How does MispriceHQ produce its probability estimates?

For every contract, we price from scratch using ~140 features — rate curves, options-implied paths, macro prints, polling, on-chain crypto flows, and platform-level liquidity. The market price is never used as a model input.

How do you measure model accuracy?

We score every prediction against the realized outcome using Brier scores and reliability diagrams. We compare against the market-consensus baseline on a per-category basis.

Are these gambling recommendations?

No. MispriceHQ publishes independent probability estimates for educational and research purposes. We do not take positions and do not route orders. Event contracts on platforms like Kalshi are CFTC-regulated derivatives — they carry risk of total loss.

← Back METHODOLOGY · CALIBRATION

How we know
we're honest.

Every probability on this site is graded against what actually happened. The categories we beat the market in are shown in blue. The ones we don't — yet — are shown honestly grey.

MARKETS RESOLVED

1,072

OVERALL BRIER

0.201

VS MARKET

-0.017

RELIABILITY DIAGRAM

N = 1,072 · TRAILING 12 MO

X-axis: what the model predicted. Y-axis: what actually happened. Perfect calibration sits on the dashed 45° line. Bubble size scales with sample count per bucket.

HOW IT WORKS, IN PLAIN ENGLISH

1 · Independent priors

For every contract, we price from scratch using ~140 features — never from the market price itself.

2 · Disagreement is signal

The gap between our prior and the market price is the edge. Sign tells direction, magnitude tells conviction.

3 · Honesty by construction

Every model is scored against realized outcomes. If we have no edge in a category, the UI shows it grey.

SCORE BY CATEGORY

Lower Brier is better. Categories where our score beats the market consensus are highlighted.

TRAILING 12 MO

Category	Model Brier	Market Brier	Δ	N	Verdict
Macro	0.182	0.211	-0.029	312	● EDGE
Crypto	0.196	0.224	-0.028	248	● EDGE
Tech	0.218	0.222	-0.004	174	○ NO EDGE
Politics	0.209	0.215	-0.006	142	○ NO EDGE
Culture	0.226	0.234	-0.008	100	○ NO EDGE
Sports	0.241	0.232	+0.009	96	○ NO EDGE — WORSE

Sports is currently worse than the market. We leave it visible on purpose — when we don't have edge, the UI says so. We are not running models on sports markets in production.

DATA SOURCES

FRED — macro time series rate curves, CPI, unemployment, PCE

BLS / BEA public releases labor & GDP first-prints

CME options chain Fed funds futures, implied path

Spot & on-chain crypto feeds ETF flows, exchange balances

Public polling aggregators topline + crosstabs, weighted

Platform order books Polymarket, Kalshi, Manifold

MODEL CARD

ARCHITECTURE GBM ensemble + isotonic calibration

FEATURES ~140

REFIT CADENCE NIGHTLY · 03:30 UTC

HOLDOUT 30-DAY ROLLING

NEVER USED AS INPUT MARKET PRICE

NOT FINANCIAL
ADVICE.

MispriceHQ publishes independent probability estimates for educational and research purposes. Models can be wrong; market microstructure can shift overnight; tails do tail. We don't take positions, we don't route orders, and we are not your fiduciary. Risk of total loss is real and material.

Event contracts on Kalshi are regulated derivatives traded under CFTC oversight; other venues operate under their own jurisdictions. By using this site you acknowledge our methodology, our calibration, and that the numbers above describe the past — not the future.

How we knowwe're honest.

How we know
we're honest.