# Market Mechanics & Probability Aggregation Methods ## Extracted from *Prediction Markets Fundamentals* (Luckner et al., 2012) --- ## 1. OPERATIONAL PRINCIPLE (Section 2.4) Prediction markets are a form of financial market where contracts whose payoff depends on the outcome of uncertain future events are traded. Traders buy and sell contracts based on their expectations regarding the likelihood of future events. **Trading prices reflect the traders' aggregated expectations** on the outcome and can be used to predict the likelihood of these events. **Core mechanism:** - A contract like "500-600 cars in 2008" pays €100 if the event occurs, €0 otherwise - If the contract trades at €45, the market-implied probability is 45% - If a trader believes the true probability is 70%, they buy at €45 (undervalued) or sell if above 70€ - The trading mechanism automatically executes matching orders (buy/sell orders that overlap) - Higher perceived probability → higher willingness to pay and reluctance to sell - The trading price thus reflects the group's aggregated belief about event likelihood --- ## 2. THEORETICAL FOUNDATIONS OF INFORMATION AGGREGATION (Section 2.3) **Hayek's insight (1945):** Markets are the most efficient instrument to aggregate dispersed information among traders. Price signals coordinate separate actions of people. **Efficient Market Hypothesis (Fama, 1970):** - **Weak form:** Prices reflect all information contained in historic prices - **Semi-strong form:** Prices reflect ALL publicly available information (accepted paradigm) - **Strong form:** All relevant information known to anyone is reflected in prices **Information aggregation process:** - People infer information from observing other traders' beliefs - They add that information to their own prior beliefs - This converges toward a common knowledge equilibrium (McKelvey & Page, 1990) - Experimental evidence (Plott, 2000): Markets with informed traders converge to correct values; markets with dispersed partial information also successfully aggregate that information --- ## 3. CONTRACT TYPES (Section 3.1) Three primary contract types (Wolfers & Zitzewitz, 2004), each corresponding to a different prediction target: | Contract Type | Payoff | Predicts | |---|---|---| | **Winner-takes-all** | $100 if event happens, $0 otherwise | Probability of event | | **Linear / Index** | $1 per base value point | Mean value of outcome | | **Spread** | Fixed cost; pays 2x if spread condition true, else $0 | Median value of outcome | - **Winner-takes-all:** Most common type. Current price reflects the probability of the event. For mutually exclusive outcomes (e.g., championship winner), sum of all stock prices = $100. - **Linear / Index:** Any positive number can be payoff (e.g., sales/$1M). Trading price corresponds to the mean aggregated belief about the outcome. - **Spread:** Stock price fixed (e.g., $1), but threshold y adjusts based on supply/demand. Price predicts the median expectation. **Unit portfolios:** Used by IEM and others — a fixed set of shares sold/bought as a bundle at a price equal to the sum of payoffs. This is risk-free for the operator. --- ## 4. TRADING MECHANISMS (Section 3.2) Six trading mechanisms are described and compared: ### 4.1 Double Auctions (DA) - **Continuous Double Auction (CDA):** Traders submit limit or market orders into an order book. Matching orders (buy price ≥ ask price) are immediately executed. Most common in PMs and financial markets. Requires sufficient liquidity. - **Call Auction (CA):** Orders collected over a period, then executed at predetermined times (e.g., opening price determination). Used by STOCCER during FIFA World Cup 2006. - **Key property:** No financial risk for operator if set up with unit portfolios. No immediate liquidity guarantee. ### 4.2 Market Scoring Rules (MSR) (Hanson, 2003) - Builds on scoring rules for evaluating forecasters - Successive forecasters adjust the current prediction; paid based on improvement (can be negative) - Modeled as tradeable shares with a continuous price function determined by the scoring rule - **AMM determines price for each share** bought/sold - **Properties:** Limited losses (upper bound known), one parameter to set (controls liquidity), provides immediate liquidity, arbitrage-free, continuous price updates - Also called LMSR (Logarithmic Market Scoring Rule) ### 4.3 Dynamic Pari-Mutuel Market (DPM) (Mangold et al., 2005) - Overcomes the static nature of standard pari-mutuel markets (horse race betting) by introducing **dynamic prices** for shares - Price depends on number of shares in market and a continuous price function - All money redistributed over winning shares - **Critical distinction:** Price does NOT directly reflect probabilities — must be transformed - **Properties:** Limited losses, one parameter (liquidity), immediate liquidity, arbitrage-free - Used by Yahoo! Buzz markets ### 4.4 Dynamic Price Adjustment (DPA) - No continuous price function; offers equal buy/sell price up to a max quantity - New price calculated after each transaction based on last executed trades within a moving window - If purchases dominate, price rises - **Properties:** NOT arbitrage-free (can be exploited), losses NOT bounded, needs 3 parameters - Slower information incorporation than MSR/DPM ### 4.5 Hollywood Stock Exchange (HSX) Mechanism - Orders collected during a "sweep period" (like call auction) - Net-movement balance calculated (buy shares - sell shares) - Multiplied by factor to determine price movement, potentially attenuated by "Virtual Specialist" - New price = old price + price movement - **Properties:** NOT arbitrage-free, losses NOT bounded, needs >3 parameters - Slowest information incorporation - Price certain only at end of sweep period (confusing for traders) ### 4.6 Empirical Comparison Results **Forecasting Accuracy (Mean Absolute Error, lower = better):** 1. DPM: 1.101 (best) 2. LMSR: 1.254 3. DPA: 2.253 4. HSX: 3.233 (worst) **Speed of Information Incorporation (mean periods to reach new price):** 1. DPM: 8.13 (fastest) 2. LMSR: 9.79 3. DPA: 18.24 4. HSX: 29.96 (slowest) **Robustness to Parameter Misspecification:** - DPM and LMSR: error increases <30% when deviating from optimal parameters - DPA: error can increase >100% - HSX: error increases >80% **Noisy Trading Robustness:** - HSX: NOT susceptible to noisy trading (best) - DPA: slightly susceptible - DPM & LMSR: most susceptible to noisy trading --- ## 5. PROBABILITY AGGREGATION — SUMMARY The key mechanisms by which prediction markets aggregate probability estimates: 1. **Price mechanism:** The market price of a winner-takes-all contract directly represents the crowd's probability estimate for the event. Price = implied probability × payout. 2. **Continuous updating:** Any trader who disagrees with the current price can trade, moving the price. All available information is continuously reflected. 3. **Incentive alignment:** Traders profit by being correct, rewarding truthful information revelation rather than preference expression. Even enthusiasts will not distort prices if it would cost them money. 4. **Contract design determines the aggregation target:** - Winner-takes-all → probability - Linear/Index → mean - Spread → median 5. **Automated Market Makers** solve liquidity problems in thin markets: - MSR/LMSR & DPM: continuous price functions, arbitrage-free, bounded losses - DPA & HSX: less sophisticated, arbitrage possible, unbounded losses **Note on DPM:** The price does NOT directly correspond to probabilities — it must be mathematically transformed, creating additional cognitive burden for traders.