What if the smartest money in the room already knew the outcome of the midterms weeks before the first ballot was counted? That is the quiet question hanging over the latest numbers coming out of the prediction markets. One platform alone just cleared more than 2.5 billion dollars in midterm trading volume, nearly double the total recorded during the entire previous election cycle. The figure is striking on its own, but the real story sits in the details: how accurately those markets called the races, how the odds shifted once results started rolling in, and what it all suggests about the future of election forecasting.
Kalshi Midterm Trading Volume Nearly Doubles Previous Cycle
The headline number is hard to ignore. Midterm election contracts on the platform generated more than 2.5 billion dollars in trading activity. That is an increase of roughly 92 percent compared with the 1.3 billion dollars seen across the 2024 cycle. Keep in mind the comparison is not perfectly apples-to-apples. The earlier period included a presidential contest that naturally draws heavier attention and capital. This year’s focus stayed on congressional seats, governorships, and a string of state-level races. Even so, the jump in volume remains impressive.
Traders appeared to lean toward Democratic control of both chambers well before Election Day. The markets did not fully capture the eventual size of those gains, yet the directional call held up. In my view, that combination of volume and directional accuracy is what makes these numbers worth a closer look. Prediction markets are no longer a sideshow. They have become a real-time signal that many analysts now watch alongside traditional polls.
How Accurate Were the Market Favorites?
By early morning on November 4, markets had resolved 487 races. Favorites won 479 of them. That works out to a hit rate of about 98.4 percent, which the platform rounded to 98 percent in its public summary. The breakdown is revealing:
- House races: 356 favorites correct out of 360
- Senate contests: 32 correct out of 33
- Gubernatorial elections: 33 correct out of 34
Only eight favorites lost across the entire set. A handful of races, including the Alaska and Mississippi Senate seats, remained unresolved at the time of the initial recap, so the final accuracy figure could still shift slightly. Still, the early numbers paint a clear picture of markets performing at a very high level.
I’ve found that simply counting winners is a blunt way to measure skill. A contract priced at 51 percent that pays out receives the same directional credit as one priced at 95 percent. That is why researchers also track Brier scores, which penalize overconfident or underconfident forecasts. The gap between those two approaches becomes important when you dig deeper into the data.
Election Night Odds Shifted in Real Time
The live feed from election night offers a different kind of insight. At 6:55 p.m. Eastern, the probability of Democratic House control sat at 97.2 percent. By 8:10 p.m. that figure had climbed to 99 percent. Senate control crossed the same 99 percent threshold roughly fifty minutes later.
Individual races moved just as quickly. One update noted that a long-serving Republican incumbent in the Texas governor’s race was no longer favored by 8:06 p.m. Later posts showed a Democratic challenger in the Texas Senate contest reaching 99 percent probability at 9:35 p.m., followed by another candidate in South Carolina at 10:10 p.m. A Texas gubernatorial challenger hit the same mark just before midnight. These were market projections rather than official calls, yet they tracked the emerging results with striking speed.
Perhaps the most interesting aspect is how markets absorbed information that traditional media was still processing. Traders appeared to update their views as soon as partial counts and exit data became available. That continuous recalibration is one of the features that sets prediction markets apart from static polling averages.
Research Tracks Forecast Improvement Over Time
A separate internal analysis examined races called by 6:26 a.m. on November 4 and tracked the quality of forecasts from three months before Election Day through the moment polls closed. Directional accuracy rose from 93.8 percent to 98.2 percent, a gain of 4.4 percentage points. Brier scores improved from 0.040 to 0.016. Lower Brier scores indicate better calibration, so the drop represents a meaningful tightening of forecasts.
The researchers behind the working paper made a deliberate choice. They anchored every comparison to prices available before a given cutoff rather than allowing post-outcome trading to inflate accuracy numbers. That methodological decision matters. Contracts can continue trading after the result becomes obvious, and including those later prices would paint an overly optimistic picture.
Simply counting winning favorites does not capture how confidently the market held its view. A 51 percent favorite that wins looks identical to a 95 percent favorite on a pure directional score.
That observation feels especially relevant this cycle. Markets correctly anticipated Democratic control of both chambers, yet they underestimated the magnitude of the gains in several places. The combination of high directional accuracy and imperfect magnitude estimates is a pattern worth watching in future contests.
Open Interest and Regulatory Attention Before Voting
In the weeks leading up to the election, combined open interest across major platforms reached roughly 250 million dollars. One platform accounted for the large majority of that figure. Regulators had already signaled they were monitoring election-related contracts for possible manipulation. Public comments from the agency overseeing these markets made clear that surveillance would be active.
At the same time, legislation was introduced that would restrict federal candidates, their spouses, dependent children, and authorized campaign committees from trading contracts tied to their own races. Proposed penalties start at 10,000 dollars or three times any net financial gain, whichever is higher. The proposal has not yet become law, but its appearance reflects growing awareness of the intersection between political campaigns and prediction markets.
In my experience, these kinds of guardrails tend to arrive after volume has already grown large enough to attract official notice. The 2.5 billion dollar figure will only accelerate that conversation.
What the Numbers Suggest About Market Design
Prediction markets thrive when two conditions are present: enough liquidity to make prices meaningful and clear resolution rules that leave little room for dispute. Midterm elections satisfy both criteria for most races. House and Senate contests usually resolve on a single night or within a few days. Gubernatorial races follow the same pattern. That clarity helps traders price information efficiently.
The jump in volume also hints at broader participation. Retail traders who once treated these contracts as a novelty appear to have become more comfortable allocating larger sums. Institutional interest is harder to measure from public data alone, yet the sheer size of the total suggests some professional capital has entered the arena as well.
One limitation remains. Markets still struggle with magnitude. Calling the direction of control is different from forecasting the exact number of seats that will change hands. Traders can be highly accurate on binary outcomes while remaining under-calibrated on the size of the eventual margin. Future market designs may try to address that gap by offering contracts that settle on continuous ranges rather than pure win-or-lose outcomes.
Comparing Markets to Traditional Polls
Outside analysis noted that markets and polls performed at roughly similar overall levels. One firm identified 14 Democratic winners who had been market underdogs a week before the election. That finding undercuts any claim of perfect foresight. At the same time, the rapid updating visible on election night gave markets a practical advantage once votes began to be counted.
Polls freeze at a specific moment. Markets keep absorbing information until the contract settles. That continuous process is both a strength and a potential weakness. It allows prices to reflect late-breaking developments, yet it also means that early prices can look less impressive once the full picture emerges.
I’ve watched enough cycles to know that neither tool is a crystal ball. The value of prediction markets lies less in absolute perfection and more in the real-time probability estimates they provide. When those estimates move sharply, they often signal that something material has changed in the information environment.
Practical Takeaways for Traders and Observers
Several lessons stand out from this cycle. First, volume can grow quickly once a platform gains regulatory clarity and public visibility. Second, directional accuracy at the 98 percent level is impressive but still leaves room for a handful of high-profile misses. Third, the biggest moves often occur after polls close, when actual vote counts begin to override pre-election assumptions.
- Watch open interest in the weeks before major elections for early signals of engagement
- Treat late-night probability spikes as information rather than entertainment
- Remember that magnitude forecasts remain harder than directional calls
- Expect regulatory attention to intensify as dollar volumes climb
These points are not revolutionary. They simply reflect patterns that became clearer once the 2.5 billion dollar figure landed. Anyone following future contests will likely see similar dynamics play out, only at larger scale.
Looking Ahead to the Next Cycle
The next presidential election will almost certainly produce even larger volumes. A contest that includes the White House tends to pull in capital that stays on the sidelines during midterms. If the growth rate observed this year continues, total election-related trading could reach levels that make today’s 2.5 billion dollar mark look modest.
That growth will bring new questions. How should platforms balance retail access with protections against insider trading? What disclosure rules make sense for candidates and their close associates? How much weight should journalists and analysts place on market prices relative to traditional polling aggregates?
None of those questions have settled answers yet. The data from this midterm cycle simply makes them more urgent. Markets have demonstrated both their predictive power and their limitations in the same set of races. The combination is what keeps the conversation interesting.
In the end, the story is less about a single platform posting a big number and more about the quiet maturation of an entire forecasting tool. Two and a half billion dollars is a lot of capital to put behind a set of political outcomes. When nearly 98 percent of those outcomes land the way the market expected, the signal becomes hard to dismiss. The underestimation of the final margins and the handful of favorite failures serve as useful reminders that no system is perfect. Still, the overall performance this cycle sets a high bar for whatever comes next. Traders, analysts, and regulators will all be watching the next set of contracts more closely because of it.
The real test will arrive when the stakes are higher and the information environment is even noisier. Until then, the midterm numbers stand as the clearest evidence yet that prediction markets have moved from experimental curiosity to serious forecasting infrastructure. Volume nearly doubled. Accuracy held near 98 percent. Odds updated in real time as the night unfolded. Those three facts alone are enough to keep the conversation going well into the next election cycle.
