UQO Working Paper No. 8· v1.2· November 2026· 64 pages
Prediction Markets as Arrow–Debreu Securities: Evidence from Polymarket
State Prices, Executable Bounds and Macroeconomic News in Blockchain-Based Contingent Claims
Polymarket state prices sum to slightly more than one dollar at recorded closes (1.018 for the June 2026 FOMC market, 1.015 for the 2024 election), but at traded prices a round trip beat the no-arbitrage bound on only 1 of 89 covered days: the overround was not executable at size.
Negative-risk events list one contract per outcome and settle exactly one at one dollar, which makes their prices analogues of Arrow–Debreu state prices. The paper tests the law of one price for the complete set against bounds derived from the platform's mechanics, using recorded closes and the public trade tape, and benchmarks the implied policy expectation against fed-funds futures. The evidence rests on two markets at daily frequency; the executable test covers mostly February to May, and the announcement and lead–lag tests have limited power.
Simon-Pierre Boucher — contact@spboucher.ai
1.018 / 1.015
mean recorded sum of state prices in the Fed market (standard deviation 0.026, 133 closes) and the election market (304 closes); the Fed sum exceeds one on 86 percent of days and falls below the discounted bound by more than half a cent on 5 of 133 days
1 of 89 / 0 of 82
covered days on which a round trip at traded prices beat the upper bound (selling a minted set) or the discounted bound (buying and holding to resolution), net of ticks; the test covers mostly February to May and leaves the early sample largely untested
0.987 / 1.021
mean sum of the last bid-side and ask-side prints of the five Fed legs, from a trade tape of 247,968 fills: traded prices bracket one dollar
0.919
correlation in levels between the Polymarket-implied expected policy change and the fed-funds-futures measure for the June meeting over 94 common sessions (mean gap −0.9 basis points); only 0.180 in daily changes. The benchmark is provisional
9.9
variance ratio of the No change state price on the 13 scheduled release days against non-release weekdays (permutation p = 0.0024, Holm-adjusted 0.019); 40.9 on the 3 FOMC days
3.29, p = 0.150
the same ratio for the expected policy change: at the edge of the minimum detectable ratio of 3.28 for 13 events, not distinguishable from one, and below one when the 18 March FOMC day is excluded; the 5 CPI releases show no excess dispersion
no lead–lag detected
between daily changes in the implied expectation and the ten-year Treasury yield, the S&P 500, the VIX or Bitcoin on 93 trading days; relations explaining less than about 8 percent of next-day variance would not have been detected
Abstract
Polymarket negative-risk events list one binary contract per outcome and settle exactly one at one dollar: their prices are analogues of Arrow–Debreu state prices. We test whether the complete set is worth one discounted dollar in the five-outcome June 2026 FOMC market (133 closes) and the seventeen-contract 2024 election market (304 closes). Recorded sums average 1.018 and 1.015, above one on most days, below the discounted bound on a few; at traded prices a round trip beat the bound net of ticks on 1 of 89 covered days: the overround was not executable at size. The implied expected policy change tracks fed-funds futures (correlation 0.919). The modal price is far more dispersed on FOMC days; the release-day excess in the expected change is indistinguishable from zero with 13 releases; no daily lead–lag relation with Treasury, equity, volatility or cryptocurrency markets is detected; Granger tests against futures are unadjusted and deferred.
JEL: G13, G14, E52, D47, C58
Key Results
Executable bounds for the complete set of claims
Because spanning is built into the contract design, the testable content is the law of one price. The bounds follow from the mechanics of minting, converting and redeeming claims, and add a time-value lower bound that makes the band asymmetric.
Recorded closes checked against the trade tape
The public tape of the five Fed contracts lets the sum of state prices be measured at prices at which trades occurred. On the five days when closes sit below the discounted bound, no complete set could have been assembled from the day's fills at or below the bound.
A risk-neutral policy expectation benchmarked against futures
Normalised state prices give an implied expected change for the June 2026 decision that agrees with the fed-funds-futures measure in levels and only loosely in daily changes; this is not evidence of informational superiority over the futures market.
Announcement-day and weekend variance with stated power
Permutation inference with multiplicity adjustment and a stated minimum detectable effect shows repricing of the modal state on FOMC days, while the excess in the expected change is not established with 13 releases. Weekend changes have about one third of weekday variance in point estimate, significant for the modal price only.
A calendar caution for blockchain data
Polymarket's daily price points are stamped at 00:00 UTC of the following day, so a merge on the time stamp misaligns calendars by one day, and a lag search that selects the smallest p-value produces spurious lead–lag findings in short samples.
Quality control
This paper went through a multi-stage quality control: 6 quality-control reports and 77 points checked. The version presented on this page is the revised manuscript. Scores are on a 0–100 scale and are a diagnostic of the manuscript, not an acceptance probability.
Score after revision
65.0/100
50.2 before quality control · +14.8
Quality control dated November 4, 2026
- Quality-control decision
- claims not supported as written → reframed and revised
- Scientific integrity
- PASS
- Reproducibility
- R1reproduced from documented analysis data
- Maturity
- Bone revision cycle from submission
- Quality-control reports
- 6
- Points checked
- 77
Scores by dimension
Before quality controlAfter revision
- Originality & contribution45.8 → 50.0
- Scientific rigor51.3 → 67.0
- Identification & methodology50.0 → 65.0
- Data & reproducibility43.3 → 78.0
- Robustness & validation54.5 → 62.0
- Writing66.3 → 70.0
- Figures & tables59.3 → 76.0
- Literature positioning50.3 → 68.0
Revised manuscript64 pages13 figures15 tables6 equations83 references26,255 words
What quality control changed
Quality control found that the evidence did not support several claims of the earlier version. The announcement calendar was corrected, missing closes are no longer imputed, the pricing conclusion is now bounded by a test at traded prices, and the release-day result for the expected policy change is no longer claimed. A fed-funds-futures benchmark and an executable-bounds analysis were added.
- Fed sample of 146 daily closes with filled gaps → 133 closes on which all five contracts have a genuine close; mean sum of state prices 1.017 → 1.018
- Overround described as within no-arbitrage bounds on every day → the recorded sum falls below the discounted bound on 5 of 133 days, and the conclusion is restated as not executable at size at traded prices (1 of 89 and 0 of 82 covered days)
- Announcement calendar of 4 FOMC, 4 CPI and 5 employment days → corrected to 3 FOMC statements, 5 CPI releases and 5 employment reports with a close
- “The expected policy change moves two to five times more on release days” (variance ratio 4.9) → 3.29, not distinguishable from one (permutation p = 0.150); the excess dispersion survives only for the modal state price (9.9, Holm-adjusted 0.019)
- Lead–lag tests re-run on a no-imputation grid of 93 trading days with multiplicity adjustment: no lead–lag relation is claimed in either direction
- Volatility models kept in an appendix as a description of the fits, without inference; subtitle changed from No-Arbitrage Bounds to Executable Bounds
Limitations stated in the paper
- The study rests on two markets at daily frequency: 133 closes, 13 scheduled releases and 93 common trading days give the announcement and lead–lag tests limited power, so the nulls are not a demonstration of informational equivalence.
- The executable test covers 89 and 82 of the 133 days and few days in the first two months, where the recorded overround was largest; a full-depth order-book reconstruction could overturn the conclusion there.
- The futures benchmark uses a provisional definition of the pre-June rate, and its Granger tests are unadjusted and deferred.
- Fees, the conversion fee and gas are omitted from the bounds, and CPI and employment-report dates were transcribed manually because the source pages could not be archived.
Data
Fed decision in June 2026 (five outcomes)
Daily price histories and an order-book snapshot retrieved on 22 May 2026 from Polymarket's public APIs; 133 days between 11 December 2025 and 21 May 2026 on which all five contracts have a genuine close, with no imputation, and about $37.8 million of cumulative volume.
Presidential Election Winner 2024 (seventeen contracts)
304 days with a complete set of seventeen closes between January and 5 November 2024, ending before the outcome was known; used to check the pricing regularities in a larger and more heavily traded state space.
Public trade tape
247,968 fills in the YES and NO tokens of the five Fed contracts, each stamped with the side of the resting order, retrieved on 4 November 2026.
Conventional-market and calendar series
Effective federal funds rate from FRED for the discount factor; SPY, VIX, ten-year Treasury yield index and Bitcoin closes from Yahoo Finance; CME 30-day federal funds futures settlements; FOMC, CPI and employment-report dates (3 FOMC statements, 5 CPI releases and 5 employment reports with a close).
Methodology
Tests of the pricing restriction
Daily sums of state prices compared with one and with the discounted bound, their persistence (half-life of 2.4 days to the sample mean) and their decline towards resolution.
Executable round trips from fills
Sums of last bid-side and ask-side prints of the five legs and round trips that sell a minted set or buy and hold a set, net of a tick allowance, on the days the tape covers.
Risk-neutral distribution and futures benchmark
Normalised prices define a distribution over the five policy states whose mean is compared with the futures-implied expected change for the same meeting, in levels and in daily changes.
Announcement-day analysis
Variance ratios of release days against non-release weekdays for the expected change and the modal state price, with permutation p-values, Holm adjustment and minimum detectable ratios; weekend against weekday variance.
Lead–lag tests
Granger tests on a no-imputation trading-day grid with lag order fixed by information criteria, conventional and HAC statistics, and Holm and Benjamini–Hochberg adjustment over the family of 24 tests.
Reproducibility
- The raw Polymarket snapshot and FRED files are committed with SHA-256 checksums; the snapshot is the only source of the Fed panel because the platform no longer serves the daily histories of the resolved contracts.
- The public trade tape is committed with its checksums, and a re-pull is checked against the manifest.
- A single entry point regenerates the processed panels and every statistic, table, figure and inline number, and a test checks that regeneration is byte-identical.
- Yahoo Finance files and the raw futures settlements are not redistributed: a fresh clone must re-download the former and obtain the latter under licence; the derived futures series is committed pending confirmation of the licence terms.
- The Federal Reserve calendar page is archived; the Bureau of Labor Statistics schedule pages could not be archived and their dates were transcribed manually.
Cite this paper
Boucher, S.-P. (2026). Prediction Markets as Arrow–Debreu Securities: Evidence from Polymarket. UQO Working Paper No. 8, Université du Québec en Outaouais.
@techreport{boucher2026wp8,
author = {Simon-Pierre Boucher},
title = {Prediction Markets as Arrow–Debreu Securities: Evidence from Polymarket},
institution = {Université du Québec en Outaouais, Département des sciences administratives},
number = {8},
type = {UQO Working Paper},
year = {2026},
month = {nov},
url = {https://www.spboucher.ai/research/wp8},
}More in Finance & Economics