years to play out. The framework does not predict short-horizon returns.

The Brina Gap is not survivorship-bias-free in the strict sense: the S&P 500 itself is a selected universe (firms must be US-listed, large-cap, and meet liquidity/profitability requirements). Application to small-cap, international, or distressed equities is future work and explicitly not claimed here.

9.4 Implications for value investing

The empirical failure of single-axis MoS on this panel is, we believe, the paper’s most consequential finding for practitioners. It is not that value investing fails — Buffett’s track record alone falsifies any such claim — but that the operational form in which MoS is typically applied (price < DCF intrinsic value) is, on a broad equity universe, a noise signal. What works is the two-axis decomposition: firm economics and market expectations, considered jointly.

This is, we contend, the unspoken practice of every successful value investor. The Brina Gap names it.

9.5 Implications for academic finance

The Brina Gap is a firm-specific, fundamentals-derived signal that produces directionally significant excess returns versus a market benchmark. It is therefore a candidate factor in factor models, in spirit similar to Fama-French’s value and quality factors but mechanically distinct: it is not a static cross-sectional sort but a year-by-year scalar computed per firm.

Whether the Brina Gap survives orthogonalization against the standard factors (market, size, value, profitability, investment) is empirical work we do not perform in full here. The tercile evidence of §8.5 indicates the Gap is not subsumed by profitability alone, but a complete factor-model test — Fama-French five-factor and Novy-Marx (2013) gross profitability regressions on a Gap-sorted long-short portfolio, with returns rather than directional hit rates — is reserved for a subsequent paper. The framework is offered here as a refinement of the conceptual case for value investing, not as a factor-model contribution.

10. Limitations

• Universe: US large-cap S&P 500 only. Applicability to small-cap, international, and emerging markets is untested.
• Forward window: 5-year horizon by design (the framework’s mechanism is multi-year gap closure). Shorter horizons are not tested.
• Sample window: 2010–2024. Includes the post-GFC bull market, the COVID shock, and the 2022 rate-shock. Does not include the 1990s tech boom or 1970s stagflation; out-of-sample robustness in different macro regimes is future work.
• Owner-earnings DCF: our intrinsic-value methodology is one of several reasonable choices. A defender of a different MoS specification can substitute via the published pipeline.
• Survivorship: survivor bias inside the index is eliminated — the panel is point-in-time and retains acquired/delisted firms with their realized terminal outcomes (deal price or delisting return; §5.5, §6.1). The residual limitation is that the S&P 500 is itself a selected large-cap universe; results are not claimed for small-cap, international, or distressed equities.
• Pending forward windows: firm-years whose 5-year window extends past the data cutoff (recent-FY observations) are not yet resolved and are excluded from the directional tests; they mature into an out-of-sample panel on which the §12 pre-registered refinements can be tested. A residual set of pre-2016 acquired firms also lacks point-in-time prices from free data sources and enters the universe with fundamentals only (no Gap).

11. Conclusion

The Brina Gap is a single, signed, falsifiable scalar that reframes the price-versus-value question by decomposing it into two independently estimable terms: the firm’s fundamental growth ceiling (\(g_f = \text{ROIC} \times \text{RR}\)) and the market-implied growth rate from a reverse DCF (\(g^*\)). It is offered first as a theoretical construct — replacing the analyst-defined intrinsic value at the heart of the Margin of Safety with two externally verifiable quantities — and second as an empirical claim.

The empirical results, on the full S&P 500 panel with pre-registered universe and citation-backed exclusions, support the theoretical claim — first as a measurement, then as a signal:

The Gap measures a real economic quantity. Its fundamental arm \(g_f\) is the realized growth ceiling of stable-ROIC firms (realized 7.4%/yr vs predicted 7.1%), with deviations explained by ROIC drift; and, holding the market’s implied growth fixed, firms whose economics can finance the priced growth re-rate while those that cannot de-rate — a ~9pp multiple spread among equally-priced firms (§6.6–6.7). As corroboration, it also carries directional content — survivorship-clean, on the complete point-in-time S&P 500. Reduced to a single mechanical sign on the broad universe, the Margin of Safety does not sort returns (51.2%) — a property of that narrow test, not a verdict on the principle; the Gap supplies the reinvestment-economics axis it misses, reaching 54.8% on strong signals (p=0.0001), 59.5% on value traps, strengthening to a 58.9% short-side screen once delisted firms are included, and beating the Margin of Safety head-to-head where the two disagree (+3.9pp, p=0.004). The Brina Gap is not a refinement of the Margin of Safety but a complementary measurement that adds the axis price-versus-value cannot see.

The directional content is asymmetric — a strongly significant negative screen (58.9%; Value Trap 59.5%, Expensive Hype 58.5%) and a chance-level positive screen on the broad universe (48.2%, with no clean positive cell once the universe is complete). Per-sector heterogeneity reveals the framework operates best on firms with stable ROIC (Utilities and Real Estate ~72%) and weakest on firms with rising or volatile ROIC (Technology 47%, Materials 42%) — a scope condition exactly aligned with both the framework’s steady-state derivation and the construct-validity evidence of §6.6.

The Brina Gap is, we contend, the quantitative form of what value investors have always sought: a measure of mispricing grounded in a firm’s own economics rather than in the observer’s assumptions. Its validity rests not on beating the market in a backtest but on measuring something real — what a business can grow versus what its price assumes — verifiably, from public filings, for any firm.

Within the modern toolkit, the evidence assembled across this paper assigns the Gap a specific seat. Stock selection needs one validated instrument per axis: the quality axis belongs to ROIC, the strongest sorter in the companion audit. The pricing axis has been held for ninety years by book-to-market and the Margin of Safety — on seniority. By the same tests the canon is judged with, the Gap return-sorts first among the pure valuation metrics, reads the expensive side of the market more accurately than the Margin of Safety, and is the only valuation metric whose claim can be checked at all. The pricing slot belongs to the Gap.

12. Pre-Registered Refinements (Future Work)

The results reveal three structural patterns that suggest refinements to the framework. These cannot be tested on the present panel without HARKing (Hypothesizing After Results are Known); they are pre-registered here as falsifiable hypotheses, to be tested on (a) the recent-FY forward-return windows of this panel as they mature and/or (b) an out-of-sample international universe (e.g., S&P Europe 350, MSCI Japan large-cap).

The original pre-registration is dated 2026-05-25; the hypotheses below are re-anchored to the full point-in-time baseline. A future revision will report the pre-registered tests and their results, including failures.

12.1 Refinement A — The ROIC-Trajectory Matrix

Motivation. §4.5 demonstrates that the MoS axis of the v2.0 Matrix contributes no incremental directional information beyond the |Gap|>3pp confidence band. §2.5 attributes this to the value-trap selection bias documented in Lakonishok-Shleifer-Vishny (1994), Sloan (1996), and Penman (2007): MoS+ is not an independent confirmation of cheapness, but a contaminated signal. A genuinely independent second axis must measure something the Gap cannot see, and must not carry MoS’s bias.

Proposed second axis: ROIC trajectory, operationalized as the comparison between marginal ROIC (\(\Delta E_{t+1} / \Delta IC_t\)) and average ROIC (\(E_t / IC_t\)):

• Marginal ROIC > Average ROIC: expanding economics (the firm’s reinvestment is producing higher returns than its historical capital base implies)
• Marginal ROIC \(\le\) Average ROIC: contracting economics

This is mechanically computable from the same panel inputs as \(g_f\), requires no analyst assumption, and directly addresses Failure Mode 2 (Quality Compounder Blind Spot, §7.2).

Proposed refined Matrix:

Pre-registered prediction A: On the out-of-sample panel, the Compounding Bargain and Terminal Decline quadrants will jointly exceed a 60% directional hit rate against a 50% null (one-sided p < 0.01). The Justified Premium and Reinvestment Mirage quadrants are predicted to be non-significant at the 5% level (i.e., the framework correctly identifies these as abstain regions).

Status note (v3.0). An exploratory in-sample check of this refinement on the present panel (script v4_trajectory_test.py, published with the code) does not support it: the Compounding Bargain cell ran below chance in-sample (≈43%), and the joint CB+TD rate (≈54%) fell short of the pre-registered 60% bar. The decisive test remains the out-of-sample one registered above, but we report the unfavorable in-sample evidence now rather than wait: it lowers our prior on Refinement A. The registration stands so that a confirmed failure is on the record — that is what a commitment device is for.

12.2 Refinement B — Asymmetric Application Rule

Motivation. §6.2 reveals a long-side/short-side asymmetry: on the full point-in-time universe the broad-universe positive screen is at chance (48.2%, with Double Discount at 47.9% and Underestimated Growth at 48.5%), while the negative screen is strongly significant (58.9%, p=\(3\times10^{-8}\); Value Trap 59.5%, Expensive Hype 58.5%). The asymmetric assumption load identified in §6.2 — positive predictions require both ROIC stability and market recognition, negative require only correct identification of unsustainable implied growth — predicts this asymmetry. (The thresholds below are re-anchored to the v3.0 full-point-in-time baseline; the original pre-registration was calibrated to the earlier survivor-panel numbers.)

Pre-registered prediction B (re-anchored 2026-06-04 to corrected baseline): On the out-of-sample panel:

• Negative predictions (Gap < −3pp) on the broad universe will exceed 55% hit rate against a 50% null (one-sided p < 0.05).
• Positive predictions (Gap > +3pp) on the broad universe will not exceed 55% hit rate (the v3.0 long/short asymmetry will replicate).
• Positive predictions restricted to stable-ROIC sectors (Utilities, Real Estate, Consumer Staples, Healthcare — as identified in §6.3) will exceed 58% hit rate against a 50% null (one-sided p < 0.05).

12.3 Refinement C — Smoothed \(g_f\) Input

Motivation. Single-year NOPAT and IC are noisy due to M&A, working-capital fluctuations, and one-time items. The v2.0 framework addresses this via boundary conditions and the cited-events database, but residual single-year noise still distorts \(g_f\). Multi-year smoothing of ROIC reduces variance without introducing analyst input.

Proposed input refinement: Replace single-year ROIC in the \(g_f\) identity with a 3-year trailing average:

\[g_f^{\text{smooth}} \equiv \overline{\text{ROIC}}_{t-2..t} \times b_t\]

Pre-registered prediction C: On the out-of-sample panel, \(g_f^{\text{smooth}}\) will outperform single-year \(g_f\) in pooled directional hit rate by at least 1 percentage point.

12.4 Refinement D — Tiered Confidence Bands

Motivation. §7.4 identifies that |Gap| \(\in\) [0, 5pp] is below the framework’s signal-to-noise floor. The v2.0 framework uses a single neutral band at 3pp; a graded tier structure should improve signal selectivity by stratifying predictions by signal strength.

Proposed tier structure:

• |Gap| \(\in\) [3pp, 10pp]: Medium confidence
• |Gap| \(\in\) [10pp, 25pp]: High confidence
• |Gap| > 25pp: Suspended (likely measurement artifact from boundary-condition-edge cases)

Pre-registered prediction D: On the out-of-sample panel, the High-confidence tier will exceed 60% directional hit rate and outperform the Medium-confidence tier by at least 3 percentage points.

12.5 What this section is and is not

This section is a commitment device. By stating these refinements as pre-registered predictions with specific thresholds, we forfeit the option of later cherry-picking which refinement to publish based on out-of-sample outcomes. A future revision will report all four predictions A–D, including any that fail.

This section is not a claim about v2.0. The v2.0 results stand as published; nothing here modifies them. Refinements A–D are testable forward predictions, not retrospective adjustments. The companion pre_registration_v3.md in the published repository contains an identical statement of these hypotheses, dated and committed at v2.0 release.

Appendix A — Worked Example: NVIDIA FY2012

NVIDIA’s fiscal year 2012 ended January 29, 2012. At that point the company was a graphics-chip vendor pre-dating the data-center / AI cycle by approximately five years. We walk through the Brina Gap computation as it would have been visible to any reader of the 10-K filed in March 2012.

A.1 Inputs (from the FY2012 10-K, SEC EDGAR)

A.2 Computed quantities

A.3 Interpretation as of January 2012

The framework saw a firm with operating economics that could sustain ~17% earnings growth from internal reinvestment alone, trading at a price that — under standard discounting — required only 3.4% growth to justify. The gap was large (+13.6 pp), the direction was clear (the firm’s economics supported much more growth than the market was pricing), and the quadrant assignment was unambiguous (Underestimated Growth).

The framework’s a priori prediction was: positive 5-year excess return versus SPY.

A.4 Realized 5-year outcome (FY2012 close to FY2017 close)

NVIDIA outperformed SPY by 40.4 percentage points per year (dividend-reinvested total return) for the five years following the FY2012 fiscal close. The framework’s prediction was directionally correct, and the magnitude of outperformance is the largest single-firm hit in the universe.

A.5 What the framework did not predict

The framework predicted direction (positive excess return). It did not predict magnitude (40.4 pp/yr is far beyond what 13.6 pp of Gap suggests for a five-year window). The mechanism of the magnitude — the data-center and AI cycle that re-rated NVIDIA’s ROIC upward by an order of magnitude — is exactly Failure Mode 2 (Quality Compounder Blind Spot, §7.2): the framework’s steady-state assumption under-predicts firms whose ROIC subsequently rises. NVIDIA’s \(g_f\) of 17% was a ceiling on the steady-state; the firm broke through the ceiling by acquiring a new addressable market (data center). The framework was conservatively right in direction, missed the magnitude, and the miss is in the favorable direction.

This is, we contend, the most useful kind of failure to have in a value-investing framework.

Appendix B — Replication

All code, the 8,820-row dataset, the 228-event corporate-events database, the manual review queue, and per-quadrant bootstrap CIs are published at the companion repository:

• pre_registration.md — pre-registration with two dated amendments
• outputs/file_01_full_dataset_rebuilt.csv — the full panel (v3.0 canonical)
• outputs/ticker_events_database.csv — citation-backed exclusion log
• outputs/statistical_results.md — auto-generated results file
• outputs/file_05_replication_guide.md — methodology details
• src/ — pipeline implementation (Python 3.12)

Any reader can reproduce every number in this paper from raw SEC filings using the published pipeline. The user-agent header for SEC EDGAR API access must comply with SEC’s terms of use.

Appendix C — Notation

Plain-language FAQ

What is the Brina Gap? The Brina Gap is a valuation metric introduced by Fabio Brina. It is defined as the difference between the growth a company can finance from its own profits (\(g_f\) = ROIC × reinvestment rate) and the growth its market price already implies (\(g^*\), recovered from a reverse DCF on enterprise value). A negative Brina Gap means the price demands more growth than the firm’s current economics fund; a positive Gap means the market is pricing in less than the firm could self-finance. Both inputs come from public filings, so the Gap is fully mechanical and reproducible. In the 2010–2024 audit it is the strongest pure valuation metric tested — ranking above book-to-market, earnings yield, EPV, and the Margin of Safety on the identical sample (Part I, §8.1) — and the only one whose central claim can be verified against what firms subsequently deliver.

What growth is the market pricing into a stock? That is exactly the quantity the Gap’s pricing arm \(g^*\) recovers — and this paper shows it is real: market-implied growth tracks the growth firms subsequently deliver (correlation ≈ 0.4–0.5) and is nearly unbiased over a 10-year horizon. The market is a good growth forecaster on average, with one systematic flaw: it over-extrapolates — the highest-expectation stocks deliver less growth than their prices imply.

Is the Brina Gap better than the Margin of Safety? They address the same question — what is the price assuming? — but the Brina Gap is mechanical and falsifiable from filings, while the Margin of Safety depends on the analyst’s own intrinsic-value estimate. Empirically the two are nearly orthogonal (they share less than 6% of rank variance), the Gap is more accurate on the directional test both can take (≈ +4pp), and on the test neither passes — picking winners among similar-quality firms — every pricing metric fails alike. See Part I, §4.6.

Is the Brina Gap a buy or sell signal? No — and no single metric is (companion audit). The Gap is the pricing half of a two-part judgment. Its validated uses are reading the growth expectation embedded in a price and flagging over-extrapolated expectations; acting on it requires the fundamentals half — see the Brina Matrix (Part I, §5).

Why doesn’t the Brina Gap predict returns on its own? Because theory says it shouldn’t — and that the best metrics don’t. The Gap is computed entirely from public filings; if markets price public arithmetic at all, a transparent metric cannot carry a standalone return edge (Part I, §2.2). Empirically the Gap return-sorts better than most famous screens — gross profitability, EPV, ROE, the Piotroski score, the Margin of Safety, and the Fama-French value factor — on the identical sample (Part I, §8.1). What it uniquely adds is not prediction but measurement: it tells you, verifiably, what growth expectation you are being asked to pay for.

What is the Brina Matrix? A two-axis diagnostic that crosses fundamental quality with the Brina Gap’s pricing read, sorting any stock into four quadrants: Underpriced Quality, Priced for Perfection, Value Trap, and Speculative/Overpriced. Its one robust rule, confirmed under every specification tested: cheap is only a bargain when quality is high — cheap plus low quality is the value trap, the worst-performing cell (≈ −4%/yr vs the market over 2010–2024).

How do you avoid a value trap? Check quality before acting on cheapness. In our data, low-quality stocks that looked cheap (positive Gap) were the worst performers — they were cheap because they deserved to be. The AT&T FY2018 case study (Part I, §6) is the textbook example: a household name, a pessimistic price, weak economics — and −13%/yr versus the market over the next five years.

Declarations

Data and code availability. The full panel dataset (8,820 ticker-years), the corporate-events database, the pre-registration documents, and the complete replication pipeline are published as companion materials (Appendix B; Zenodo and the project repository). Every number in this paper is reproducible from public SEC filings using the published code.

Use of AI tools. The author developed the Brina Gap framework, designed the research methodology, collected and analyzed all data, and drew all conclusions independently. Claude (Anthropic) was used solely as a writing assistant to improve the readability and formatting of the prose.

Competing interests. The author is the founder of Zyberno.com, a venture that applies the Brina Gap framework, and therefore has a professional and prospective financial interest in the framework — though he presently draws no income from the venture and funds its operation himself. This is disclosed in the interest of full transparency. The study is deliberately designed to be resistant to author bias: the universe and methodology were pre-registered before any results were computed (with dated amendments); every firm-year exclusion is documented with a verifiable citation; the framework’s failure modes, weakest sectors, and discount-rate sensitivity are reported in full rather than suppressed; and the complete dataset and replication pipeline are published so that any reader can independently reproduce — or refute — every result in this paper.

Funding. This research received no external funding.

References

Bessembinder, H. (2018). Do stocks outperform Treasury bills? Journal of Financial Economics 129(3), 440–461.

Buffett, W. (1986). Letter to shareholders of Berkshire Hathaway Inc. (introducing the owner-earnings concept).

Damodaran, A. (2002, and continuing editions). Investment Valuation: Tools and Techniques for Determining the Value of Any Asset. Wiley.

Fama, E. F., and French, K. R. (1993). Common risk factors in the returns on stocks and bonds. Journal of Financial Economics 33(1), 3–56.

Graham, B., and Dodd, D. L. (1934). Security Analysis. McGraw-Hill.

Graham, B. (1949). The Intelligent Investor. Harper & Brothers.

Greenblatt, J. (2005). The Little Book That Beats the Market. Wiley.

Greenwald, B., Kahn, J., Sonkin, P., and van Biema, M. (2004). Value Investing: From Graham to Buffett and Beyond. Wiley.

Koller, T., Goedhart, M., and Wessels, D. (2020). Valuation: Measuring and Managing the Value of Companies (7th ed.). McKinsey & Company / Wiley.

Lakonishok, J., Shleifer, A., and Vishny, R. W. (1994). Contrarian investment, extrapolation, and risk. Journal of Finance 49(5), 1541–1578.

Mauboussin, M. J., and Rappaport, A. (2001). Expectations Investing: Reading Stock Prices for Better Returns. Harvard Business School Press.

Novy-Marx, R. (2013). The other side of value: The gross profitability premium. Journal of Financial Economics 108(1), 1–28.

Ohlson, J. A. (1995). Earnings, book values, and dividends in equity valuation. Contemporary Accounting Research 11(2), 661–687.

Penman, S. H. (2007). Financial Statement Analysis and Security Valuation. McGraw-Hill (3rd ed.).

Penman, S. H. (2010). Accounting for Value. Columbia University Press.

Ross, S. A., Westerfield, R. W., and Jaffe, J. F. (various editions). Corporate Finance. McGraw-Hill. (See Chapter 5 on the sustainable growth identity.)

Sloan, R. G. (1996). Do stock prices fully reflect information in accruals and cash flows about future earnings? The Accounting Review 71(3), 289–315.

Bordalo, P., Gennaioli, N., La Porta, R., and Shleifer, A. (2019). Diagnostic expectations and stock returns. Journal of Finance 74(6), 2839–2874.

Brina, F. (2026). Which Stock Screens Actually Work? A Survivorship-Free Audit of 16 Fundamental Metrics — ROIC, Value, Momentum, and Quality — on the S&P 500 (2010–2024). Companion working paper. Zenodo: 10.5281/zenodo.20650537.

Claus, J., and Thomas, J. (2001). Equity premia as low as three percent? Evidence from analysts’ earnings forecasts. Journal of Finance 56(5), 1629–1666.

Easton, P. (2004). PE ratios, PEG ratios, and estimating the implied expected rate of return on equity capital. The Accounting Review 79(1), 73–95.

Fama, E. F., and MacBeth, J. D. (1973). Risk, return, and equilibrium: Empirical tests. Journal of Political Economy 81(3), 607–636.

Gebhardt, W., Lee, C., and Swaminathan, B. (2001). Toward an implied cost of capital. Journal of Accounting Research 39(1), 135–176.

Higgins, R. C. (1977). How much growth can a firm afford? Financial Management 6(3), 7–16.

La Porta, R. (1996). Expectations and the cross-section of stock returns. Journal of Finance 51(5), 1715–1742.

Lockwood, L., and Prombutr, W. (2010). Sustainable growth and stock returns. Journal of Financial Research 33(4), 519–538.

Miller, M. H., and Modigliani, F. (1961). Dividend policy, growth, and the valuation of shares. Journal of Business 34(4), 411–433.

Working Paper v3.0 — comments to fabio@fabiobrina.com. Data, corporate-events database, figures, and the complete replication pipeline are published with this paper (Zenodo and the Zyberno Research repository). A plain-language companion is maintained at Zyberno.com.