Comparison of prediction platforms and prediction markets

Prediction markets and prediction platforms differ in how they elicit forecasts from users. Generally, prediction markets use some trading mechanism and users are rewarded for accuracy by earning some real or token money. Usually, in this they are competing against other users.

With prediction platforms, users make independent forecasts which are independently evaluated against the observed outcome. They are not rewarded by making profitable trades, but rather receive scores or points depending on how accurate their predictions are.

Both prediction markets and prediction platforms may or may not involve monetary incentives. Markets can either operate based on tokens or on real money. For prediction platforms, points or scores that forecasters can earn may translate into monetary rewards through prizes awarded to good forecasters.

Advantages of prediction markets

 * The current price is completely open and transparent, so anyone watching the market knows immediately what the market thinks
 * Users earn money/points in proportion to how much information they provide. There is no incentive to just copy the current market prize
 * It is straightforward to switch from token money to real money incentives
 * In theory, large money markets should provide excellent information

Disadvantages of prediction markets

 * Markets are usually zero-sum (if someone gains then another one loses). There is therefore an incentive not too share information that goes beyond that provided through the fact that one is buying / selling
 * Not trivial to allow trading (i.e. making forecasts) on continuous quantities such as future GDP
 * Little incentive to invest in long-term markets, because money will be tied up for a long time. In play-money markets this can be solved by offering participants interest-free loans of tokens they can bet on a question
 * The spread between bid and ask may be large, as rational investors would require a high risk-premium to be willing to risk losing all their assets
 * Markets may need large subsidies to attract sufficient interest

Advantages of prediction platforms
- Potentially fewer incentive problems with long-term questions, assuming that forecasters are more intrinsically motivated

- Easy to incorporate continuous forecasts

- Easy to integrate existing forecasting efforts in different domains, e.g. epidemiology, as predictions come in similar formats

Disadvantages of prediction platforms

 * Not trivial to switch from scores/points to monetary rewards, as assigning monetary rewards to the top forecasters makes the reward not be proper anymore. Forecasters e.g. may have an incentive to report beliefs that are more extreme than the ones that they actually hold to get a chance to score high in a tournament with payouts
 * Herding problem: people are incentivised to copy the aggregate forecast (e.g. the community median)

Incentive problems
see paper "Alignment Problems With Current Forecasting Platforms", https://arxiv.org/pdf/2106.11248.pdf