5D Library › Prediction Market Microstructure, Bregman Information Projection and Cross-Book Arbitrage Bounds · 中文

Preface: Prices, Probabilities and Capital

Note

This is not investment, trading or betting advice; please read this first. This book studies the mathematics of prediction-market prices. It does not advise anyone to take part in a prediction market, betting or any trade, does not assess any platform, and does not discuss regulation in any jurisdiction. Apart from the published research it cites, all data are synthetic, generated by the programs printed in the book. Its fee formulas, order books and costs of capital are illustrative, not any platform’s terms.

Prediction markets are often described as “a readout of the wisdom of crowds”: a contract at 0.20 means the event has a 20 % chance. That reading is useful in many settings, but it squeezes three different things into one number: how traders’ beliefs aggregate into a price; how the market mechanism pays for information; and how much room a price has to depart from a probability when capital is locked up. This book separates the three, writes each as a runnable model, and finds where “price = probability” holds, where it fails, and how much of an apparent arbitrage can actually be executed.

Like the other books of 5D Library, it falsifies first, then sets bounds, and reports negative results honestly — including where its own models need more data than exist to be tested.

What the book covers

Chapter 1 uses the models of Manski and of Wolfers and Zitzewitz to show that a price is a quantile or a weighted mean of the belief distribution, and that dispersed beliefs and bounded prices are enough to overprice longshots. Chapter 2 builds Hanson’s logarithmic market scoring rule, checks its duality with the softmax, KL divergence and Bregman projection, and works out how the liquidity parameter trades subsidy against slippage. Chapter 3 builds the no-arbitrage band of capital lock-up and explains why near-certain long-dated contracts sit at 0.92 to 0.96. Chapter 4 computes the executable arbitrage in a book of exclusive outcomes: walking the book, fees, lock-up and stale quotes. Chapter 5 puts logically related contracts into a convex polytope and uses a hand-written Frank-Wolfe projection to find arbitrage certificates. Chapter 6 asks how much data it takes to test and correct a price bias. Chapter 7 reads the Guanzi’s “the clever gain tenfold” and the Shiji’s Bai Gui, “what others discard I take, what others take I let go”, as information rent and the cost of carry. Chapter 8 gives a four-step check and a pre-registered test of the lock-up discount.

This book borders several other books of 5D Library and recalls them only briefly where they overlap: Book 19 treated the Kelly criterion and non-ergodicity; Book 26 probability judgement, Brier scores and the wisdom of crowds; Book 33 pre-registration and decision rules (which Chapter 8 follows); Book 39 order-book depth exhaustion and cointegration arbitrage; Book 45 order-book microstructure and adverse selection; Book 49 information geometry and the KL divergence. This book does not repeat them.

Claims corrected on checking

  1. A price is not the mean belief. With risk-neutral traders and equal budgets, the price is the quantile at which the share of traders above it equals it: a population whose mean belief is 0.100 trades at 0.167 (p01). Under log utility the price is exactly the wealth-weighted mean, 0.0989; when risk attitude goes with belief it can move to 0.1444 (p02).
  2. The favourite-longshot bias needs no one who loves risk. With unbiased but dispersed beliefs alone, contracts priced between 0.05 and 0.10 average 0.084 against a true probability of only 0.024 (p03).
  3. Every LMSR trade is an information projection. A trader’s expected gain is exactly bb times a KL divergence, 3.3614 in the example; the market maker’s worst-case loss is exactly bln⁡nb \ln n, 69.31 with two outcomes (p04). Subsidy times price impact tends to 34.66, which does not depend on bb (p05).
  4. Near-certain long-dated contracts sitting at 0.92 to 0.96 is not a failure. With a true probability of 0.99 and a 14.5 % cost of capital, the lower edge of the no-arbitrage band is 0.9252 at six months and 0.8646 at one year (p06); the same mechanism lets the favourite-longshot bias survive in long-dated contracts (p07).
  5. A “Dutch book” whose asks sum to less than 1 is only a ceiling. In the example the best case is +7.0 % a year, and with a fee of 0.07P(1−P)0.07\,P(1-P) every size loses (p08); arbitrage created by stale quotes loses money on average when executed leg by leg (p09).
  6. Incoherence among related contracts can be proved from the prices alone. Projection yields a portfolio that gains in all 8 states, but plain Frank-Wolfe is still unconverged after 400 steps and its certificate bound fails; the pairwise variant converges exactly (p10). Pairwise logical rules catch only 92.9 % of incoherent quotes (p11).
  7. The bias is small enough to be hard to confirm. With 1,000 independent contracts a test has a power of only 0.575 (p12); with a within-event correlation of 0.3, 1,000 contracts are worth 139 independent ones (p13); a flexible correction makes forecasts worse at every sample size (p14).
  8. “Tenfold” and “what others discard I take” are two different sources of profit. This is an interpretation of the classical texts, written as a model and a hypothesis, and the quotations have been checked against the standard punctuated editions (p15).

How to read the boxes

Marked boxes separate published evidence, runnable Python calculations (16 programs, standard library only, fixed random seeds, each actually run at build time with its real output printed beneath it), classical sources, notes and the author’s hypotheses; every hypothesis states what result would show it wrong.

Disclosure

The author uses the works of 5D Library, including this book, as the methodological basis of his community courses, the 5D Ascension Cards (Dimensional Shift Card Decks) and the Yafu AI knowledge base, and therefore has a commercial interest in the subject of this book. The book promotes no course, card deck, software, exchange, prediction-market platform, signal service or fund, provides no link to or recommendation of any trading platform, and accepts no sponsorship from any financial firm or platform. It is not investment, trading or betting advice. The test in Chapter 8 uses public archive data only, is to be run and analysed by people independent of any trading decision, and its code and protocol are published openly for anyone to check.

Fan, Y. (2026). Prediction Market Microstructure, Bregman Information Projection and Cross-Book Arbitrage Bounds: Logarithmic Market Scoring Rules, Favorite-Longshot Calibration Bias, Lock-Up Opportunity Cost and Frank-Wolfe Combinatorial Pricing [预测市场微观结构、布雷格曼信息投影与跨盘口无套利边界] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090166 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).