# Prediction Markets Theory and Applications — Structured Extraction

**Source:** Leighton Vaughan Williams (ed.), *Prediction Markets: Theory and Applications*, Routledge International Studies in Money and Banking, 2011.
**Extraction Date:** 2026-05-30
**Scope:** 283 pages, 16 chapters covering theory, mechanisms, case studies, and practical implementation

---

## 1. HOW PREDICTION MARKETS ARE CONSTRUCTED AND OPERATED

### 1.1 Core Concept

A prediction market offers contingent contracts whose payoff is tied to the outcome of an uncertain event. Traders buy/sell these contracts; the market price reflects the collective probability assessment. A contract that pays $1 if event X occurs and $0 otherwise will trade at price $p, where p is the market's consensus probability of X.

### 1.2 Market Mechanisms (Chapter 4 — Yiling Chen)

Three desirable properties for prediction market mechanisms:
- **Liquidity** — participants must be able to find counterparties and trade whenever they want
- **Expressiveness** — freedom to express information, e.g. via combinatorial betting languages
- **Bounded budget** — the market institution's potential loss must be capped

#### Auctioneer Mechanisms (risk-free matching)

**Call Markets**: Orders are batched and matched at pre-specified times. A market-clearing price is determined where demand = supply. Uses k-double auction rules. Provides more stable prices in thin markets than CDA, but delays price discovery.

**Continuous Double Auctions (CDA)**: Orders matched in real-time. Maintains an order book with highest bid < lowest ask (bid-ask spread). Used by IEM, Intrade, NewsFutures, HP and Google internal markets. Suffers from thin market problems when few traders exist.

**Generalized Call Mechanisms**: Perform multilateral matching across different contracts. Enable combinatorial prediction markets (e.g., betting on Boolean combinations of outcomes). The order-matching problem is modeled as a linear integer program.

#### Pari-Mutuel Markets

All bets pooled; winners split the pool proportionally. Provides infinite liquidity (participants always can bet). Payoff per share is not fixed until pool closes, creating last-minute betting incentives.

#### Automated Market Maker Mechanisms

Solve the no-trade theorem problem by providing a positive-sum game for participants. The market maker bears risk.

**Market Scoring Rules (MSR)** — Hanson's mechanism:
- Starts with initial probability estimate r⁰
- Participants can change the estimate at any time
- Pay scoring rule payment for old estimate, receive payment for new estimate
- Proper scoring rules (logarithmic, quadratic) ensure truthful reporting is incentivized
- Worst-case loss is bounded: LMSR max loss = b·log(n); QMSR max loss = b(n-1)/n

**Cost Function-Based Market Makers** (equivalent to MSR):
- Cost function C(q) tracks total money spent as function of shares held
- Instantaneous price of contract i = ∂C(q)/∂q_i
- Need p_i(q) ≥ 0 for all i and Σp_i(q) = 1 (no arbitrage)

**Logarithmic Market Scoring Rule (LMSR)** cost function:
```
C(q) = b · log( Σ e^(q_i / b) )
```
Where b controls the liquidity/price-sensitivity parameter. Higher b = more liquidity (price moves less per unit trade). Used by Inkling Markets and Microsoft's internal prediction markets.

**Dynamic Pari-Mutuel Markets (DPM)** — hybrid of pari-mutuel and CDA:
- Cost function: C(q) = K · √( Σ q_i² )
- Price function: p_i(q) = q_i / Σ q_j
- Payoff per share for correct outcome k: (K · √(Σ q_i²)) / q_k
- Provides infinite liquidity; prices vary dynamically
- Used by Tech Buzz Game

### 1.3 Contract Types

| Type | Description | Payoff |
|------|-------------|--------|
| Binary / Winner-take-all | Predict yes/no outcome | $1 if event occurs, $0 otherwise |
| Proportional share | Predict continuous variable | Pays proportionally to outcome value |
| Index | Predict magnitude | Pays based on value relative to threshold |
| Spread | Predict margin of victory | Pays if margin exceeds/falls short |
| Combinatorial | Multiple events combined | Boolean formula determines payoff |

### 1.4 Market Operation Steps

1. **Define the event** — unambiguous, objectively verifiable, with clear end date
2. **Write contracts** — mutually exclusive and exhaustive outcomes
3. **Seed the market** — initial probability estimate (e.g., $0.50)
4. **Select participants** — open public, restricted expert, or corporate employees
5. **Fund endowments** — real money, play money, or virtual currency with prizes
6. **Set trading mechanism** — CDA, call market, automated market maker, or pari-mutuel
7. **Monitor and resolve** — settle contracts against objective criteria at end date

---

## 2. REAL-WORLD CASE STUDIES

### 2.1 Iowa Electronic Markets (IEM)

**What**: Academic prediction market at University of Iowa (1988-present).
**Focus**: US presidential elections, congressional races, other political events.
**Key Results**:
- 1988 presidential election: Proportional share markets predicted vote shares within **0.1%** (Bush) and **0.2%** (Dukakis) — closer than any major polling agency
- 1992 three-way race: Average absolute error of **0.2%** across three candidates
- Markets consistently beat opinion polls and political pundits
- **Factors explaining accuracy variance** (Berg et al., 2008):
  1. Presidential markets perform better than lower-profile elections
  2. Markets with more volume near election perform better
  3. Markets with fewer contracts (candidates/parties) predict better
- Architecture: CDA with real money (but small stakes, up to $500)

### 2.2 TradeSports / Intrade

**What**: Real-money online prediction market, primarily sports then expanded to politics/finance.
**Timeline**: Founded ~2000; closed US sports operations Nov 2008 due to UIGEA; continued non-sports via Intrade.
**Key Features**:
- Binary options paying $0 (loss) or $100 (win) per 10-contract lot
- In-play sports betting — unique for testing real-time information processing
- Massive archived data enabling research on price efficiency and trader behavior
- Notable contract dispute: North Korean missile test — DOD never issued statement, contract expired at $0 despite media reports
**Legal Demise**: US Unlawful Internet Gambling Enforcement Act (UIGEA) 2006 prohibited fund transfers to gambling sites. Founder of BetOnSports.com sentenced to 4 years prison (2009).

### 2.3 Google Internal Prediction Markets

**What**: Largest known corporate prediction market experiment (2005-2007+).
**Scale**: Thousands of internal markets deployed.
**Focus**: Product launch dates, project deadlines, business outcomes.
**Key Findings** (Cowgill et al., 2009):
- Evidence of optimism bias (especially in newer recruits)
- Markets became less biased over time as collective trading experience increased
- Markets reasonably efficient
- Google employees motivated more by **status** (leaderboards, T-shirts) than cash — "nobody noticed the $15 we gave people, but everyone notices T-shirts"
- Play money with prizes was sufficient incentive

### 2.4 Hewlett-Packard Internal Markets

**What**: Early corporate pioneer (late 1990s).
**Focus**: Printer sales forecasting.
**Key Result**: Information markets beat forecasts from HP's own sales department (the "experts").
**Mechanism**: CDA-based internal platform called BRAIN.

### 2.5 Taiwan XFuture

**What**: Exchange of Future Events, run by National Chengchi University (2006-present).
**Scale**: 14,938 futures, 2,115 events, 1.98M submissions, 98,000 matches, 240M trading volume. Registered traders from 121 countries.
**Architecture**: Futures market with margin-based trading (unique design). Margin calculated on worst-case assumption.
**Types**: Winner-take-all and share markets. Topics: politics, economics, finance, Cross-Strait affairs, sports, entertainment.
**Hit Rate**: **87.79%** across 172 political events with 575 futures contracts (151/172 correct).
**Taiwan Experience** (2010): "An overall accuracy rate of 87.79% indicates that it works in Taiwan as well as it works in other places."
**Notable**: Also includes TAIPEX (Taiwan Political Exchange, 2003) and AI-ECON FX (founded 2006).

### 2.6 Hollywood Stock Exchange (HSX)

**What**: Play-money prediction market for movie box office.
**Scale**: ~1.8M registered users.
**Focus**: Opening weekend box office, Oscar winners.
**Results**: More accurate than film critics. Play-money predictions comparable to real-money markets for movies (Gruca et al., 2008).

### 2.7 Other Corporate Users

| Company | Use Case | Source |
|---------|----------|--------|
| **General Electric** | "Imagination Markets" — idea evaluation | Spears et al., 2009 |
| **Microsoft** | Internal prediction markets | Chapter 5 |
| **Best Buy** | Sales forecasting with diverse participants | Dvorak, 2008 |
| **Eli Lilly, Pfizer, Novartis, GSK** | Pharmaceutical R&D forecasting | Chapter 5 |
| **Siemens, Intel, IBM, Yahoo, France Telecom** | Experimented with prediction markets | Graefe, 2009 |

### 2.8 NewsFutures

**What**: Play-money prediction market platform.
**Compared to**: TradeSports (real money) across sports and non-sports events.
**Finding**: Play money performed as well as real money for sports events; real money marginally better for non-sports events (Rosenbloom & Notz, 2006).

### 2.9 Experimental Lab Markets

**Plott and Sunder (1982, 1988)**: Foundational experiments showing markets can aggregate dispersed information into accurate prices under rational expectations.
**Forsythe et al. (1992)**: Identified **marginal traders** (22 out of 192) as key drivers of pricing accuracy.
**Rietz (2005)**: Laboratory evidence of bubbles and no-arbitrage violations in prediction markets, even with known probabilities.
**Hanson et al. (2006)**: Manipulation attempts in lab markets failed — non-manipulators step up as counterparties, neutralizing distortions.

### 2.10 Notable Failures and Challenges

- **Berlin state election (1999)**: 10-day manipulation episode (Hansen et al., 2004) — exception, not norm
- **Thin market failures**: Low-liquidity contracts produce unreliable prices
- **Information mirages**: Markets may price as if insiders exist when none do (Camerer & Weigelt, 1991)
- **Bubble episodes**: Experimental asset markets show bubbles are a legitimate concern
- **Legal/regulatory barriers**: UIGEA (US) forced TradeSports closure; gambling laws restrict real-money markets
- **Insider trading concerns**: Managers reluctant to use markets for major strategic decisions

---

## 3. THEORETICAL FRAMEWORK

### 3.1 Hayek Hypothesis (Hayek, 1945)

Markets aggregate dispersed private information through the price mechanism. The price becomes a **sufficient statistic** — it summarizes all available information without requiring any single individual to possess all the knowledge. This is the foundational theoretical justification for prediction markets.

### 3.2 Efficient Markets Hypothesis

Market prices fully reflect all available information. In prediction markets:
- **Weak form**: Current prices reflect all historical price data
- **Semi-strong form**: Prices reflect all publicly available information
- **Strong form**: Prices reflect all information, including private/insider knowledge
Testing EMH in prediction markets avoids the **joint hypothesis problem** (true value is revealed when the event resolves), making prediction/sports markets ideal laboratories.

### 3.3 Wisdom of Crowds (Surowiecki, 2004; Page, 2007)

Four conditions for a wise crowd:
1. **Diversity** of opinion (each person has private information)
2. **Independence** (opinions not dominated by others)
3. **Decentralization** (people draw on local knowledge)
4. **Aggregation** (mechanism to turn judgments into collective decision)

Prediction markets satisfy all four: they incentivize diverse participation, enable anonymous independent trading, aggregate via price mechanism.

Scott Page demonstrated formally that the wisdom of crowds depends not only on individual accuracy but also on **cognitive diversity** — diverse problem-solving approaches improve collective prediction.

### 3.4 Information Aggregation

**Key insight**: Markets weight opinions by conviction (how much money someone puts at stake) rather than equally as in polls. This solves the problem of identifying who has genuine expertise.

**No-trade theorems** (Milgrom & Stokey): Rational risk-neutral agents should not trade in zero-sum markets because a willingness to trade signals private information. Automated market makers solve this by providing a positive-sum environment where the market maker subsidizes trading.

**Information mirages**: Markets may falsely reveal information not actually held by any trader — price paths driven by inferences from others' trades rather than genuine information.

### 3.5 Proper Scoring Rules

A scoring rule S = {s₁(r), ..., sₙ(r)} is **strictly proper** if a risk-neutral expert maximizes expected score by reporting truthful probabilities. The general form:
```
s_i(r) = G(r) - Σ r_j · G'_j(r) + G'_i(r)
```
where G(r) is any bounded strictly convex function.

**Logarithmic**: s_i(r) = a + b·log(r_i)
**Quadratic**: s_i(r) = a + 2b·r_i - b·Σ r_j²

Market scoring rules convert proper scoring rules into sequential trading mechanisms where each participant pays the previous probability estimate and receives the new one.

### 3.6 Favorite-Longshot Bias

Systematic tendency for longshots (low-probability outcomes) to be overpriced and favorites (high-probability outcomes) to be underpriced relative to true probabilities. Explained by:
- Risk-preference (bettors overweight small chances of large payoffs)
- Utility-maximizing behavior under CRRA preferences
- Information asymmetries

### 3.7 Manipulation Resistance

Both theory (Hanson & Oprea, 2009) and experiment (Hanson et al., 2006; Rhode & Strumpf, 2004) suggest manipulation is very difficult:
- Manipulators create profit opportunities for informed traders who correct prices
- Manipulation may *aid* accuracy by increasing attention and liquidity
- Effects are at most short-term

---

## 4. PRACTICAL IMPLEMENTATION DETAILS

### 4.1 Contract Design Rules

1. **Unambiguous definition**: "Will WMD be found in Iraq by date Z?" not "Are WMD in Iraq?"
2. **Objective resolution authority**: Pre-specify the source (e.g., government agency, official statistic)
3. **Clear end date**: Long-term markets are less motivating
4. **Mutually exclusive, exhaustive outcomes**: All possible states covered
5. **No mid-course redefinition**: Ortner (1998) describes a software delivery market that broke when the client changed the deadline

### 4.2 Liquidity & The LMSR Parameter

The LMSR cost function C(q) = b·log(Σ e^(q_i/b)) has a critical parameter **b** that controls liquidity:
- Higher b = price changes more slowly per unit trade (more liquidity)
- Lower b = price more responsive to trades
- Max loss = b·log(n) for n outcomes
- **Practical rule**: Choose b based on expected trading volume and acceptable worst-case loss

### 4.3 Participant Selection

| Strategy | Used By | Rationale |
|----------|---------|-----------|
| Open to public | IEM (public markets) | Maximize diversity and information |
| Restricted experts | IEM Health Markets | Domain-specific accuracy |
| Employee-wide | Best Buy, Google | Cognitive diversity across functions |
| Selective invitation | GE Imagination Markets | Controlled participation |

**Paradox**: Uninformed ("noise") traders may be **necessary** — they provide cover for informed traders to profit, solving the no-trade problem. But too much noise can harm efficiency.

### 4.4 Incentive Design

| Incentive Type | Examples | Effectiveness |
|----------------|----------|---------------|
| Real money | TradeSports, IEM (private) | Gold standard but legally restricted |
| Play money + prizes | Google, NewsFutures | Often as effective as real money |
| Play money only | Hollywood Stock Exchange | Sufficient for high-interest topics |
| Status/leaderboards | Google | Surprisingly effective (T-shirts > $15) |

**Key finding**: Play money markets can perform as well as real money markets (Servan-Schreiber et al., 2004; Gruca et al., 2008), especially for high-engagement topics.

### 4.5 Thin Market Mitigation

Solutions to low-liquidity problems:
1. **Call markets** (batch orders) — more stable prices than CDA
2. **Automated market makers** (LMSR) — always available to trade
3. **Dynamic pari-mutuel markets** — infinite liquidity by design
4. **Price correction techniques** — wavelet smoothing, median filters (Chen & Wu, 2009)
5. **Seed money** — market maker provides initial liquidity

### 4.6 Insider Trading & Confidentiality

**Concerns for corporate markets**:
- Employees who know non-public info become legal "insiders"
- If they trade, it constitutes insider trading in company stock
- Sensitive information may leak to competitors
- Market prices visible to all employees can affect morale (Pygmalion effect)

**Mitigations**:
- Use play money only (reduces legal exposure)
- Mask/obscure aggregate forecasts
- Restrict market to trusted participants
- Use "lite" versions (e.g., prediction polls instead of full markets)

### 4.7 Factors Influencing Forecast Accuracy

From CMXX movie exchange study (Skiera & Spann):
- **Price volatility** (endogenous): High volatility → high forecast error. Removing top 20% volatile movies reduced MAPE from 211.9% to 97.13%
- **Number of screens** (exogenous): Low distribution → high error. Removing bottom 20% low-screen movies reduced MAPE to 69.79%
- **Trading volume**: Higher volume → more accurate predictions
- **Event profile**: High-profile events → more accurate predictions
- **Number of options**: Fewer contracts → better predictions

### 4.8 Recommended Architecture for Building a Prediction Market Today

1. **Choose mechanism**: LMSR cost-function market maker (most practical)
   - Pros: Always liquid, bounded loss, no thin market problems
   - Cons: Market maker subsidizes trading, b parameter tuning requires care
2. **Set b parameter**: Based on expected number of traders and acceptable subsidy
3. **Implement cost function**: C(q) = b·log(e^(q₁/b) + e^(q₂/b) + ...)
4. **Compute prices**: p_i = e^(q_i/b) / Σ e^(q_j/b)
5. **Handle trades**: Payment = C(q_new) - C(q_old); negative = sale proceeds to trader
6. **Define contracts**: Binary (winner-take-all) for yes/no events; proportional for continuous outcomes
7. **Seed initial state**: Start with q = (0, 0, ..., 0) or pre-purchase shares to set initial prices
8. **Resolve**: On event resolution, pay $1 per share for correct outcome, $0 for others

### 4.9 Legal and Regulatory Landscape (circa 2011)

- **United States**: UIGEA (2006) restricts real-money betting. IEM operates under CFTC no-action letter. Intrade continued accepting US bets from outside US. TradeSports forced to close US sports operations.
- **European Union**: More permissive. Betfair (UK) operates as largest prediction exchange with 2M+ members.
- **Asia**: Taiwan's XFuture operates freely. Japan's General Election Hatena (2005). New Zealand's ipredict (2008).

### 4.10 Barriers to Mainstream Adoption (Chapter 5 — Croxson)

Despite theoretical promise and empirical evidence, as of 2011:
- Only **9% of executives** reported prediction markets deployed in their organizations (McKinsey 2009 survey)
- Most usage described as "evaluating or running limited trials"
- Key barriers: legal concerns, insider trading fears, sensitivity of information, lack of clear evidence vs. alternatives, organizational resistance, need for unambiguous contract design

---

## Key Referenced Works

| Author(s) | Year | Contribution |
|-----------|------|-------------|
| Hayek | 1945 | "The Use of Knowledge in Society" — price as information aggregation mechanism |
| Plott & Sunder | 1982, 1988 | Foundational lab experiments on information aggregation in markets |
| Forsythe et al. | 1992 | IEM success; marginal trader theory |
| Hanson | 2002 | Market Scoring Rules (LMSR) |
| Wolfers & Zitzewitz | 2004 | Survey of prediction markets; price biases |
| Berg et al. | 2008 | 12 years of IEM research; accuracy factors |
| Cowgill et al. | 2009 | Google internal markets study |
| Arrow et al. | 2007 | Statement calling for regulatory accommodation of prediction markets |
| Chen | 2011 | Comprehensive mechanism taxonomy (Chapter 4) |

---

## Cross-References to Other Knowledge Base Entries

- See `/root/hermes-knowledge/extractions/market-making-lmsr.md` for detailed LMSR implementation guide
- See `/root/hermes-knowledge/extractions/information-aggregation-theory.md` for deeper theoretical treatment
- See `/root/hermes-knowledge/extractions/betting-exchange-mechanics.md` for Betfair-specific implementations