Notes · 25 September 2026
The stale-price flag misses five.
An asset manager sends four years of prices and the positions of fifteen funds, and asks what the platform holds on 26 June 2026. The export has 761,005 price rows. On that date the funds hold 2,133 positions in 669 securities, EUR 13.55 billion in total. The data are synthetic; there is no live fund behind them.
A flag is someone else's count
The price vendor marks prices it considers stale with its own flag. On the reporting date the flag sits on 40 positions. It is tempting to take that as the number of positions valued at a price that did not move.
The prices say something else. 42 positions in 14 securities are valued at a price exactly equal to the previous trading day's close: the quote most likely did not move. That is 2.27% of gross value. Of those 42, the vendor flag misses five. And on three positions the flag is set although the price did move.
The client's own data dictionary already said the flag is not a complete list. So the document prints both counts, 40 by the vendor flag and 42 by the prices, and says which is which.
The figure the total does not show
The headline, EUR 13.55 billion, is right. But 37.2% of gross value sits in instruments that are suspended (14.4%) or delisted (22.8%) on that date, carried at their last price. How to value them was the owner's reading, chosen before any figure was computed, and it is the first thing a reader of the total needs to know.
What I take from it
- A field in an export is somebody's opinion about the data. When the data can answer the same question directly, count both and print both.
- Decide how to treat the awkward instruments before counting, not after the total looks strange.
- All 106 figures of the report were recomputed from the export, and the receipt carries a checksum of every file, so anyone can see the document was not changed afterwards.