Mu and Prediction Markets (Part 5)


I asked ChatGPT about how prediction markets work. The response:

  1. A company creates a prediction market for a specific event, such as the outcome of an election or the success of a product launch.

  2. The company issues contracts representing the possible outcomes of the event. For example, in an election prediction market, there might be contracts representing the victory of each candidate.

  3. Participants in the market can buy and sell these contracts. The prices of the contracts fluctuate based on supply and demand, and can be interpreted as the market’s prediction of the probability of each outcome occurring.

  4. As the event approaches, the prices of the contracts may change based on new information or changing market conditions.

  5. After the event occurs, the contracts are settled based on the actual outcome. Those who hold contracts representing the winning outcome are paid the full value of the contract, while those holding contracts for the losing outcomes receive nothing.

From the introduction of the 2011 book “Prediction Markets: Fundamentals, Designs, and Applications” by Stefan Luckner, Jan Schröder, Christian Slamka, et al: “The basic idea of prediction markets is to trade contracts whose payoff depends on the outcome of uncertain future events. Although the final payoffs of the contracts are unknown during the trading period, rational traders should sell contracts if they consider them to be overvalued and buy contracts if they consider them to be undervalued. Until the outcome is finally known, the trading prices reflect the traders’ aggregated beliefs about the likelihood of the future events. In efficient markets, all the available information is reflected in the trading prices at any time.”

Here is a graphic from the book which explains the working:

There are different types of contracts in prediction markets as this table from a paper on “Prediction Markets for Economic Forecasting” by Erik Snowberg, Justin Wolfers and Eric Zitzewitz shows:

Cultivate Labs writes about the two primary mechanisms to enable trading:

Continuous Double Auction (CDA): A continuous double auction (often abbreviated as CDA) is a mechanism for matching buyers and sellers of a stock. In a CDA, the market maker keeps an order book that tracks bids and asks. If I come along and say that I’d like to buy a share stock A for $5, that is recorded in the order book as a bid for 1 share at $5. On the flip side, if you own a share of stock A and are willing to sell that share for $5, that is recorded as an ask. If the bid & ask for two traders match, like in our example (I want to buy stock A for $5, you want to sell it for $5), then the trade is executed. A continuous double auction is also used in traditional stock markets like the NYSE.

Automated Market Makers: One issue with using a continuous double auction in a prediction market is that liquidity can be a problem. Most prediction markets have far fewer participants than an exchange like the NYSE. If I make a bid for $5 and there is no one out there selling the same stock for $5, then I can’t make my trade. If there’s no one to take the other side of my trade, the market would be said to have low or poor liquidity. To alleviate this problem, platforms use what’s known as an automated market maker. In this setup, the platform acts as the “house,” taking the opposite side of all trades. Doing so ensures that participants are always able to make a trade, effectively creating or “making” the market.

The Logarithmic Market Scoring Rule (LMSR), proposed by Robin Hanson in 2002, has become the de facto market-maker mechanism for prediction markets. As the abstract puts it: “In practice, scoring rules elicit good probability estimates from individuals, while betting markets elicit good consensus estimates from groups. Market scoring rules combine these features, eliciting estimates from individuals or groups, with groups costing no more than individuals. Regarding a bet on one event given another event, only logarithmic versions preserve the probability of the given event. Logarithmic versions also preserve the conditional probabilities of other events, and so preserve conditional independence relations. Given logarithmic rules that elicit relative probabilities of base event pairs, it costs no more to elicit estimates on all combinations of these base events.”

Before we discuss Mu and Prediction Markets, we will take a detour to discuss Superforecasting.

Published by

Rajesh Jain

An Entrepreneur based in Mumbai, India.