We compare sportsbook and prediction market prices across the same markets to try to tease out how anonymity vs maker competition affects spreads. We see these two forces as specific market structure elements that affect liquidity and aim to tease out their aggregate effect.
A sportsbook knows a lot about you before you even place your first bet. They use all the information they can get (gender, funding method, access device) to discriminate on total available liquidity (betting limits) and prices (max payouts). Prediction market makers, on the other hand, do not have this information. However, in a prediction market, makers are always competing to provide you the best price.
Sportsbooks are the only counterparty to all traders betting on their platforms. All bets on Fanduel / Draftkings are filled by in-house liquidity, akin to an internal market making arm for a CEX exclusively filling all orders. This is also the impetus behind why sportsbooks care about counterparty identity as an input to a probabilistic model of sharpness - the user’s win is the platform’s loss. Combined, this results in high amounts of liquidity for squares / losing bettors and limited size for sharps.
Prediction market makers, on the other hand, operate on an anonymous orderbook. Furthermore, unlike in perps, the main risk is not latency relative to the dominant price-discovery venue. It is trading against privately informed flow, where no external reference price exists to reveal that the maker’s quote is stale. This means compared to sportsbooks, prediction market makers must widen spreads to combat this issue.
However, prediction markets also open up competition among different makers (and Polymarket gives out liquidity incentives today). This is a counterbalancing force to the lack of counterparty information that should lower prices compared to sportsbooks.
The following dashboard examines both hypotheses by looking at shared markets across 4 venues (Draftkings, Fanduel, Polymarket and Kalshi) for items that do not have resolution differences (sports games) to try to distill the effects of anonymity vs maker competition on spreads.
Finding 1: For liquid markets on prediction markets, prices are ~100-300 bps better than sportsbooks. Even considering 150-175 bp fees at 0.50 for Polymarket, Kalshi, PMs still have materially better prices than sportsbooks.
This is clearly expected. https://x.com/annanay/status/1959948619550785964?s=20
Take any OTC market and move it to a central limit order book… spreads tighten, price discovery improves, and counterparty risk evaporates…
We moved from markets where monopoly liquidity providers had pricing power and price-insensitive users, then added maker competition.
One caveat worth mentioning is that the prediction market value used is the best ask; it is not guaranteed that the best ask has large amounts of liquidity. There are cases where the best ask is < 10usd, but there are no observed cases where the delta price between best ask and next best ask is larger than 5c.
Finding 2: Despite increased maker competition and even rebates for makers, the competition factor doesn’t push prediction market prices to be significantly better than de-vigged sportsbook odds (entertain us in how we devig this)
Devigged sportsbook odds represent the fair price assessment by the sports book including an aggregate of its counterparty information including knowledge it obtains from its sharp betting flow and its risk management engine among other factors. We can therefore, given the limitations of public information, assume a comparison between the devigged book odds and the prediction market odds is the price of anonymity.
We calculate fair probability as:
over_round = sum(raw_implied_probs)
fair_prob_i = raw_prob_i / over_round
For example on a 50/50 cointoss, a sportsbook may quote -110/-110 (bet 110 usd to win 100 on both sides). That signifies 52.38% and a 52.38% implied probability of the event to happen. Fair odds are 50/50 - so you are getting a worse price. Here’s an example of the above calculation:
Over_round = 104.76%
Fair_prob_a = 52.38 / 104.76 = 50%
Here, 50% represents the sportsbooks’ interval view of the fair odds (barring odds shading/risk management...etc) and a 4.76% vig, which accrues over large numbers as margin to the sportsbook.
In the fee adjusted fair diff view, we compare the devigged fair odds against the de-feed (maker fee) price on prediction markets. We are in essence trying to see “what is the fair underlying price that makers think is fair for this market?”. The comparison is not perfect because the prediction market still embeds the maker's margin (including liquidity incentives).
The same holds for devigged prices compared against no fee prediction market best asks.
Note that Kalshi has fees on all sports markets while Polymarket does not charge maker fees - and therefore the distribution is unaffected. Furthermore, both Polymarket and Kalshi have rebate programs for makers, but we ignore the effect of the rebate program (if anything, the rebate program makes the “real odds to the maker” worse.
Our current hypothesis as to why this is the case is that the additional amount of information sportsbooks have on their users - specifically counterparty information - allows them to offer tighter spreads because they have in effect a counterparty aware last look. Polymarket specifically allows makers to see all executed trades in real time, so resting orders can already react to sharp flow in real time to adjust odds, so the fair diff should reflect just the tighter book spreads as well as their risk management activities.
Finding 3: Longer tail markets on prediction markets have wide spreads vs sportsbooks.
The distribution above has a fatter left outlier tail. This represents less popular or far out markets that have 10-50% spreads on prediction markets (ie. FA Cup, UEFA Europa Conference League, Liga Mx). The lines with more than 10% pred market and sportsbook differences tend to have less than 1,000 usd volume on the prediction market.
We hypothesize this is due to anonymity and lack of maturity on prediction markets compared to sportsbooks w.r.t longer tail games and leagues.
The conclusions we can draw from this is
Non-anonymity/flow-enrichment decreases spreads. Maker competition decreases spreads. In prediction markets in particular, where long tail markets have trouble with liquidity depth and spreads, adopting sportsbook based market structure that have proven to be able to service massive OI will allow users to source deeper liquidity at tighter spreads on a wider variety of markets with the tradeoff
Sharp flow that is payout limited on sportsbooks can still source liquidity due to maker competition albeit likely at a worse price. To allow for privacy, users should be able to pay for optional anonymity but pay a higher fee which can be used to compensate makers who take on toxic flow.
If you are interested in opinions / ideas around prediction market market structure, we’re always happy to chat! Dm us on X.
Thanks to @allquantor, @Mikey0x_ and @_helmass for review
Originally posted on X.