Reconciling District Prediction Markets and Seat-Count Buckets in U.S. Midterm Elections

arXiv:2611.13427 · stat.AP · Submitted 2026-11-10 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Statistics. The day in statistics: methods, theory and the inference under them.

Tom: We're looking at the paper: "Reconciling District Prediction Markets and Seat-Count Buckets in U.S. Midterm Elections".

Jane: Here's the gist of the paper. The prediction market platform Kalshi offers individual markets for the winner in each U.S.

Tom: congressional district and an aggregate market for the total number of Democratic House seats, and we develop an applied statistical framework for reconciling these two market layers since the aggregate seat count is determined by the collection of individual district outcomes The framework combines a latent-Gaussian dependence structure with a log-odds adjustment to district probabilities and a power adjustment to aggregate seat-count probabilities, with parameters estimated by divergence minimization Applied to a July 30, 2026 snapshot, the three-parameter primary model achieves close reconciliation, with forward and reverse KL divergences both approximately 0.0013 while yielding a positive latent national correlation, a negative district log-odds shift, and flattening of the aggregate target distribution The framework provides a diagnostic approach for assessing cross-market consistency in hierarchically structured prediction or financial markets Pruning redundant reasoning tokens Prediction markets are trading platforms where participants buy and sell contracts based on the outcomes of future events These markets can represent the same underlying outcomes at different levels of aggregation In the context of U.S.

Jane: congressional elections, for example, an individual district contract concerns the winner of a single congressional race while the aggregate seat-count market concerns the total number of seats won nationally by a political party However, prices across these two types of markets need not exhibit probabilistic consistency even though the underlying contracts depend on the same collection of district election results How the training signal is built Our framework introduces a correlated latent-Gaussian structure, a log-odds adjustment to the input district probabilities, and a power adjustment to the aggregate seat-count probabilities We fit the model by minimizing a divergence measure between the model-implied aggregate distribution, derived from the district inputs and latent dependence structure, and the adjusted aggregate market distribution The snapshot analysis uses the July 30, 2026 snapshot to assess robustness across divergence criteria and explore model extensions while the historical daily analysis examines how market reconciliation evolves over time including periods of relative stability and substantial volatility Where the gains come from The model provides a probabilistic framework that connects individual district markets with the aggregate seat-count market To introduce dependence while preserving the marginal district probabilities, we use a latent-variable framework related to Bayesian probit modeling for binary data and Gaussian copula methods in quantitative finance The parameter σ 2 nat in [0, 1) represents the fraction of unit latent variance allocated to the shared national factor and is also the correlation between the latent scores of any two distinct districts Increasing σ 2 nat shifts latent variance from the district-specific disturbances to the common national factor, thereby increasing dependence while preserving the marginal probabilities π i and the expected seat total E(S) Reconciliation through adjustments The model incorporates three distinct components: a correlated latent-Gaussian structure, an adjustment to input district probabilities, and an adjustment to the market bucket probabilities Fitting the model therefore allows reconciliation through adjustments to both district and aggregate market probabilities rather than relying on district dependence alone The two parameters can therefore compete to capture similar distributional adjustments during optimization Dynamic market behavior The historical daily analysis shows that the fitted parameters exhibit a period of substantial volatility during the spring, approximately from March 27 to May 8 Reconciliation in the earlier period typically involves an upward district log-odds adjustment, whereas the later period involves a downward adjustment The spring window contains all 36 instances in which γ reaches either its lower or upper optimization bound over the entire historical period The snapshot and historical daily results demonstrate how the individual district and aggregate seat-count markets can be closely reconciled using a parsimonious three-parameter model By combining a correlated latent-Gaussian structure with targeted adjustments to both district-level probabilities and aggregate bucket probabilities, we provide a mathematically concrete bridge between two distinct levels of market pricing Beyond its descriptive capabilities, our modeling framework offers potential practical value as a diagnostic tool for analyzing cross-market consistency More broadly, our work develops an applied statistical framework for reconciling probabilities across different levels of aggregation in hierarchical prediction markets References Albert J H and Chib S (Bayesian analysis of binary and polychotomous response data) Ali S M and Silvey S D (A general class of coefficients of divergence of one distribution from another) Byrd R H Lu P Nocedal J and Zhu C (A limited memory algorithm for bound constrained optimization) Cartea A Jaimungal S and Penalva J (Algorithmic and High-Frequency Trading) Chen S X and Liu J S (Statistical applications of the Poisson-binomial and conditional Bernoulli distributions) Cherubini U Luciano E and Vecchiato W (Copula Methods in Finance) Golub G H and Welsch J H (Calculation of Gauss quadrature rules) Kalshi (Democratic house seats) Kalshi (New Jersey’s 7th congressional district election, 2026) Kalshi (Orderbook responses) Predexon (Prediction market data and orderbook history API) Page L and Clemen R T (Do prediction markets produce well-calibrated probability forecasts?) Platt J C (Probabilities for SV machines) Snowberg E and Wolfers J (Explaining the favorite-long shot bias: Is it risk-love or misperceptions?) Thaler R H and Ziemba W T (Anomalies: Parimutuel betting markets: Racetracks and lotteries) Wolfers J and Zitzewitz E (Prediction markets) U S Census Bureau (Geographic terms and definitions)

Title and authors: Jane: So to kick off, we're looking at the title "Reconciling District Prediction Markets and Seat-Count Buckets in U.S. Midterm Elections" and who wrote it—Michael Tan and Han Xiao. It’s a very specific problem they are trying to solve for the prediction market platform Kalshi.

Tom: It sounds like they're focusing on how to bridge the gap between individual district contracts, which are about one race, and the aggregate market that tells you how many seats a party will win nationally.

Lu: The authors describe their framework as combining a latent-Gaussian dependence structure with adjustments for both district probabilities and aggregate seat-count probabilities, which is a specific statistical approach.

Meng: It seems they are using Bayesian probit modeling for the binary data from the districts and Gaussian copula methods, which gives it some strong grounding in modeling dependent events in finance.

Tom: So, when you put it all together, they are aiming to reconcile those two market layers by fitting this model over time to see how well it works across different election outcomes.

Jane: They also use data from Kalshi spanning from January eight through August thirty-one of two thousand twenty-six to test this framework in two main parts: a snapshot analysis and a historical daily analysis.

Tom: The snapshot part uses the July thirty two thousand twenty-six date to check how robust the three-parameter model is across different divergence criteria.

Lu: And the historical daily analysis examines how market reconciliation evolves over time, including periods of relative stability and substantial volatility.

The paper's summary: Jane: So what they are actually summarizing in this paper is that this three-parameter model does achieve close reconciliation when applied to the July thirty two thousand twenty-six snapshot. They show a way to connect the individual district predictions smoothly to the aggregate seat count market using that latent Gaussian structure.

Tom: They demonstrate how combining that structure with targeted adjustments to both district and aggregate market probabilities provides a mathematically concrete bridge between those two distinct levels of market pricing.

Lu: The authors state that fitting this model allows reconciliation through adjustments to both district and aggregate market probabilities rather than just relying on the dependence between the districts alone.

Jane: That means they can capture similar distributional adjustments using either parameter, so they compete to find the best fit for the overall structure when modeling these election outcomes.

Tom: And when they look at their historical daily data, there’s a dynamic behavior we see; specifically during a period from March twenty-seven to May eight of two thousand twenty-six they see substantial volatility.

Meng: What does that volatility actually mean for deployment? Is it unstable pricing or just a phase where the underlying market dynamics are changing rapidly?

Jane: The historical daily analysis shows that the fitted parameters themselves exhibit this period of substantial volatility during the spring, and reconciliation in earlier periods usually involves an upward district log-odds adjustment.

Lu: That specific spring window contains all thirty-six instances where gamma reaches either its lower or upper optimization bound over the entire historical period, which is a concrete marker of that instability.

The paper's improvements: Tom: So to summarize the improvements they suggest with this model, they show that this three-parameter model does achieve close reconciliation when applied to the July thirty two thousand twenty-six snapshot. It uses a latent-Gaussian structure and targeted adjustments for both district and aggregate market buckets.

Jane: They demonstrate how combining that structure with those adjustments provides a mathematically concrete bridge between those two different levels of market pricing, which is what makes the work practical.

Lu: The implication is that this framework offers potential practical value as a diagnostic tool for analyzing cross-market consistency in hierarchical prediction or financial markets.

Meng: It’s useful for diagnosing where things are misaligned, which is important when you’re trying to build a stable system that actually ships reliably to production.

Lalam: The work also develops an applied statistical framework for reconciling probabilities across different levels of aggregation in hierarchical prediction markets, which is a significant step forward in modeling these structures.

Tom: So the snapshot and historical daily results demonstrate how individual district and aggregate seat-count markets can be closely reconciled using this parsimonious three-parameter model.

Jane: It’s a solid piece of work that gives us a way to see how different market tiers interact when they are all based on the same underlying election outcomes.

Lu: The framework provides a concrete statistical method for connecting individual district markets with the aggregate seat-count market using latent variables and adjustments.

Meng: It’s interesting to see how this connects to uncertainty analysis papers we've been looking at, like BayesJudge, because it deals with conflicting judgments across different levels of data structure.

Tom: It certainly does connect that way; it shows that even when you have individual models, you still need a layer to check if they're all coherently aligned.

Conclusion: Jane: So we've been looking at how these prediction markets can be messy because they have two different levels of counting—individual districts versus the total number of seats—and this paper, "Reconciling District Prediction Markets and Seat-Count Buckets in U.S. Midterm Elections," shows a way to make sense of that mess using a three parameter model.

Tom: Exactly; it takes those individual district predictions and connects them smoothly to the aggregate seat count market by using a latent Gaussian structure, which basically models the hidden dependencies between districts better than just looking at them separately.

Lu: The authors fit this model by minimizing a divergence measure between what their aggregate distribution looks like based on the district inputs and this latent dependence structure, and what the actual aggregate market distribution is.

Meng: And they use these adjustments—a log-odds adjustment for the district probabilities and a power adjustment for those aggregate seat count probabilities—to force them to match up properly.

Tom: The snapshot analysis on that July thirty two thousand twenty-six date shows the three parameter primary model actually does achieve close reconciliation, with things like forward and reverse KL divergences being very small, around zero point zero zero one three.

Jane: That means they found a positive latent national correlation in their model, a negative log-odds shift for the districts themselves, and it flattens out the aggregate target distribution nicely.

Lu: The core gain here is providing a probabilistic framework that connects those individual district markets with the aggregate seat-count market by introducing dependence while keeping the individual district probabilities intact.

Meng: It’s a diagnostic tool, which is useful because it lets us see how cross-market consistency holds up when we look at historical data to see if things are stable or volatile.

Tom: The historical daily analysis also showed that the fitted parameters exhibit a period of substantial volatility during the spring, specifically from March twenty-seven to May eight and that window contains all thirty six instances where one of those adjustment parameters hits its lower or upper bound.

Jane: So for someone just listening to the show, this means they’ve built a mathematical bridge between two different ways people price political outcomes at different scales.

Lu: The framework is really about showing how you can adjust both district and aggregate market probabilities at the same time during the optimization process, which is a strong mechanism for capturing distributional adjustments.

Meng: It’s an applied statistical framework, which is good because it shows we can actually use this to analyze cross-market consistency in hierarchical prediction markets, not just theoretical math.

Tom: So that's what we were looking at with "Reconciling District Prediction Markets and Seat-Count Buckets in U.S. Midterm Elections," a solid model for bringing order to those two different levels of betting data.

Michael Tan, Han Xiao

Union County Magnet High School · Department of Statistics, Rutgers University

stat.AP

Submitted: 2026-11-10

Updated: 2026-11-10

Comments: 14 pages, 5 figures

Project page: https://www.census.gov/programs-surveys/popest/about/glossary/geo-terms.html

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 88/100

The gist: The prediction market platform Kalshi offers individual markets for the winner in each U.S.

Key concepts

Correlated Latent-Gaussian Structure
This structure introduces a hidden layer of shared factors influencing all districts simultaneously. It models how the outcomes of individual districts are not independent but are linked through these unobservable, correlated latent variables, which helps capture underlying national trends.
Log-Odds Adjustment to District Probabilities
This adjustment modifies the initial probabilities derived from individual district markets. It accounts for systematic biases or dependencies between districts that a simple model might miss, ensuring the district predictions align better with the overall aggregate outcome.
Power Adjustment to Aggregate Seat-Count Probabilities
This component specifically adjusts the probability distribution of the total number of seats won nationally. It ensures that the aggregate market's pricing reflects the true relationship between individual district results and their combined total, not just a simple sum.

Summary

The prediction market platform Kalshi offers individual markets for the winner in each U. p. 0S. congressional district and an aggregate market for the total number of Democratic House seats, and we develop an applied statistical framework for reconciling these two market layers since the aggregate seat count is determined by the collection of individual district outcomes

The framework combines a latent-Gaussian dependence structure with a log-odds adjustment to district probabilities and a power adjustment to aggregate seat-count probabilities, with parameters estimated by divergence minimization

Applied to a July 30, 2026 snapshot, the three-parameter primary model achieves close reconciliation, with forward and reverse KL divergences both approximately 0.0013 while yielding a positive latent national correlation, a negative district log-odds shift, and flattening of the aggregate target distribution

The framework provides a diagnostic approach for assessing cross-market consistency in hierarchically structured prediction or financial markets

Pruning redundant reasoning tokens

Prediction markets are trading platforms where participants buy and sell contracts based on the outcomes of future events These markets can represent the same underlying outcomes at different levels of aggregation In the context of U.S. congressional elections, for example, an individual district contract concerns the winner of a single congressional race while the aggregate seat-count market concerns the total number of seats won nationally by a political party However, prices across these two types of markets need not exhibit probabilistic consistency even though the underlying contracts depend on the same collection of district election results

How the training signal is built

Our framework introduces a correlated latent-Gaussian structure, a log-odds adjustment to the input district probabilities, and a power adjustment to the aggregate seat-count probabilities We fit the model by minimizing a divergence measure between the model-implied aggregate distribution, derived from the district inputs and latent dependence structure, and the adjusted aggregate market distribution The snapshot analysis uses the July 30, 2026 snapshot to assess robustness across divergence criteria and explore model extensions while the historical daily analysis examines how market reconciliation evolves over time including periods of relative stability and substantial volatility

Where the gains come from

The model provides a probabilistic framework that connects individual district markets with the aggregate seat-count market To introduce dependence while preserving the marginal district probabilities, we use a latent-variable framework related to Bayesian probit modeling for binary data and Gaussian copula methods in quantitative finance The parameter σ 2 nat in [0, 1) represents the fraction of unit latent variance allocated to the shared national factor and is also the correlation between the latent scores of any two distinct districts Increasing σ 2 nat shifts latent variance from the district-specific disturbances to the common national factor, thereby increasing dependence while preserving the marginal probabilities π i and the expected seat total E(S)

Reconciliation through adjustments

The model incorporates three distinct components: a correlated latent-Gaussian structure, an adjustment to input district probabilities, and an adjustment to the market bucket probabilities Fitting the model therefore allows reconciliation through adjustments to both district and aggregate market probabilities rather than relying on district dependence alone The two parameters can therefore compete to capture similar distributional adjustments during optimization

Dynamic market behavior

The historical daily analysis shows that the fitted parameters exhibit a period of substantial volatility during the spring, approximately from March 27 to May 8 Reconciliation in the earlier period typically involves an upward district log-odds adjustment, whereas the later period involves a downward adjustment The spring window contains all 36 instances in which γ reaches either its lower or upper optimization bound over the entire historical period

Conclusion

The snapshot and historical daily results demonstrate how the individual district and aggregate seat-count markets can be closely reconciled using a parsimonious three-parameter model By combining a correlated latent-Gaussian structure with targeted adjustments to both district-level probabilities and aggregate bucket probabilities, we provide a mathematically concrete bridge between two distinct levels of market pricing Beyond its descriptive capabilities, our modeling framework offers potential practical value as a diagnostic tool for analyzing cross-market consistency More broadly, our work develops an applied statistical framework for reconciling probabilities across different levels of aggregation in hierarchical prediction markets

References

Albert J H and Chib S (Bayesian analysis of binary and polychotomous response data) Ali S M and Silvey S D (A general class of coefficients of divergence of one distribution from another) Byrd R H Lu P Nocedal J and Zhu C (A limited memory algorithm for bound constrained optimization) Cartea A Jaimungal S and Penalva J (Algorithmic and High-Frequency Trading) Chen S X and Liu J S (Statistical applications of the Poisson-binomial and conditional Bernoulli distributions) Cherubini U Luciano E and Vecchiato W (Copula Methods in Finance) Golub G H and Welsch J H (Calculation of Gauss quadrature rules) Kalshi (Democratic house seats) Kalshi (New Jersey’s 7th congressional district election, 2026) Kalshi (Orderbook responses) Predexon (Prediction market data and orderbook history API) Page L and Clemen R T (Do prediction markets produce well-calibrated probability forecasts?) Platt J C (Probabilities for SV machines) Snowberg E and Wolfers J (Explaining the favorite-long shot bias: Is it risk-love or misperceptions? p. 1) Thaler R H and Ziemba W T (Anomalies: Parimutuel betting markets: Racetracks and lotteries) Wolfers J and Zitzewitz E (Prediction markets) U S Census Bureau (Geographic terms and definitions) p. 1

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Model reconciliation framework for hierarchical prediction markets: The improved system can reconcile individual district prediction market prices with aggregate seat-count market probabilities using a latent-Gaussian structure, a log-odds adjustment to district probabilities, and a power adjustment to aggregate seat-count probabilities.

  2. Diagnostic tool for cross-market consistency: The system can be used as a diagnostic tool for analyzing cross-market consistency by fitting the model over time to characterize periods of relative alignment or misalignment between individual and aggregate markets.

  3. Dynamic market response analysis: The improved system can track how different market tiers respond to macro-level shocks by examining shifts in fitted parameters and divergence values, such as the substantial fluctuations observed during the spring of 2026.

Related papers