Prediction Markets

The Challenge

Prediction markets (PMs) are hard to market-make. Each collection of all bet slips has a fixed, real value. But which slip receives and pays out all this value at maturity, irrevocably, is uncertain. Additionally, prediction markets are interesting precisely in situations of high information asymmetry. Whoever supplies liquidity faces the risk of selling undervalued bet slips to an insider.

To cope with this (and even make large profits), betting platforms typically use two tricks:

• high fees (rakes)
• edge

The edge could be:

• Market-specific (advanced prediction algorithms or employing leading experts for the kind of markets selected on offer)
• Banning users who are too good
• Others, such as:
  • delays of live stream
  • manipulation (withholding of information, marketing)

Sustainable real-money prediction markets on general topics that are permissionless cannot rely such edge used in the betting space, nor can fees be high or users will lose money on average and be disincentivized.

Liquidity Issues

Adverse Selection

The problem of thin PMs is mentioned as motivation for the invention of the first automated market maker (a special AMM well suited for PMs with conditional questions) by Robin Hanson. However, the passive AMM will always lose money against even the most basic arbitrage, for example after the result is in. At the resolution of the market, this loss is cemented as permanent. Previous efforts have found that these costs are typically higher than the value brought by the information aggregation, and therefore have not lead to profitable businesses in two decades. Research has produced AMMs that take fees dynamically but these are still too passive to avoid high fees when they are not needed.

P2P

The only way to run prediction markets is therefore p2p. Even experts and potential insiders often disagree with each other. The concept of debate is as old as human civilization. Therefore, the expectation is that, at least for some interesting and controversial topics with passionate communities, trading volume can be substantial not just for irrational all-in bets (wager with any odds on a favorite) such as seen in sports betting. Interestingly, it is not irrational to participate in such entertainment, as it is very difficult to lose money:

• if the market is accurate and takes negligible fees, the expectation value is zero on both sides for each binary market (No expected loss no matter what one bets on.)
• if the market is not accurate but the user picks a side randomly, there is still no loss expected across markets, because one can both win and loose with the same probability.

Frontrunning

When there are no MMs willing to participate, CLOBs technically still allow p2p trades (one needs to make a small modification such as to match bids with bids on the opposite side, and similar with asks, besides the usual bids and asks of the same asset). To this end, some predictors would have to set limit orders, which is a UX challenge¹.

More importantly, this approach is not viable for the reason of speed-based adverse selection against such users in a way that is not just simply exploiting some total passivity of LPs, but something very human: it takes time to update limits whenever new information clearly changes the odds. (This might be a sudden reveal of the outcome itself, a factor such as a goal being scored in a sports game, a press conference or the release of a report, a whistle-blower posting something...)

As a consequence, even the best predictor tend to lose money just for being a little bit slower than someone with decent predictions and faster execution. CLOB-based prediction markets are therefore exploitative in the long run against its best human predictors, which is unacceptable and would limit their adoption.

The Solution

We conclude that what is needed is a p2p trading structure that can be slowed down sufficiently for people to share their belief and then, over time, gradually and automatically invest in and trade bet slips when the market prices them cheaper than what the user believes are the winning chances.

Furthermore, trading should happen in a way that speed is not relevant beyond the time duration in which someone upholds a controversial opinion, such that all orders are streamed at a reasonably (potentially very slow for long-duration markets) time scale such that everyone gets the same price at any moment.

And this is, of course, the GLOB.

Footnotes