AI-generated research, checked but not peer-reviewed. It may contain errors.

E817　 KILL　 Below costsTested · October 2026

Thin election markets show no lag worth trading after costs

On the 2024 US election night, thin Senate and governor contracts followed a 5-cent presidential move by a gross 2.14 cents in 15 minutes, indistinguishable from zero and below the 3-cent cost.

prediction-markets election-night lead-lag attention thin-markets event-contracts

Readable summary Full research record JSON recordFeedback on E817

On election night one stream of vote counts moves every race in a state, but traders and market makers watch the biggest market. The belief: thin Senate and governor markets update minutes after the liquid presidential market of the same state. We found 48 moves of at least 5 cents in the presidential contract, spread over 11 state pairs, and copied each move in the thin contract for 15 minutes. The gross gain was +2.14 cents per contract. After a 3-cent round-trip cost it was -0.86 cents, with an interval from -3.6 to +1.6 cents. Only 3 of 11 pairs had a positive mean.

Signals48in 11 state pairs, 5-cent moves in the leader market

Gross move captured+2.14 cper contract over 15 minutes, before the 3 c cost

Net per contract-0.86 cpair-clustered 95% interval -3.6 to +1.6 c

Pairs with positive net3 of 11rule required at least two thirds

01 · The idea being tested

On election night, thin down-ballot markets react minutes after the liquid presidential market of the same state, so following the big market's move in the small one pays.

Why people believe it

People argue that attention and market-making capital are scarce and concentrate on the largest, most visible market. A vote count that reprices the presidential race also bears on the Senate and governor races in the same state, but fewer people watch those markets, so their prices should adjust later.

Why it would persist The argument, as stated before testing

The down-ballot markets hold far less money, so the cost of quoting them and of trading them is high relative to the small lag; few traders would bother to watch every pair at once on the night.

02 · How it was tested

We took every state that had a presidential winner market and a Senate or governor winner market on an on-chain prediction market, and watched the minute-by-minute prices of one fixed side in both. Whenever the presidential contract moved by 5 cents or more in 5 minutes, we simulated buying or selling the thin contract in the same direction and closing it 15 minutes later, paying a flat 3-cent cost. As a control we swapped the roles: the thin contract led and the presidential contract followed, to see whether any lag is one-way.

Pass required

At least 6 pairs with signals and at least 40 signals; mean net above 0 with pair-clustered bootstrap 95% lower bound above 0; positive mean in at least two thirds of pairs; mean above 0 without the best pair; and the reverse-role control has a smaller mean.

Killed if

Any of the pass conditions fails.

Costs

Flat 3 cents per round trip, set before the test as a spread and fee allowance for thin markets. Prices are in cents per contract; one cent is one percentage point of probability.

Controls

Reverse-role control: the same rule with the follower as leader and the presidential contract as follower. A real one-way lag should show a smaller result there. A second control using a national presidential market as the leader was not available.

03 · What happened

Why it was rejected

The gross lag of +2.14 cents is smaller than the 3-cent cost, so the net is -0.86 cents; the interval includes 0, only 3 of 11 pairs were positive and the mean without the best pair was -2.0 cents.

48 signals in 11 state pairs (44 follower markets, 40 of them with a same-state presidential leader). Gross move captured: +2.14 cents per contract over 15 minutes; net of the 3-cent round trip: -0.86 cents.
The pair-clustered 95% interval for the net is -3.6 to +1.6 cents. 3 of 11 pairs had a positive net mean, against the two thirds the rule required.
Without the best pair the net mean was -2.0 cents.
In the reverse-role control (49 signals in 12 pairs) the gross move was -0.04 cents and the net, after the 3-cent cost, -3.04 cents.
In the 5 minutes before each signal the follower had already moved +1.6 cents in the leader's direction (signed), so part of the gross move happens before the entry.
The rule failed on the lower interval bound, on the share of positive pairs and on the without-best-pair condition. It passed the minimum pairs (11 against 6) and signals (48 against 40) and the reverse-role comparison, but passing those minimums is not a result: the rule needed every gate.

The lag is smaller than the cost

Mean result per signal in cents. The gross move is positive, but after the 3-cent round trip the net sits below zero and its 95% pair-clustered interval spans zero. Without the best pair the net is lower still.

Chart data (cents per contract)

	Mean	95% low	95% high
Gross move, before cost	2.14	Not reported	Not reported
Net of 3 c cost (n 48)	-0.86	-3.6	1.6
Net without the best pair	-2	Not reported	Not reported
Reverse-role control, gross	-0.04	Not reported	Not reported

Gross +2.14 c, not distinguishable from zero

Gross 15-minute move when the presidential contract leads versus when the thin contract leads. The presidential-leads case is +2.14 cents, not distinguishable from zero, and under the 3-cent cost line.

Chart data (cents per contract)

	Gross move per signal
Thin contract follows presidential	2.14
Presidential follows thin (control)	-0.04

04 · What it means

Why it failed Interpretation · not established

The gross move of +2.14 cents over 15 minutes is not distinguishable from zero, and even taken at face value it is smaller than the spread and fee a trader pays in a thin market. The follower had already moved +1.6 cents in the leader's direction before the signal; whether that reflects fast reaction was not tested.

Lesson

A small average lead of one market over another is not a trade. In thin markets the cost of crossing the spread is of the same order as the whole move, so a lead-lag of this size is worth nothing to a taker. A reverse-role control is what separates a one-way lag from two markets that simply move together.

Scope of this result

Covers one election night (2024 US general election), Senate and governor winner contracts against the presidential contract of the same state, one fixed side per state, 1-minute prices, a 15-minute hold and a flat 3-cent cost. The national-leader control was not available.

A rejected experiment applies to the tested conditions and does not prove an idea can never work. bp = basis point, 0.01%.

1. Test design

Pre-registered

Yes: rule and criteria fixed before the test data was examined

Data

1-minute price history of presidential-winner contracts (leaders) and Senate or governor winner contracts (followers), one fixed side per state — an on-chain prediction market, public price-history endpoint (anonymous); 1 minute; 2024-11; public; window from 23:00 UTC on election eve to 12:00 UTC the next day

Universe

44 follower markets (Senate and governor winner markets), of which 40 had a presidential-winner market in the same state. 11 state pairs produced at least one signal. The followed side is the same party's contract in both markets, fixed in advance from public candidate lists.

Signal

A signal at minute t when the leader price moved at least 5 cents over the previous 5 minutes (|L(t) - L(t-5)| >= 5 c). Signals are non-overlapping within a pair: the next one at least 15 minutes later. Skipped when the follower price was at or below 3 cents or at or above 97 cents (race already decided).

Execution

Enter the follower at its price one minute after the signal, in the direction of the leader's move; exit at the price 16 minutes after the signal (15-minute hold). Net = signed follower move minus cost.

Costs

Flat 3 cents per round trip, set before the test as a spread and fee allowance for thin markets. Prices are in cents per contract; one cent is one percentage point of probability.

Controls

Sample split

No split: one election night. Results reported per state pair, and without the best pair.

Pass criterion

Kill criterion

Any of the pass conditions fails.

Power

One night gives few independent events: 48 signals in 11 pairs. The pair-clustered interval on the net spans -3.6 to +1.6 cents, so a net effect of about 2 cents in either direction cannot be separated from zero.

2. Result

Observations

Instruments

Net mean

-0.86

95% interval

-3.6 to +1.6

Median

—

Worst

—

Finding 1

48 signals in 11 state pairs (44 follower markets, 40 of them with a same-state presidential leader). Gross move captured: +2.14 cents per contract over 15 minutes; net of the 3-cent round trip: -0.86 cents.

Finding 2

The pair-clustered 95% interval for the net is -3.6 to +1.6 cents. 3 of 11 pairs had a positive net mean, against the two thirds the rule required.

Finding 3

Without the best pair the net mean was -2.0 cents.

Finding 4

In the reverse-role control (49 signals in 12 pairs) the gross move was -0.04 cents and the net, after the 3-cent cost, -3.04 cents.

Finding 5

In the 5 minutes before each signal the follower had already moved +1.6 cents in the leader's direction (signed), so part of the gross move happens before the entry.

Finding 6

The rule failed on the lower interval bound, on the share of positive pairs and on the without-best-pair condition. It passed the minimum pairs (11 against 6) and signals (48 against 40) and the reverse-role comparison, but passing those minimums is not a result: the rule needed every gate.

3. Interpretation

Why it failed

Lesson

Scope

4. Reproducibility

Steps

1) For each US state, find the presidential-winner market and any Senate or governor winner market on an on-chain prediction market, and take the Yes token of one fixed side in each. 2) Download 1-minute price histories from 23:00 UTC on election eve to 12:00 UTC the next day and forward-fill gaps. 3) At each minute where the leader moved at least 5 cents in 5 minutes, and at least 15 minutes after the previous signal in that pair, enter the follower one minute later in the leader's direction and exit at minute 16, skipping followers priced at or below 3 or at or above 97 cents. 4) Subtract 3 cents and bootstrap by state pair; repeat with the roles swapped for the control.

Grade

COMPLETE