ValueScreener

Lesson 4 of 8

The retro-test and the eight

Eight figures can be rebuilt from an old filing and tested backwards. The credit ratios are mostly not among them, and what to do about that.

Javier Sanz, founder

By Javier Sanz

Founder of ValueScreener. Built Ninety Nine, a securities services provider, and sold it to a tier-one global bank; ran operations and internationalization at Alpaca, the brokerage API.

Who writes thisEditorial processLinkedIn (opens in a new tab)X (opens in a new tab)

In one minute

  • Eight figures can be rebuilt from old filings on the same formula the live value uses, so a condition on them can be tested backwards before it is trusted.
  • Most of the credit-specific figures are not among the eight, so a condition on one of them can only be checked forward.
  • A condition you can only check forward is still a condition. It is one where the answer arrives by waiting.

Before relying on a condition, it is worth knowing whether it would ever have fired. For eight figures that question has an answer, and for the rest it does not.

Which conditions can be tested against filed history

Testable backwards

  • Current ratio, debt to equity.
  • Gross margin, operating margin, net margin.
  • Return on equity.
  • Free cash flow.
  • One-year revenue growth.

Forward only

No backward test

  • Interest coverage, net debt to earnings, operating cash flow to debt.
  • Any maturity or covenant condition.
  • Return on invested capital, the composite scores.

The left column can be checked before it is trusted. The right column can only be checked by waiting.

Illustrative round numbers, not any company's filings.

What the strict test is

A figure passes if it can be rebuilt from a filing several years old, using the exact formula the live figure uses today. Not a similar formula and not an approximation: the same one, on inputs the old filing carries in the same way.

Eight figures pass in ValueScreener: current ratio, debt to equity, free cash flow, gross margin, net margin, operating margin, one-year revenue growth and return on equity. Return on invested capital does not, and neither do the composite scores.

Why the credit figures mostly fail it

Because their inputs are less consistently available across filed years than the eight. An interest expense that is zero in some years and positive in others, a balance sheet with no current split, an earnings measure not defined by any standard: each of those breaks the reconstruction in a different way.

Two of the eight are credit figures, which is worth noticing. The current ratio and debt to equity both pass, so a condition on either can be tested backwards at a filer that reports the inputs. The ones that fail are precisely the ones this track found most useful, which is an awkward result and the honest one.

Two mistakes a backward test does not protect against

A standard changing under the series. The lease standard moved operating leases onto balance sheets for fiscal years beginning after 15 December 2018, so every ratio with assets or liabilities in it changed basis on that date without any business changing. A backward test across that transition tests two different things and reports one number.

And overfitting the level. A reader who tries several levels and keeps the one that fires at the interesting moments has fitted the level to the history, which is what the screens and factors course spends a lesson on. The test tells you whether a level would have fired; it does not tell you whether it should have.

What to do with a forward-only condition

Write it, and write beside it that it has no backward test. That single note is what stops a reader in three years treating a rail that has never fired as evidence that nothing has happened.

A condition you can only check forward is still a condition, and the answer arrives by waiting.

Check yourself

What is the strict test a figure has to pass to be retro-testable?

  1. That it appears in every annual report
  2. That it can be rebuilt from an old filing using the exact formula the live figure uses today
  3. That it is audited

Show the answer

Answer: B.

Not a similar formula and not an approximation. Eight figures pass; return on invested capital and the composite scores do not.

A condition on free cash flow would never have fired across seven filed years. What does that tell you?

  1. That the condition is well chosen
  2. That it is measuring something that does not vary here, or is set where the company has never been, and should be changed before it is relied on
  3. That the company is financially sound

Show the answer

Answer: B.

Unwelcome information delivered cheaply is what the test is for. The alternative is finding out in five years, reassured by silence throughout.

What does a backward test not protect against?

  1. A figure with missing data
  2. A standard changing under the series, and fitting the level to the history by trying several
  3. A condition written in the wrong direction

Show the answer

Answer: B.

A test across the lease transition tests two different things and reports one number, and a level chosen because it fired at the interesting moments has been fitted rather than derived.

Sources

  • Toll Brothers, Inc., annual reports on Form 10-K for fiscal years 2019 through 2025, for operating cash flow and capital expenditure; fiscal 2025 accession 0000794170-25-000112, filed 19 December 2025.
  • ASC 842, Leases, effective for public business entities in fiscal years beginning after 15 December 2018.

What this leaves out

  • This is a heterodox tradition. The mainstream account of the same phenomenon differs, and nothing here is a forecast or a view on any security.
  • The eight retro-testable figures are the eight in this product. A reader working from filings by hand can reconstruct anything they are willing to compute.
  • Seven fiscal years is the window checked above. A longer history at the same company could contain a year in which the condition fired, and the test reports only the window it covers.
  • Nothing here is a view on Toll Brothers or on any security.

A condition shown, not set

Lesson 4 of 8 of Guardrails that follow a reading. The condition here is written out as an example. Setting one on a company you hold, and having it checked nightly, happens in the product, not here.

Free audit

Research tool, not investment advice. See our methodology for how the product reads a filing and what it refuses to compute.

Once a week, what changed in the filings of US-listed companies, why it can matter to a thesis, and what deserves a closer review. Grounded in filings, not market noise.