3 Sep 2026
Phronopolis
A journal of minds & margins
EN NL

The Ledger and the Expiration Week

John Herschel · 3 Sep 2026
The Ledger and the Expiration WeekA jagged, crystalline topography of midnight blue and silver shards fractures in sharp relief, suggesting data that is simultaneously solid and dissolving. Pale grey atmospheric haze softens distant peaks, while cool, directional light from the left catches foreground geometry, rendering it with sharp vector precision and soft radial blur for background mist. Palette: Midnight Blue, Silver, Pale Grey, Deep Charcoal. Mood: uncertain, luminous, structurally fragile. Render with layered linear gradients for depth and a glow filter on highlights to evoke the tension between signal and noise.

Dear Mr. Baily,

The ledger came back from the copyist yesterday, and with it the thing I want to put before you plainly: the expiration-week premium, 0.35 percent on average, holds in the series I reconstructed - and I no longer trust my own instinct about what that means.

Provenance first, as always. The figures derive from the return series for equity index weeks, partitioned by expiry calendar, tested against the residual weeks under the same construction. Four of the five confirming runs draw from partitions of the same underlying tape - re-sliced, not independent. Treat the confirmation as partial. The fifth run, built on a different construction with different weighting, gives a weaker and noisier result. Whether that fifth run is a poor instrument or an honest null, I cannot yet say, and the distinction matters more than the 0.35 itself.

You will remember our arguments about the nebulae - how a chance alignment of two objects at different distances can imitate a physical association, and how only parallax or independent distance evidence separates the pair from the accident. Here the accident and the mechanism wear the same face. Gamma hedging by the dealers would produce exactly this signature, and so would a dozen artefacts of the calendar: settlement conventions, reporting dates, the simple fact that expiries cluster. The correlation is real in the data; its parentage is unproven.

I tested it the way I would test a suspected double star. Observe again, under different conditions, at different times, and see whether the pairing survives. I re-ran the partition with the three largest expiration weeks excluded - the result diminished but did not vanish. I shifted the window by one day - it degraded. Both outcomes are consistent with a real but fragile effect, and both are consistent with a systematic bias that decays at the edges. The resolving power of this data is simply not equal to the question we are asking of it.

No reliable decomposition exists for the hedging-flow component, because the flow data is not public and the proxy series available to us are constructed from the same exchange records that generated the original signal.

The phenomenon continues unresolved. The next expiry falls in ten days; I shall be watching, and the notebook is open.

← Lab