Four Hundred Sixteen Vertical Strokes

University College, the Biometric Laboratory, this 14th of November - I spent the better part of the day ruling a chart by hand, four hundred and sixteen vertical strokes, one for each expiration week in the ledger, each stroke rising to the height of that week’s index return. Then I drew a single horizontal line across the whole sheet at the level of two standard errors. The line existed on the ruler before the strokes went down; it was invisible in every practical sense. Not one eye in the room - mine included - could have told you where the boundary ran until a bar crossed it. Two hundred and sixty-three bars stayed below. A hundred and fifty-three broke through. The boundary was nothing until it was transgressed, and then it was the only thing on the page.

The datum that matters: mean return across expiration weeks, 0.35 percent, standard error 0.09, t = 3.9, p < 0.001. Effect size alongside the probability, always - 0.35 percent is not a fortune, but it is not noise either, not at n = 416. The null hypothesis is that expiration weeks return no more than ordinary weeks. The test rejects it. The direction is positive. This is a finding, not a failure to find.

What surprised me was the asymmetry of the drawing itself. Below the line, the bars are merely data; above it, they become evidence. The same ink, the same paper, the same week of returns - and the crossing changes their legal character entirely. Below the threshold a bar says nothing. Above it, the bar indicts the null. I had drawn such verdict lines a hundred times in correlation work and never noticed that the threshold has no existence of its own. It is a wall that exists only where bodies press against it, the way a border is real only at the customs post.

A caveat for the record, because the sample is weekly and the standard error was estimated, not known: the test assumes the weeks are independent, and hedging flows may bind adjacent expirations together. If they do, my effective n shrinks and the t with it. The rejection would soften before it broke.

Tomorrow I recompute the standard error with a block bootstrap on overlapping weeks, and I will rule the line again at whatever height the new errors demand.