Who controls the Senate after November?
50,000 simulated elections, seeding each race with its polling average, correlated polling error, and a prediction-market prior. Where the race stands today — current polling averages blended with prediction-market odds. A work in progress from Policy y Peaches.
Today’s climate (Republican president) implies a national D+6.5 margin. Measured against the 2024 House result, that is a uniform D+9.3 swing applied to every state’s fundamentals prior. It moves thinly-polled races the most.
Baseline: 44 safe D seats + 47 safe R seats; 9 competitive races simulated. Democrats need 51 seats for control.
Market odds also feed the per-race blend (weight 25% market / 75% polls).
- Polling margin
- D +6.8
- P(D), polls
- 86%
- P(D), blended
- 88%
- Polymarket
- 94%
- Kalshi
- —
- Polling margin
- D +9.0
- P(D), polls
- 92%
- P(D), blended
- 86%
- Polymarket
- 66%
- Kalshi
- —
- Polling margin
- D +8.1
- P(D), polls
- 90%
- P(D), blended
- 91%
- Polymarket
- 93%
- Kalshi
- —
- Polling margin
- D +5.3
- P(D), polls
- 79%
- P(D), blended
- 77%
- Polymarket
- 70%
- Kalshi
- —
- Polling margin
- D +13.3
- P(D), polls
- 98%
- P(D), blended
- 93%
- Polymarket
- 81%
- Kalshi
- —
- Polling margin
- D +3.3
- P(D), polls
- 66%
- P(D), blended
- 63%
- Polymarket
- 55%
- Kalshi
- —
- Polling margin
- D +2.3
- P(D), polls
- 59%
- P(D), blended
- 45%
- Polymarket
- 3%
- Kalshi
- —
- Polling margin
- R +0.2
- P(D), polls
- 38%
- P(D), blended
- 38%
- Polymarket
- 39%
- Kalshi
- —
- Polling margin
- R +1.0
- P(D), polls
- 32%
- P(D), blended
- 32%
- Polymarket
- —
- Kalshi
- —
How the simulation works
- Each race’s polling average margin is converted to a win probability using a fat-tailed Student-t error model: a national error (σ = 2.899) shared by every race plus an independent per-race error (σ = 5.144), sized to historical Senate polling misses. A shared tail shock allows unusually large misses to move several races together.
- The national midterm climate — the president’s approval and the generic congressional ballot — is folded into each race’s fundamentals prior as a uniform D+9.3 swing versus the 2024 House result. It mainly lifts thinly-polled seats; heavily polled races already price the climate in.
- Prediction-market odds are blended in at 25% weight — markets aggregate information polls miss, and the weight is a tunable model parameter.
- We then simulate 50,000 elections. The shared national error is what makes sweeps (all close races breaking one way) more likely than independent coin flips would suggest.
- This shows where the race stands today, given current polling and market prices. A work in progress from the team at Policy y Peaches — learn more here.
A work in progress from the team at Policy y Peaches. Learn more here.
Last updated: Sep 20, 2026, 12:01 AM