How Accurate Are Net Worth Estimates? A Source Audit
Most published celebrity wealth figures cite nothing. The reliable ones trace to a filing with a date, a filer and a legal consequence for lying: a Form 4, a bankruptcy schedule, a divorce exhibit, a set of Companies House accounts. Everything else is either a disclosed-methodology estimate or repetition that has been mistaken for corroboration.
Mostly, no. A wealth figure is only as good as the document underneath it, and for the overwhelming majority of named individuals there is no document at all: there is a number that appeared somewhere around 2011 and has been copied since.
The figures that are good are very good, because they come from filings made under legal compulsion by people with liability for getting them wrong. A Form 4 filed two business days after an insider trade tells you exactly how many shares changed hands and at what price. A bankruptcy schedule signed under penalty of perjury lists assets and liabilities line by line. A divorce exhibit produced under discovery has been through opposing counsel.
This page grades the difference. What each evidence class proves, how far behind reality it runs, which direction it distorts, and what weight it should carry. That grading is published here as the Paper Trail Grade rubric, and it is the standard every figure on this site has to meet before it appears.
Where a figure can actually come from
Ten evidence classes cover essentially everything. The lag column matters as much as the verifiability column, because a perfectly attested figure from four years ago is a historical fact rather than a current one.
Evidence class What it proves Statutory deadline or typical lag Directional bias Weight Grade SEC Form 4 and Schedules 13D and 13G Exact share counts, transaction prices and beneficial ownership for insiders and 5% holders Form 4 within two business days of the transaction. Schedule 13D within five business days of crossing the threshold since the 2023 amendments took effect None. Both over- and under-statement carry liability Highest for the specific holding, silent on everything else A Bankruptcy petition schedules and Statement of Financial Affairs A complete claimed asset and liability picture, signed under penalty of perjury Schedules due within 14 days of the petition under the bankruptcy rules Low. Debtors have an incentive to minimise assets, and a trustee to catch them Very high, but only as at the petition date A Court and divorce financial disclosures Sworn income and asset statements, sometimes with returns attached as exhibits Filed at the litigation stage, frequently sealed and released only in part Low. Both parties have an incentive to understate their own position Very high where the exhibit itself is available, not where only a reported summary is A SEC 10-K and DEF 14A proxy statements Named executive officer compensation, beneficial ownership tables and related-party payments 10-K due 60, 75 or 90 days after fiscal year end depending on filer status. Proxy filed annually None on the disclosed items. Silent on anything outside the registrant High for the covered person, zero for anyone not an officer or director B Companies House accounts, confirmation statements and the PSC register Net assets, shareholdings and persons with significant control at 25% thresholds for UK companies Private company accounts due nine months after the accounting reference date. Reforms phasing in expanded filing requirements Downward on small-company accounts, which historically omitted the profit and loss account High for the company, indicative for the individual B Land Registry titles and US county recorder deeds Ownership, transfer date and consideration paid, plus registered charges Days to weeks after completion. England and Wales title registers show price paid on sales from April 2000 None on price paid. Silent on current value and on the mortgage balance today High for the asset, misleading if treated as equity B Union scale minimums and league collective agreements The floor and the mechanism: what the contract type must pay and how the money is structured Published on execution and superseded on renegotiation. Term dates are the whole story Downward. Scale is a minimum, and established performers work above it High for the mechanism, zero for the individual amount C Published estimate with a disclosed methodology That a defined method was applied. Bloomberg publishes a per-person methodology note; Forbes describes liquidity discounts applied to private holdings Annual or rolling, with inputs not disclosed Varies by publisher. Private-asset valuation is the discretionary step Moderate. Usable if attributed as an estimate with the method named D Statement by a publicist, agent or the subject That an interested party said a thing on a date Immediate, and never withdrawn when it stops being true Strongly upward for a rising career, strongly downward during litigation Low. Record the claim and the claimant, not the number E Aggregator repetition That several sites carry the same figure The figure ages indefinitely. Ten-year-old numbers are routine Upward and sticky. Nothing revises it when a filing later contradicts it Zero. Concurrence between copies is not corroboration F Work down the grade column when checking a figure. If the best available class is D or worse, the honest output is a range with the method named, or no number at all rather than a false precision. The Paper Trail Grade, A through F
Grades attach to the document class, not to the person and not to the plausibility of the number. A wildly implausible figure in a bankruptcy schedule is still grade A evidence of what was claimed under oath. A perfectly sensible figure with no source is still grade F.
The worked example below follows one modelled songwriter's publishing catalogue through each grade, showing how the same underlying subject gets stated differently depending on what evidence exists.
Grade Evidence standard How the same figure is stated at this grade What an entry may claim A Primary document filed with a regulator or a court, dated, and carrying perjury or statutory liability for misstatement Catalogue sold for $18.4M. Consideration stated in the buyer's Form 8-K dated 12 March 2024 The figure as fact, with the filing and its date B Statutory filing without individual attestation: company accounts, registry records, proxy tables The writer's holding company filed accounts at Companies House showing net assets of £6.2M at 31 December 2024 The figure as fact, flagged with the filing lag C A contractual instrument that establishes the mechanism but not the individual amount: a union agreement, a league CBA, a rate determination The writer share is set by the statutory mechanical rate under the Phonorecords IV determination covering 2023 to 2027. The rule is documented; this writer's total is not A modelled range with the mechanism cited. Never a point figure D A published estimate with a disclosed methodology and undisclosed inputs A wealth index estimates $40M for the catalogue using a published methodology whose per-person inputs are not shown The estimate, attributed to the publisher, with the method named E An attributed statement by an interested party A representative told an outlet the catalogue is worth north of $100M The claim and the claimant. The number is not repeated as fact F Repetition with no traceable origin Six sites carry $40M. None cites anything. The earliest traceable instance is a 2013 post Nothing. The entry records that no sourced figure exists Grade the document, then decide what may be printed. This is the rule that stops a plausible-sounding number from being laundered into a fact by being typed confidently. Notice the grade C row. The catalogue mechanism is fully documented, the rate is a matter of public record, and yet the individual figure remains unknown. That combination is by far the most common situation in this subject, and it is the one where careless reporting does the most damage, because the existence of a real rule gets treated as evidence for an invented total.
Repetition is not corroboration
Six sources agreeing means something only if they are independent. In wealth reporting they almost never are.
The pattern is easy to trace once you look for it. A round number appears, usually attributed to nobody or to an unnamed source close to the subject. A second site picks it up and cites the first. A third cites the second. Within eighteen months there are twenty instances and no origin, and any attempt to check produces a circle.
There is a diagnostic. Search the exact figure with a date restriction and work backwards to the earliest instance. If the trail dead-ends at an undated post with no attribution, the figure has no source and never had one. If it dead-ends at a filing, you have your document.
Round numbers are the other signal. Real documents produce $18.4 million and £6.2 million. Publicists and aggregators produce $20 million and $100 million. A wealth figure that is a suspiciously round number in the currency of the country where it was first published is usually an estimate that hardened.
Lag: the date on the document is the date of the number
Every filing describes a moment that has already passed, and the gap varies enormously by class.
An insider transaction report is two business days behind. A UK private company's accounts can be nine months behind the year end they describe, so a set filed in September 2026 may be reporting a position at 31 December 2025, which is eighteen months stale by the time anyone reads it. A bankruptcy schedule is a snapshot at the petition date and starts ageing immediately.
This is why an as-at date belongs beside every figure. Without it, a reader cannot tell whether they are looking at a current position or at a photograph taken before a divorce, a sale, a market fall or a career change. A number carrying its date can be assessed. A number without one cannot be assessed at all, which is a large part of why the aggregator model persists: undated figures never look wrong.
Which way each source lies
Bias in this field is systematic rather than random, which makes it correctable if you know the direction.
Anything filed by a debtor or a divorce litigant skews low. Both have an incentive to minimise, both are constrained by penalty of perjury and by an opponent with discovery rights, and the net result is a floor rather than a measurement. Treat those figures as at least this much rather than exactly this much.
Anything supplied by a representative skews high, and it skews high asymmetrically: the claim is made when the career is rising and never retracted when it is not. Publicist figures are the origin of an enormous share of the numbers still circulating, and their original purpose was leverage in a negotiation rather than accuracy.
Registry and land records skew high in a different way, by describing gross rather than net. A title register shows a purchase price and any registered charge, but not the outstanding balance on that charge. Counting a $12 million property as $12 million of net worth when $9 million of mortgage is outstanding is a common and very large error.
Aggregator figures skew high and, more importantly, skew stale. They ratchet in one direction, because adding an earnings year is easy and subtracting a divorce settlement requires somebody to read a court file.
Estimates with a method are not the same as estimates
A published wealth index that documents how it values private holdings is doing something meaningfully different from a site that types a number. The method can be argued with, which is the point.
Where these estimates get soft is the valuation of anything not publicly traded. A private company is typically marked against public comparables and then discounted for illiquidity, and the size of that discount is a judgement call worth tens of millions on a large position. Two competent analysts applying different discounts to the same holding will differ by 30% without either being wrong.
So the honest treatment is to name the publisher, name the method, print the number as an estimate, and never merge it with a document-derived figure to produce a total. Mixing a grade A holding with a grade D valuation and reporting a single sum produces a number that inherits the weakest input's reliability and the strongest input's apparent authority. That combination is how bad figures acquire credibility.
Checking a figure yourself, in order
The sequence matters, because the first hit is usually the best one available.
Start with the regulator. EDGAR covers US insider and beneficial ownership filings and is free and full-text searchable. Companies House covers UK companies, directorships and the persons with significant control register, also free. Both will tell you within minutes whether a named person holds anything reportable.
Move to the courts. PACER covers US federal filings, including every bankruptcy petition and the schedules attached to it. State court portals cover divorce and civil judgments, with varying access. County recorders hold deeds and tax lien notices; the Land Registry holds the equivalent for England and Wales.
Then the industry instruments. Union agreements, league collective bargaining agreements and rate determinations are all published, and they establish what a given contract type must pay. That is grade C evidence: excellent for the mechanism, useless for the individual.
If none of that yields anything, the finding is that no sourced figure exists. Write that down. An entry recording that a widely repeated number has no traceable origin is more useful to a reader than a number, and it is the only output that stays true when the source finally surfaces.