DV-PMI Surveillance · Surprise

Resolution Surprise

Cumulative snapshot through 2026-02-28

In markets that resolved opposite to the consensus pre-resolution price (the "high surprise" tercile), flagged wallets show 0.054 excess accuracy while the non-flagged population shows -0.396. The pooled regression coefficient on the flagged indicator is 0.477 (t = 49, n = 2.0M trades). Flagged wallets are positive in every surprise tercile; the population is negative in every tercile. The pattern is consistent with ex-ante knowledge of outcomes that ran counter to the prevailing market consensus.

Surprise tercile	Flagged EA	Non-flagged EA	Gap
low	+0.046	-0.394	+0.441
medium	+0.166	-0.224	+0.390
high	+0.054	-0.396	+0.450

Excess accuracy is wallet accuracy minus max(price, 1-price), the accuracy achievable by mechanically following the price. Surprise terciles are sorted by absolute deviation of final outcome from the pre-resolution consensus price.

Regression summary

Model	n	β flagged	t flagged	β surprise	t surprise	β interaction	t interaction	R²
(1) Market surprise	2.0M	0.477	49.4	-0.136	-101.5	0.166	9.09	0.012
(2) + Volume	2.0M	0.456	48.4	-0.134	-99.9	0.138	7.79	0.015
(3) Trade surprise + Volume	2.0M	0.514	80.1	0.101	97.7	0.018	1.61	0.015

Interpretation

The "surprise tercile" sort isolates markets in which the realized outcome diverged most strongly from what trading prices had implied. In a clean market, no trader should systematically profit from these markets because the consensus price reflected all public information and was wrong only because of an unforecastable terminal outcome. The data show otherwise. Wallets flagged by the orthogonality test maintain positive excess accuracy across all surprise terciles, including in markets where the population was systematically wrong. The gap widens at higher levels of surprise.

Two interpretations remain on the table: (a) flagged wallets had ex-ante information about the outcome that the market consensus lacked; (b) flagged wallets had ex-ante information about which markets were likely to be mispriced ex-post, which is a softer version of the same claim. Either is inconsistent with the strong-form efficient-pricing null. Neither, by itself, establishes the presence of material non-public information.

Methodology

Each trade is assigned to a surprise tercile based on the absolute difference between the market's terminal outcome and its pre-resolution mid-price (T-1 day mid). The reported regression is OLS of trade-level excess accuracy on a flagged-wallet indicator, surprise magnitude, their interaction, and (in model 2-3) trade volume. The interaction term in model (1) is positive and significant, indicating flagged wallets do disproportionately well in high-surprise markets.

Limitations

Surprise is measured at the trade level using the T-1 mid-price. A subset of trades occur within the T-1 window itself, mechanically loading on the surprise; the trade-surprise specification in model (3) addresses this by using the price prevailing at trade time rather than at T-1.

Source paper

Della Vedova, J. (2026). Detecting Informed Trading in Prediction Markets: An Orthogonality Test. SSRN. Section: Resolution Surprise Test.

Data

surveillance_surprise_latest.json

Other surveillance indices

← All surveillance indices · Informed Trading (PII) · Insider Timing · Adverse Selection · Wash Trading

Cite this index

@misc{dellavedova2026surprise,
  title        = {Resolution Surprise Index},
  author       = {Della Vedova, Joshua},
  year         = {2026},
  url          = {https://jdellavedova.com/surveillance/resolution-surprise}
}