What does the Capital reverse-DCF & ROIC engine predict?

It answers: What growth, exit multiple and return on capital does today's price already require, and is that more or less than the business has actually delivered? Output: Req g, Exit ×, Perp g, Hurdle, ROIC−WACC and one verdict chip per row. Horizon: 7-year explicit period plus a perpetuity; no forward-return horizon is claimed. Coverage: 10,631 listed names above each market's own USD cap floor; Financials and Real Estate masked.

How was the Capital reverse-DCF & ROIC engine tested?

Training data not stated, test data not stated, out-of-sample design: none. Round-trip forward↔reverse assertion: 4.2e-15 median error, 3,158 rows; Req g rank stability under β noise: Spearman 0.979; Forward-return validation: none — never tested.

What growth and return on capital does a stock's price already require?

Capital reverse-DCF & ROIC engine (M-002). Short answer: the hub rates it FIT FOR LIMITED USE.

Data as of 6 Nov 2026.

Open M-002 in the live Model Hub All 30 models and verdicts

Verdict: Capital reverse-DCF & ROIC engine (M-002) is rated “FIT FOR LIMITED USE — RELIABLE WITH CONDITIONS” in its validation review dated 1 Sep 2026.
Key result: Round-trip forward↔reverse assertion: 4.2e-15 median error, 3,158 rows.
Key result: Req g rank stability under β noise: Spearman 0.979.
How it was tested: the record does not state separate training and test periods; out-of-sample design: none.
Horizon: 7-year explicit period plus a perpetuity; no forward-return horizon is claimed. Coverage: 10,631 listed names above each market's own USD cap floor; Financials and Real Estate masked.
Stage: Production. Version v3; the hub record is as of 6 Nov 2026.

The verdict

FIT FOR LIMITED USE — the ranking replicates and the arithmetic is verified; the levels, the sign of the headline read and any claim of predictive power are not supported. No forward-return evidence exists for any column on this face.

Development summary: Conceptually sound and arithmetically verified, and nothing on this face has ever been tested against forward returns. It is built to be ranked and argued with, not quoted: a 1pp error in WACC moves Req g by 7.7pp.

What the review allows

Rank-ordering names within a sector on Req g / Hurdle / Exit × — β-noise Spearman 0.979-0.998.
Reading what a price already assumes, as the prompt for a research argument in the 70% value sleeve.
ROIC − WACC as a first gate: below zero, growth subtracts value and the price columns answer a question that does not matter yet.

What the review rules out

Quoting Req g or Hurdle to a decimal — the honest band is roughly ±8-11pp from beta estimation error alone.
Reading the SIGN of Hurdle as a fact: 5.2% of rows flip on beta noise before any modelling error.
Treating a verdict chip as a forward-return claim — not one of the five has a measured base rate.
Applying it to Financials or Real Estate (masked) or to the 51% of names with negative free cash flow.
Automatic sizing, entry or exit; and any use inside an options screen, where the horizon does not exist.

What it predicts

Item	Detail
Question it answers	What growth, exit multiple and return on capital does today's price already require, and is that more or less than the business has actually delivered?
Method	Reverse DCF (7y fade)
How it works	Reverse DCF — 7y fade path, mid-year discounting, self-funding terminal; bisection inverts price into the assumption that justifies it. ROIC−WACC beside it.
Inputs	Enterprise value & market cap, TTM free cash flow & EBITDA, 5y beta (cost of equity), Debt & cost of debt, 5y revenue CAGR (Hurdle base)
Output	Req g, Exit ×, Perp g, Hurdle, ROIC−WACC and one verdict chip per row
Horizon	7-year explicit period plus a perpetuity; no forward-return horizon is claimed
Coverage	10,631 listed names above each market's own USD cap floor; Financials and Real Estate masked
Markets	US, India, Japan, Korea, Hong Kong, Taiwan, Canada, Singapore, UK, Germany, France, Switzerland, Netherlands, Italy, Spain, Sweden, Denmark, Norway, Poland, Finland
Model type	Model · Structural
Stage	Production
Version	v3, first recorded 22 Aug 2026

Validation evidence

Training data: not stated. Test data: not stated. Out-of-sample design: none.

Headline results from the model's development record, as the hub publishes them.
Measure	Result	Compared with	Note
Round-trip forward↔reverse assertion	4.2e-15	0 (exact)	median error, 3,158 rows; both directions call one _fade_path so they cannot drift
Req g rank stability, β ~ N(0, 0.20)	Spearman 0.979	1.000 = perfect	36.5% of rows change decile; top-decile survival 90%
Req g level sensitivity to WACC	+7.73pp per +1pp	±1pp of CoE from β s.e. alone	90th percentile +11.89pp — the levels are not quotable
Terminal-reinvestment correction (v2)	median Req g 1.6% → 16.4%	the 2026-08-22 build	281 of 918 negative hurdles flipped positive; 76 rows re-chipped Burns capital
Hurdle base: 5y revenue CAGR vs TTM	Spearman 0.222 vs 0.178	TTM, the prior base	9,100 matched observations, 1,979 names, forward 5y realised CAGR; MAE 0.102 vs 0.142
Rows with both Exit × and Req g present and unflagged	13% (469 / 3,509)	100% of rows rendered	US tape, 2026-08-23 — the real size of the trustworthy universe

The hub's standard metrics for this model class (only those the record measured).
Metric	Sample / window	Compared with	Value	Range (CI)
Round-trip error	nightly · tape 2026-09-01/02	0 (exact)	4.2e-15	n/a
Observations	round-trip rows · reviewed build	n/a	3,158	n/a

What the validation review found

Is the design sound? The reverse-first design is right for a 10k-name tape and the file's own docstring says why. Mid-year terminal discounting re-derived independently and confirmed correct. The v2 correction — the terminal must fund its own growth — removed the one real economic error (at 12% ROIC the old terminal cash flow was overstated 50%; at 5% ROIC, 339%), and the bias had been largest exactly on the capital-hungry names the face flags as cheap. Remaining conceptual gap: one equation in two unknowns (growth vs discount rate), with Req g's Spearman against WACC at +0.573.

Data quality. 2026-09-01 full run: 1 RED · 6 AMBER · 12 OK across 26 datasets. The RED was the known SEC-family blackout (five US columns 100% null against a 40% baseline) caused by a NameError, not a data problem; the 2026-09-02 file shows sbc_rev 46.0%, fcf_sbc_yield 67.5% and debt_raised_pct 40.3% restored. Scanner throttles return HTTP 200 with nulls — guarded upstream by MIN_SCAN_ROWS = 3000 and MIN_KEEP_FRAC = 0.5. 405 rows with no TTM revenue growth are handed 5% with no flag. Universe is survivorship-biased (today's listings).

Backtest. None exists, and this is the single largest gap. valuation history began 2026-08-22, so no column, percentile or chip has ever been scored against forward returns. The closest thing to a base rate is the persistence study: over four years, 9.6% of US filers sustained >25% revenue growth and 20.5% >15%; deceleration is near-universal (93.8% of the 25-50% growth bucket decelerated). That supports the fade convention and imposes a harsh prior on any positive Hurdle, but it is not a test of the model's output.

Against simple benchmarks. Exit × correlates +0.894 with today's EV/EBITDA across the tape, so the plain multiple is a fair benchmark for how well the reverse DCF ranks — and the reverse DCF is not measurably better at ranking. Its independent content is Re-rate and Hurdle. Separately, the 8-year cross-sectional study of alternative valuation models found the reverse DCF earns its place on Industrials, Consumer Discretionary, Consumer Staples, Communication Services and Energy, and should be replaced on Financials, Real Estate, Utilities and Health Care — which is what M-003 does.

Independent replication. Every number in REVERSE_DCF_EXPLAINED was re-derived from raw inputs for two named companies and matched the shipped valuation. The round-trip assertion is a continuous replication test and passes at 4.2e-15. The mid-year exponent was re-derived from scratch; the '5% error against the benchmark' recorded in the source comments was the benchmark being wrong.

Sensitivity. WACC +1pp → Req g +7.73pp (p90 +11.89pp); worked examples at +3pp on five large caps. β ~ N(0, 0.20) → rank correlations Exit × 0.998, FDep 1.000, Perp g 0.985, Hurdle 0.980, Req g 0.979; decile churn up to 36.5%; Hurdle sign flips on 5.2% of rows. Beta window: 5y vs 1y moves 8% of verdicts. FADE_CAP relaxation 25%→50% moves Exit × −34.5% on the 18.9% of rows where it binds.

Monitoring. Present: the nightly round-trip assertion, qa_checks C3 (null-rate drift vs a frozen baseline), C7 (percentile integrity) and C8 (DCF identities: wacc ∈ [0.05,0.20], implied_x > 0, req_cagr inside its fade band). Absent: any outcome loop, any input-drift PSI, and any champion/challenger — all three are unavailable until the history file matures. is not yet wired into.

Limitations

No forward-return evidence exists for any column, chip or percentile on this face. Everything below is about internal correctness, not predictive power.
Levels are WACC-leveraged about 7.7:1 (Req g) and 1:1 (Perp g). Beta standard error alone puts a ±8-11pp band around a printed Req g.
The SIGN of Hurdle — the headline read — flips on 5.2% of rows from beta noise before any modelling error is counted.
Only 13% of US rows carry both Exit × and Req g present and unflagged; 48% carry Req g at all, and terminal_bet is the most common chip (825 of 1,733 chipped US rows) because FDep is below 30% on most of the tape.
Financials and Real Estate are masked entirely (0% coverage), and Utilities and Health Care were measured at 24.7% and 33.4% Exit × coverage — the surviving quarter of Utilities is the quarter not currently building anything, which is a systematic bias.
Exit × correlates +0.894 with today's EV/EBITDA, so as a RANKING it is largely a re-expression of the multiple; the independent content sits in Re-rate and Hurdle.

Questions

Is the Capital reverse-DCF & ROIC engine validated?: Its validation review dated 1 Sep 2026 rated it FIT FOR LIMITED USE — the ranking replicates and the arithmetic is verified; the levels, the sign of the headline read and any claim of predictive power are not supported. No forward-return evidence exists for any column on this face.
What does the Capital reverse-DCF & ROIC engine predict?: It answers: What growth, exit multiple and return on capital does today's price already require, and is that more or less than the business has actually delivered? Output: Req g, Exit ×, Perp g, Hurdle, ROIC−WACC and one verdict chip per row. Horizon: 7-year explicit period plus a perpetuity; no forward-return horizon is claimed. Coverage: 10,631 listed names above each market's own USD cap floor; Financials and Real Estate masked.
How was the Capital reverse-DCF & ROIC engine tested?: Training data not stated, test data not stated, out-of-sample design: none. Round-trip forward↔reverse assertion: 4.2e-15 median error, 3,158 rows; Req g rank stability under β noise: Spearman 0.979; Forward-return validation: none — never tested.

Source: the Model Hub's public model register, as of 6 Nov 2026. The M-002 page in the live Model Hub carries the full development and validation record, monitoring and open decisions. All 30 models and their verdicts.

Terms used on this page

Out-of-sample: results measured on data the model never saw while it was built or tuned.
Base rate: how often the outcome happens with no model at all.
Percentage points (pp): the plain difference between two percentages: 12% against 10% is 2 pp.
Standard error (s.e.): the typical size of the error in an estimate; a result more than 2 standard errors from zero is unlikely to be luck.
Confidence interval: the range that likely contains the true value (95% unless stated); [a, b] on this page is that range.
Survivorship: a sample that leaves out companies that later delisted, which flatters historical results.
PSI: population stability index: how far today's inputs have drifted from those the model was built on.
MAE: mean absolute error: the average size of a forecast miss.