Where the numbers come from.
Articles about the provenance and the checking of market data, written from our own dated verifications. No buy or sell recommendations.
Articles about the provenance and the checking of market data, written from our own dated verifications.
A DCF model is five lines of arithmetic. The number it returns is not the point.
How to build a discounted cash flow model step by step, with an example you can redo by hand, and the three inputs where it quietly breaks.
Illustration · the article's worked example, made-up round figures (€100M cash flow, 9% WACC, 2% perpetual growth).
November 2026
You bought a world fund. 72.6% of it is the US.
We opened the holdings of an MSCI World ETF as of 30 October 2026: 1,250 companies, 72.6% based in the US, and the top ten are the same as in the S&P 500.
October 2026
Telefónica fell 13% in five years. Holding it returned 27%.
We measured the 34 IBEX 35 stocks listed for five years: adding dividends lifts the median return by 25.6 points and flips the sign for three of them.
What we found wrong this quarter, with names and figures
First quarterly data quality report: the dividend and EPS figures we found wrong between July and October 2026, which ones were visible, why, and what stops it.
A DCF model is five lines of arithmetic. The number it returns is not the point.
How to build a discounted cash flow model step by step, with an example you can redo by hand, and the three inputs where it quietly breaks.
Same chart, two different supports
A support drawn by eye depends on who draws it. The rules that make a support, a resistance or a trendline something you can check against the bars.
September 2026
A stock with a beta of 0.4 can still go to zero
Beta measures how much a stock moves when the market moves. It does not measure whether the business can fail, and those are different questions.
EBITDA is not profit. The gap has a name.
Two companies with identical EBITDA can produce $140M of free cash flow and $35M. What separates them is the line EBITDA removes on purpose.
ETF vs index fund: the fee is not what decides
They track the same index and cost about the same. What separates them is which one hands you a capital gains bill you did not ask for.
Profit did not go up. EPS did.
A company that buys back 10% of its shares raises earnings per share by 11% without earning a cent more. When that creates value and when it does not.
A stock split does not make you richer. Not by a cent.
You hold twice the shares at half the price. Your money is exactly the same. What does change, and almost nobody mentions, is something else.
The same fund, at 0.85% and at 2.81%
Measured across 332 funds: the median charges 1.23% a year. And the same fund costs three times more depending on which share class you are sold.
Contango and backwardation: what the oil curve says
With the next contract at 88 it signals scarcity; at 94, oversupply. What the shape of the futures curve measures, and why almost nobody publishes it.
Half a point of WACC. Between 9% and 16% of the price.
A DCF looks objective because it ends in a number with decimals. Move the discount rate half a point and watch how far that number travels.
A 149% operating margin. Nobody had noticed.
A water utility served with $136.7M of revenue where there was $658.1M. Another reported negative revenue. Nine wrong figures, and the four causes behind almost all of them.
Ex-dividend date: why you bought and got nothing
You bought before the payment date and no dividend arrived. The ex-dividend date is why, and confusing it with the payment date is extremely common.
The same stock, two different P/E ratios
One site says P/E 18, the other says 26. Neither is lying: they are dividing by different earnings. Which ones, how much it changes, and how to tell.
An inverted yield curve is not a prediction
The 10-year at 4.79% means little on its own. Whether it sits 40 basis points above or below the 2-year is what separates growth from recession.
Asking more often does not make the figure exist sooner
Two hours after eight companies filed results, six were still being served with the previous quarter. Polling harder would have changed nothing: the figure was not there.
Why a market figure needs a second source
Auditing your reader proves you read the source correctly. It does not prove the source is right. What happened on 4 September 2026, and why averaging two providers is the wrong fix.
The Spanish originals are at celsmar.com/blog, and each article links to its counterpart.
How this is written.
Four rules, and none of them bends for a better headline.
- Every article starts from a check
- Nothing gets written because a topic is searched a lot. An article is written when there is a measurement of our own behind it (an audit, a cross-check, an error found), and it is published with its date and its method so that anyone can dispute it.
- None of them says what to buy
- There are no buy or sell recommendations here, no model portfolios and no price targets. Celsmar is not authorised as an investment firm, and this is information and analysis, not advice.
- Figures carry their origin
- Every figure in an article says when it is from and which measurement it comes from. When the data does not exist or could not be checked, we say so instead of rounding it towards something more presentable.
- What goes wrong, too
- The errors we find in our own data are published with the before and the after. A data tool whose blog never reports a failure tells you nothing.
How this is written · 4 rules
Every figure in the terminal carries its date and its source.
Open it and check a company you know well. No card. 2 AI reports a month. Cancel anytime.