Forward the Original Title: In-depth on prediction markets and why they are still cool (for certain markets)
Prediction markets are contract-based markets that track the outcome of specific events.
Traders buy shares in a market (priced 0 < x < 100), and depending on the event’s outcome, those shares are either worth 0 or 100.
The only constraints on a prediction market’s existence are a willing external party to create the market and traders willing to purchase contracts for both sides.
There are three different types of prediction markets:
There are several different real-world practical applications for prediction markets:
Probability theory is a framework for quantifying uncertainty. Probability is present in every aspect of life, from simple, everyday choices to research and risk assessment. Probability allows individuals to make logic-based decisions by understanding the likelihood of an event occurring.
Accurate probability is the representation of the true odds of an event occurring, free of manipulation and bias.
The most verifiable way to prove the accurate probability of an event occurring is by backing those odds with the most valuable good in the world: money.
In a world where prediction markets are utilized as the first probability source, biased and manipulated odds are not considered because nothing quantifiable backs those odds.
There are two main methods of liquidity systems in prediction markets.
AMMs (automated market makers) are smart contracts that hold liquidity for an asset pair. Users can swap each asset through the liquidity pool at an exchange rate determined by a formula. AMMs charge fees on each trade, which are returned to the liquidity pool and allocated pro-rata to each liquidity provider.
CLOBs (central-limit order books) are a liquidity system with two options: maker and Taker. Maker orders place bids below and ask above the mark price.
The duality of CLOB fees is that, depending on the trader’s needs, there are differentiating fees through maker and taker orders.
Users taking away from the liquidity through a taker order will pay more fees than a user adding to market stability through maker orders. In some extreme markets, taker orders can subsidize maker orders into collecting a positive fee. This is impossible through AMMs - everyone pays the same fee regardless of whether you assist in market stability.
This enhances liquidity through condensation into price bands that traders utilize. CLOB Systems also allows large orders to be placed and fulfilled at specific prices. They also reduce the probability and likelihood of sandwich attacks because their price impact is less predictable than that of AMMs, which is dangerously predictable.
Liquidity is arguably the most significant issue prediction markets face.
It is not +EV for market-makers to provide liquidity on prediction markets because of the tail risk of being stuck with zeroed-out shares.
Until prediction markets reach a stage where there is enough demand to market-make to keep markets efficient, markets need to be subsidized.
Subsidizing liquidity is integral to justifying the risk: reward ratio market-makers face when market-making prediction markets.
The concepts outlined below show that it is possible to attract liquidity with the right incentives.
There are four popular avenues that markets can take here:
Directly subsidizing liquidity from protocol profits is not viable in the long run. However, in the short term, it is perfectly feasible and very similar to the CAC (customer acquisition cost) paid by traditional sportsbooks to attract users.
In the crypto world, if a protocol isn’t subsidizing you to do something, even as simple as depositing ETH in a lending dAPP, there is a high plausibility that a competitor is willing to subsidize you to do the same action. I expect prediction markets like Polymarket and Thales to continue subsidizing for as long as possible (especially with native tokens once released).
LLMs are a form of artificial intelligence that integrates machine learning to analyze significant data sets and solve tasks.
In the future, I find it very probable that LLMs will act as the ‘creators’ of markets. Prediction markets rely on clean-cut rules to provide resolution.
With some abstract markets, there are many possible loopholes.
An example of this is the recent market on Polymarket for the approval of the Ethereum ETF. The rules stated “an approval” of the spot Ether ETF by May 23rd for the market to resolve YES. Still, they did not clarify if 19B-4s alone constituted enough to determine the market or if the S-1s (set to be approved later) were also needed.
Having LLMs create rules for markets greatly reduces the probability of loopholes in abstract markets. Additionally, on the off chance that a market has a loophole, LLMs can act as the ‘resolver’ to prevent other dispute resolution methods from taking place (which introduce attack vectors mentioned below).
Integrating decentralized LLMs like Bittensor can also prevent manipulation in rule construction and dispute resolution.
While LLMs are not yet polished or accurate enough to have any fundamental, meaningful role in the construction of these markets, in the future they will be. As such, I expect many prediction markets to switch from manual market creation to strictly LLM to prevent disputes.
The process for a market on Polymarket to be initialized and resolved is the following:
If there is a dispute over the resolution, UMA’s governance holders vote on the resolution. UMA’s DAO effectively acts as a supreme court.
Utilizing the UMA DAO as a resolution court leaves markets prone to potential manipulation attacks:
As of writing, $UMA has a 291.4 circulating market cap and a volume/mc ratio of only 18%. It would be incredibly difficult and expensive to acquire 51% of the circulating tokens. Additionally, it would be significantly unprofitable, as Polymarket does not have nearly enough volume to warrant an attack this way.The probability of an attack occurring through a single address owning 51% of the circulating supply is improbable.
Another attack possibility is through bribes. Suppose an attacker can convince large holders to vote alongside him (either through them also participating in the attack or bribes). In that case, the probability of a DAO vote succeeding in rewarding the shares that should have been zeroed out is high.
The above scenarios are ultimately very unlikely and short-term as AI expands and LLMs (large-language models) can act as resolution sources.
In prediction markets, asymmetric information is the concept that a party has more information on the outcome of an event than the party they are trading against.
If there is a market for whether Variational will release its token before June 1st, an insider at Variational can buy up shares of the outcome they know will occur.
Blockchains cannot decipher if a party has access to asymmetric information. While blockchains make monitoring and analyzing transactions simple, they cannot assess the reasoning behind a transaction. This is because networks do not have a way of connecting perfectly anonymous addresses to their real-life identities.
Thus, it is not technically possible to asses whether an anonymous address that places a prediction has access to asymmetric information.
Oracle front-running is the concept that a trader has access to asymmetric information before an Oracle, thereby allowing them to place bets or trades that they know will be profitable.
In prediction markets, if an event is effectively resolved but the market is still tradeable, this creates an attack where traders with knowledge that the event is resolved can buy up shares trading at a discount to their actual value.
Sportsbooks solved this issue by creating a short delay in placing bets to allow their oracles to process data and subsequently adjust market odds. This protected sportsbooks from individuals at an actual sports game betting as soon as they saw something happen. This is not plausible for prediction markets because some traders can access asymmetric information weeks/months before market resolution.
While some argue this makes an efficient market, this problem creates a significant issue for market-makers because of adverse selection.
If market-makers are trading against people who consistently are better-informed than them, they will face consistent losses and eventually stop market-making, leading to less overall liquidity.
J.P. Morgan has estimated that the daily notional value of 0DTE options trading has reached approximately $1 trillion.
This surge is representative of the power 0DTE options offer to capitalize on intraday market movements with cheap leverage. In crypto, this is no different, people are hungry for leverage.
The liquid flow of prediction markets is well-suited for 0DTE options. This is because with financialized products here is always a way to hedge/arbitrage with spot/perps resulting in tight spreads and efficient pricing.
This effectively solves the liquidity crisis hindering the expansion of speculative markets like elections.
With ODTE options, retail can still “hit big”:
While this is an exuberant example, the point remains that retail can still hit many multiplies on their initial position without needing to go through complex routes.
0DTE options offer a hyper-gamified experience for retail while also being the easiest route for them to use leverage.
Forward the Original Title: In-depth on prediction markets and why they are still cool (for certain markets)
Prediction markets are contract-based markets that track the outcome of specific events.
Traders buy shares in a market (priced 0 < x < 100), and depending on the event’s outcome, those shares are either worth 0 or 100.
The only constraints on a prediction market’s existence are a willing external party to create the market and traders willing to purchase contracts for both sides.
There are three different types of prediction markets:
There are several different real-world practical applications for prediction markets:
Probability theory is a framework for quantifying uncertainty. Probability is present in every aspect of life, from simple, everyday choices to research and risk assessment. Probability allows individuals to make logic-based decisions by understanding the likelihood of an event occurring.
Accurate probability is the representation of the true odds of an event occurring, free of manipulation and bias.
The most verifiable way to prove the accurate probability of an event occurring is by backing those odds with the most valuable good in the world: money.
In a world where prediction markets are utilized as the first probability source, biased and manipulated odds are not considered because nothing quantifiable backs those odds.
There are two main methods of liquidity systems in prediction markets.
AMMs (automated market makers) are smart contracts that hold liquidity for an asset pair. Users can swap each asset through the liquidity pool at an exchange rate determined by a formula. AMMs charge fees on each trade, which are returned to the liquidity pool and allocated pro-rata to each liquidity provider.
CLOBs (central-limit order books) are a liquidity system with two options: maker and Taker. Maker orders place bids below and ask above the mark price.
The duality of CLOB fees is that, depending on the trader’s needs, there are differentiating fees through maker and taker orders.
Users taking away from the liquidity through a taker order will pay more fees than a user adding to market stability through maker orders. In some extreme markets, taker orders can subsidize maker orders into collecting a positive fee. This is impossible through AMMs - everyone pays the same fee regardless of whether you assist in market stability.
This enhances liquidity through condensation into price bands that traders utilize. CLOB Systems also allows large orders to be placed and fulfilled at specific prices. They also reduce the probability and likelihood of sandwich attacks because their price impact is less predictable than that of AMMs, which is dangerously predictable.
Liquidity is arguably the most significant issue prediction markets face.
It is not +EV for market-makers to provide liquidity on prediction markets because of the tail risk of being stuck with zeroed-out shares.
Until prediction markets reach a stage where there is enough demand to market-make to keep markets efficient, markets need to be subsidized.
Subsidizing liquidity is integral to justifying the risk: reward ratio market-makers face when market-making prediction markets.
The concepts outlined below show that it is possible to attract liquidity with the right incentives.
There are four popular avenues that markets can take here:
Directly subsidizing liquidity from protocol profits is not viable in the long run. However, in the short term, it is perfectly feasible and very similar to the CAC (customer acquisition cost) paid by traditional sportsbooks to attract users.
In the crypto world, if a protocol isn’t subsidizing you to do something, even as simple as depositing ETH in a lending dAPP, there is a high plausibility that a competitor is willing to subsidize you to do the same action. I expect prediction markets like Polymarket and Thales to continue subsidizing for as long as possible (especially with native tokens once released).
LLMs are a form of artificial intelligence that integrates machine learning to analyze significant data sets and solve tasks.
In the future, I find it very probable that LLMs will act as the ‘creators’ of markets. Prediction markets rely on clean-cut rules to provide resolution.
With some abstract markets, there are many possible loopholes.
An example of this is the recent market on Polymarket for the approval of the Ethereum ETF. The rules stated “an approval” of the spot Ether ETF by May 23rd for the market to resolve YES. Still, they did not clarify if 19B-4s alone constituted enough to determine the market or if the S-1s (set to be approved later) were also needed.
Having LLMs create rules for markets greatly reduces the probability of loopholes in abstract markets. Additionally, on the off chance that a market has a loophole, LLMs can act as the ‘resolver’ to prevent other dispute resolution methods from taking place (which introduce attack vectors mentioned below).
Integrating decentralized LLMs like Bittensor can also prevent manipulation in rule construction and dispute resolution.
While LLMs are not yet polished or accurate enough to have any fundamental, meaningful role in the construction of these markets, in the future they will be. As such, I expect many prediction markets to switch from manual market creation to strictly LLM to prevent disputes.
The process for a market on Polymarket to be initialized and resolved is the following:
If there is a dispute over the resolution, UMA’s governance holders vote on the resolution. UMA’s DAO effectively acts as a supreme court.
Utilizing the UMA DAO as a resolution court leaves markets prone to potential manipulation attacks:
As of writing, $UMA has a 291.4 circulating market cap and a volume/mc ratio of only 18%. It would be incredibly difficult and expensive to acquire 51% of the circulating tokens. Additionally, it would be significantly unprofitable, as Polymarket does not have nearly enough volume to warrant an attack this way.The probability of an attack occurring through a single address owning 51% of the circulating supply is improbable.
Another attack possibility is through bribes. Suppose an attacker can convince large holders to vote alongside him (either through them also participating in the attack or bribes). In that case, the probability of a DAO vote succeeding in rewarding the shares that should have been zeroed out is high.
The above scenarios are ultimately very unlikely and short-term as AI expands and LLMs (large-language models) can act as resolution sources.
In prediction markets, asymmetric information is the concept that a party has more information on the outcome of an event than the party they are trading against.
If there is a market for whether Variational will release its token before June 1st, an insider at Variational can buy up shares of the outcome they know will occur.
Blockchains cannot decipher if a party has access to asymmetric information. While blockchains make monitoring and analyzing transactions simple, they cannot assess the reasoning behind a transaction. This is because networks do not have a way of connecting perfectly anonymous addresses to their real-life identities.
Thus, it is not technically possible to asses whether an anonymous address that places a prediction has access to asymmetric information.
Oracle front-running is the concept that a trader has access to asymmetric information before an Oracle, thereby allowing them to place bets or trades that they know will be profitable.
In prediction markets, if an event is effectively resolved but the market is still tradeable, this creates an attack where traders with knowledge that the event is resolved can buy up shares trading at a discount to their actual value.
Sportsbooks solved this issue by creating a short delay in placing bets to allow their oracles to process data and subsequently adjust market odds. This protected sportsbooks from individuals at an actual sports game betting as soon as they saw something happen. This is not plausible for prediction markets because some traders can access asymmetric information weeks/months before market resolution.
While some argue this makes an efficient market, this problem creates a significant issue for market-makers because of adverse selection.
If market-makers are trading against people who consistently are better-informed than them, they will face consistent losses and eventually stop market-making, leading to less overall liquidity.
J.P. Morgan has estimated that the daily notional value of 0DTE options trading has reached approximately $1 trillion.
This surge is representative of the power 0DTE options offer to capitalize on intraday market movements with cheap leverage. In crypto, this is no different, people are hungry for leverage.
The liquid flow of prediction markets is well-suited for 0DTE options. This is because with financialized products here is always a way to hedge/arbitrage with spot/perps resulting in tight spreads and efficient pricing.
This effectively solves the liquidity crisis hindering the expansion of speculative markets like elections.
With ODTE options, retail can still “hit big”:
While this is an exuberant example, the point remains that retail can still hit many multiplies on their initial position without needing to go through complex routes.
0DTE options offer a hyper-gamified experience for retail while also being the easiest route for them to use leverage.