Reviews
Part of Field guide to estimate methodology: the five-step pipeline
Getting highest estimate methodology right the first time
Highest estimate methodology rankings for 2027: why estimators cannot be ranked for accuracy without a truth to check against, and what can be compared instead.
Pages that rank the best sources for financial estimates run into a problem before they begin. Ranking estimators by accuracy requires knowing the right answer, and if the right answer were public nobody would need the estimators. A highest estimate methodology works from filings instead, and the order of the steps is what keeps it honest.
What to take away
- Accuracy cannot be ranked without ground truth, and there is none here.
- Agreement between estimators measures copying, not correctness.
- What can be compared is disclosure, and that comparison is worth running.
- A highest estimate is a bound, not a forecast, built from filings outward.
- Every assumption that pushes the number up gets printed beside the number.
The eight steps of a highest estimate, in order
1. Fix the quantity before touching a number. Whose net worth, on which date, in which currency, gross or after tax. A dated figure for one person can be checked. "The family's wealth" cannot.
2. Start with the documented slice. SEC Form 4 lands within two business days of a purchase or sale by an officer, director or holder above 10 percent. The DEF 14A proxy carries the summary compensation table for named executives. Schedule 13D follows a stake above 5 percent of a class, and Form 13F comes quarterly from managers above $100 million.
3. Price every disclosed holding on one stated date. Use that day's closing price and write the date into the copy. Filings are free on EDGAR.
4. Value private assets by a named method. The last disclosed transaction price, a stated multiple of a comparable public company, or book value. Name the one you used.
5. Apply the upper-bound rules out loud. No marketability discount, no minority discount, no haircut for tax or pledges. Discounts for lack of marketability in the restricted-stock studies typically run 20 to 35 percent. A highest estimate drops them on purpose and says it did.
6. List what you cannot see. Trusts, private debt, art, pledged shares, obligations still being settled.
7. Print the arithmetic and the assumptions. Someone else should be able to re-run the whole sum in an afternoon.
8. Date-stamp and version the figure. Keep the old versions, so a reader can see what changed and which filing changed it.
Follow that sequence and the checkable parts anchor the modeled ones. Skip step one and the rest of the work has no fixed target to hit.
Why an accuracy ranking cannot be built
To rank estimators by accuracy you need, for a sample of cases, the true value. Then you measure each estimator's distance from it.
No true value exists for private wealth: no register, no audit, no later event reveals it, because estates settle privately after taxes and obligations, years later, and describe a different quantity anyway.
The absence of any correction mechanism is covered on a separate page here. It is why no scoreboard can exist.
So a ranking of estimators is either built from something other than accuracy, or built from nothing.
The proxies people use instead, and why each fails
Agreement with other estimators. Treated as a proxy for correctness. It measures how closely a compiler tracks the consensus, and where the consensus is inherited from a common ancestor it measures copying. An estimator who deviated because they had better information would score badly.
Reputation of the publisher. Measures brand, resources and history. Those correlate with care and not with access to private liabilities, which nobody has.
Length of the methodology page. Measures how much was written about process. A long description of care is not a set of instructions somebody else could follow.
Frequency of updating. Measures editorial activity. Refreshing an inherited number more often produces a newer inherited number.
Number of sources cited. Measures citation volume. If the citations are other articles descending from one origin, the count adds apparent support and no information.
Each proxy is measuring something real. None of them is measuring accuracy, and presenting any of them as a quality ranking is the same error the genre makes about people, applied to publishers.
What can be compared honestly
Disclosure. Not whether an estimator is right, which is unknowable, but whether they have told you enough to judge them. That comparison is entirely feasible, and it takes about fifteen minutes per publisher.
| Question | What a good answer looks like |
|---|---|
| What is being measured? | The person, the date, the currency, and whether the figure is gross or net of tax. |
| Which holdings are public? | Each one listed with its exchange and the date it was priced. |
| How were private assets valued? | A named method: last transaction, comparable multiple or book value. |
| What was left out? | Trusts, private debt, pledged shares, and anything the estimator could not see. |
| Is it a point estimate or a bound? | A stated range, plus the assumption that produces the top of it. |
| When was it updated, and why? | A date and the event that triggered the refresh, such as a Form 4. |
| Who signed it? | A named author or desk, not a brand. |
Disclosure questions to ask
- Is the population defined?
- Is the quantity named and used consistently?
- Is there a valuation date?
- Are the valuation rules given?
- Are liabilities addressed?
- Is uncertainty published per entry?
- Are corrections marked?
A high score in this comparison does not prove an estimator is accurate. It proves accountability, the only property a reader can verify.
The conventions behind this comparison are long established in other fields. Statistical agencies publish their designs in full, as the Bureau of Labor Statistics does in its Handbook of Methods. Survey researchers have recorded what must be released with a result, in the disclosure standards of the American Association for Public Opinion Research.
The one genuine accuracy test available
There is a narrow case where accuracy can be checked. Where an estimator makes a claim about something that is disclosed, such as a company officer's reported compensation or shareholding, you can compare their figure against the filing through the SEC's EDGAR full text search.
The one checkable accuracy test
Does the estimator claim a disclosed figure?
compare against the SEC filing
accuracy cannot be checked
That test only covers the documented slice, which is the part least likely to be wrong. It is still worth running, because an estimator who cannot get the checkable part right has told you something useful about the rest. How much of an executive figure is documented and how much is modeled is on executive net worth.
Why we publish no ranking of sources
Because we would be ranking by proxies while implying accuracy, which is the exact failure this site exists to describe. What we publish instead is mechanism, on estimate methodology, the reasons the list format resists uncertainty, on richest rankings, and the reason lifetime totals are the least testable claims of all, on career earnings.
Common questions
Is any estimator clearly better than the others?
Some are clearly more transparent, and that is a real and checkable difference. Better in the sense of closer to the truth cannot be established by anyone, including them.
What if an estimator's figure is later confirmed by the subject?
A confirmation from an interested party without a document is an assertion agreeing with an estimate. It is weak evidence, and it is also the strongest that this subject usually offers.
Should a reader prefer the most cautious estimator?
Prefer the most explicit one. Caution that is never stated cannot be assessed, and an estimator who publishes their assumptions lets you re-run the reasoning yourself.
