Many prediction markets would be better off as batched auctions
18Edward Turner
3William Howard
8Brendan Long
2[comment deleted]
15Joern Stoehler
7Douglas_Knight
9William Howard
4Joern Stoehler
4interstice
7Douglas_Knight
2interstice
3Pat Myron
10Douglas_Knight
6Edward Turner
8the gears to ascension
3William Howard
3Ben
3the gears to ascension
7ProgramCrafter
3William Howard
6Pat Myron
2Sonata Green
New Comment
Currently on a non-compete so I can't say what I actually think here but a few things that may help:
1) Some counters on why you might want HFT: https://blog.headlandstech.com
2) CLOBs offer a genuine service: "if you are desperate to trade I am here to trade with you right now ... you just need to pay a fee by crossing the spread"
3) A lot of what you say has been thought about a LOT more in real markets already, look up things like dark venues/midpoint venues/systematic internalisers/retail service providers/periodic auctions
4) I think adding an end of day auction to prediction markets would likely be a benefit
5) Market makers make money off of retailers overall, they also typically add a genuine service of liquidity and minimising volatility, there are clear examples of this, you should happily pay (some amount) for that service
On footnote 2: yes they will almost certainly have an API for automated trading.
On footnote 3: markets will react as fast as is efficient to do so. If you can make enough profit by being faster then the market will. Note real markets are in ways already auctions, you cannot actually trade instantaneously, exchanges have clock cycles you may have to wait for.
2Some counters on why you might want HFT: https://blog.headlandstech.com
Thanks! In reading about this, I saw that that batch auctions had been proposed and tried in various financial markets, but they hadn't caught on, so it's interesting to see some of the pushback.
I'm not super convinced by the arguments in that post. I think a lot of them are begging the question wrt whether to prefer continuous or batched trading. I.e. given that trading is continuous, HFT does provide a lot of utility in various ways, e.g. reducing spreads between distant markets. But the reason those spreads exist is because continuous trading creates the possibility for something to happen in New York while your order is on its way from Chicago. You could equally solve that problem (at least partially) by imposing some kind of delay or using batched auctions with a shared logical time.
The one point which may be a crux is "Feedback Loops in Global Supply and Demand Negotiation". If it is the case that there is meaningful price discovery done by back-and-forth negotiation over hundreds or thousands of "market ticks", then this is a point against batched auctions.
Also I think the interesting case on prediction markets now is on relatively thinly traded markets, where reactions still tend to be on "human" timescales (someone doomscrolling twitter for news vs someone doing a monthly forecast, rather than two bots competing against each other). I think a lot of the logic about HFT doesn't really apply in this case.
if you are desperate to trade I am here to trade with you right now ... you just need to pay a fee by crossing the spread
Part of my claim is that this use case doesn't generate a lot of utility in prediction markets, and that serving it comes a cost to the majority[1] who wouldn't mind waiting.
(I may come back and respond to more later)
- ^
In some information-weighted sense
(I'm not the original person you're responding to but I came here to say something similar to them)
given that trading is continuous, HFT does provide a lot of utility in various ways, e.g. reducing spreads between distant markets. But the reason those spreads exist is because continuous trading creates the possibility for something to happen in New York while your order is on its way from Chicago. You could equally solve that problem (at least partially) by imposing some kind of delay or using batched auctions with a shared logical time.
I think this isn't why HFT reduces spreads (and I think the HFT front-running thing is a conspiracy theory with basically no evidence?). HFT lets market makers have high volume while reducing the risk of each individual transaction.
The risk of your counterparty knowing something you don't is always there, whether it's prices on another exchange or reading announcements faster than you or having insider information. Market makers are fine with that risk because they can protect against it by selling slowly and making a lot of sales. If I offer to sell you shares of 10 shares X of $1 and you buy all of them, I might offer a few more and see what happens, but if you continually buy all of them no matter the price, I'm going to stop offering because clearly you know something I don't. But I can do all of that negotiation in seconds with HFT, so I can offer 10 shares at a time and still sell tens of thousands per day. But if my only choice is to offer one big block of shares per day, then there's a huge risk that I'll lose money on that trade and I need to increase the bid/ask spread to stay in business.
1I recently collected a large premium on reaction time for the various IMO gold manifold markets bc I coincidentally saw OpenAI's announcement on twitter only 10min late.
Collecting the premium involved filling various limit orders that I confidently guess were long outdated, i.e. that would've been removed a week before the IMO if the owner had bothered to think about them at that time, in addition to being short outdated, i.e. removal after the OpenAI announcement if only the owner had as fast a reaction time as I lucked into.
I expect for long running markets, 24h batched auctions increase liquidity, and thus accuracy, bc more aggressive limit orders / price-volume tails will get specified, since there's less risk of forgetting to cancel the limit order before the information landscape changes. I.e. right now there's often less intelligence and information contained in old limit orders vs new orders bc of reaction times, and so aggregating all information should "weigh" them less in a smart way (i.e. take older orders as less strong Bayesian evidence). Often I think we don't care for super fast aggregation, and care more that traders get higher incentives to provide aggregate-able evidence e.g. via specifying their confidence curves at lower friction from platform UX and adversarial selection in one direction (i.e. if you specify two tails to capture volatility, one gets bought up entirely if you react slower than others to new information). Which is why expect that batched auctions would lower that friction, and lead to higher liquidity and more accurate forecasts, at merely lower time resolution.
That makes it sound like there's a lot of room for big improvements from small changes. Like just add expiration dates to limit orders. Then have the UI suggest reasonable dates. Maybe 1/2 of the time remaining. In an absolute sense that is a very long time, but surely it is better for almost all users than the status quo of all of the time remaining.
Expiration dates on limit orders achieve something in this direction, but it's not quite the same thing that batched trading would achieve.
Consider the case where you place an order based on your current forecast, and then some event happens while you're not watching the market:
- Under the current system, supposing you set an expiration time of 1 day: If the event happens within the day, you will still get scooped. If it happens after 1 day, then you can't get scooped but you also lose the opportunity to make the trade at your endorsed price. For thinly traded markets it may well take more than 1 day for sufficient liquidity to show up, so you have to make this tradeoff between waiting for liquidity and risking getting caught with your trousers down.
- Under a batched auction system, with a batch interval of 1 day: If the event happens within the first day, as long as you notice it before the auction executes everything is fine. If it happens after the first day, it's still fine as long as you check the market on that day, so you might set expiration dates on your limit orders according to the rule "as long as I can commit to checking on this once per day".
If you do fail to react before the auction executes: You are still likely to have your order taken up at a more favourable price, because the auction executes at a single price. Looking at the first image in the post, suppose you are the buyer with a limit order at $14. In continuous trading, your order would have been taken at around $14, as soon as the marginal traders showed up to act on the new info. But under batched trading, more sellers would show up before the auction takes place, pushing the price lower. So you are still safer leaving your order sitting there in the market, as long as you trust that the bulk of orders at the time the auction executes are from "live" traders (who are actively reacting to information)[1].
- ^
Now I think about this this could actually push in the direction of lower accuracy, because people can afford to be more blasé about leaving low info orders in the market
1There already are optional expiration dates, and I use them a lot whenever I can guess the next time new information may become available that I'd want to have a chance to react to.
Half of time remaining is a good guess, for at least about 33% of my bids I'd say (?) So this would speed up bidding somewhat for me, and likely get a lot of people who don't use expiration dates to use them, and save their skin.
For the IMO markets one problem would be that the markets were set up to close at EOY, not during the IMO. I do not know why. So for anyone who bid in 2025 the default proposal would've been too late regardless.
Polymarket and manifold already have optional expiration dates, and Polymarket suggests a default expiration of "end of today".
The default on Polymarket is no expiration. It is only after actively choosing to have an expiration date that it goes to 1 day. I suspect that is so short that it encourages people not to choose it. Similarly, the default with Manifold is not expiring. But if you do click to change, it offers 5 options from immediate to 1 month.
Both sites encourage market orders over limit orders. Polymarket makes limit orders visible but not default. Manifold buries them under "advanced." So the people Joern profited off of were sophisticated enough to choose the "advanced" options but not sophisticated enough to choose good expiration dates.
Encouraging people to use limit orders is a step in the direction of a smoother market. Limit orders with 1 month expiration are somewhat like a sequence of monthly auctions. But they may be harder to understand and require more UI work.
1The default on Polymarket is no expiration. It is only after actively choosing to have an expiration date that it goes to 1 day
Hmmm you're right, guess I misremembered.
I recently collected a large premium
mana is purely a play-money and not exchangeable for sweepcash, money, or any other goods and items
Manifold's accuracy lags because premiums collected aren't nearly as large as their digits imply
To what extent do these arguments apply to the stock market? Has any stock market tried this? Why didn't the people complaining about HFT who founded IEX do this? Is this legal (eg, compliant with NBBO)? Even if it isn't legal in America, lots of countries have their own stock markets and could try this or even impose it.
Yes, nothing here is unique to 'prediction markets', CBOE have addressed this. It is implemented and traded via BATS in Europe.
does this affect incentive to provide reasoning? eg right now I might buy, then explain, in the hope that other people will believe me, also buy, and I can sell high (and if my claims were truthful and predictive, then this doesn't screw them).
if trades are public like manifold right now, this might result in trading strategies like "mimic the most successful traders" becoming plausible. if I know a list of predictors who are reliably accurate, and I know whose offers are whose before the trade goes through, I can make the same curve of buys or asks they do. I haven't worked this out carefully but I think I'd need to match the entire market on the other side in order to keep from moving the price by doing this, but it should be possible to do this in a way that doesn't affect final price, and takes the same action as someone else. Which, from a prediction standpoint, could be very good, if I know one person has domain knowledge.
On the flipside, if bid/ask sequencing information isn't available, this might obscure fine-grained detail about who was correct first? not sure about that.
right now I might buy, then explain, in the hope that other people will believe me, also buy, and I can sell high
I think this would still work, as long you don't explain before your buy order goes through 🙃
if trades are public like manifold right now, this might result in trading strategies like "mimic the most successful traders" becoming plausible
I think this is right, because you could copy a successful trader in the same transaction they are buying in, so they would lose their first mover advantage (previously people could copy, but only after they picked up shares at the best price). This would be a good reason to not have individual orders public, at least before a batch trade goes through.
Interestingly real stock markets don't have public orders for call auctions, they report an Indicative Match Price, and some kind of aggregated information about buy/sell imbalance (details vary). Whereas for continuous trading orders are public. Presumably this is for a similar reason.
It would maybe create a bit of a free-rider problem, if you had a situation where valuable data could be acquired at a cost (by doing a survey or experiment, or having fast messengers to run to London from Waterloo), then you kind of get exploited by anyone who just copies you.
On the other hand, in the immediate (non-equilibrium) mode the "copy the best trader" strategy is helping the market make good predictions. If their really is someone out there who is making perfect predictions, then having a lot of money follow their lead is going to result in better predictions.
as long you don't explain before your buy order goes through
might even make sense to build this into the UI. noob-facing flavortext might be like "After-Trade Comment: make a comment that only becomes public after your order goes through, to explain your reasoning once you've bet on it? if you're honest and correct, and others know that, others might match your bet, allowing you to sell early if you'd like to de-risk"
realizing that there's still incentive for timeliness that might be awkward. it's still the case that the person who trades last has the best position to update off whatever information the order book provides. This might cause people to put fake orders in and change them last-minute. possibly avoidable by, eg, limiting how often you can change your order; perhaps for a 1 day call auction, you can't change your order more often than an hour after you placed it? perhaps this only makes sense for a limited period before the end of the call auction, eg if you have an order in the book 2 hours before it closes, you can't change that order. that limits the time period where someone has to come back to check if someone did a fakeout.
alternately, just don't provide any information from the unclosed call auction whatsoever; the price updates once a day, and you get no further information.
This seems to apply to prediction markets which have hundreds of traders, while many have only a few (3-5, sometimes up to 15).
Some of the points apply more obviously to high volume markets, but I actually think this would be a bigger improvement on thinly traded markets. Mainly because of the aggregation of liquidity and "better ergonomics". With 3-5 people, the issue of whether any of the participants are even checking the market at any given time is more of a problem, so there is a bigger benefit to having an agreed-upon time at which to expect everyone to have their orders ready.
100ms rather than 1 second, or even 1 second rather than 1 hour. If you can think of a case where it would be useful to have a very fast reaction to the sort of question that appears on prediction markets
Great post. Tangentially noting the vast majority of 2025's prediction market volume is live events ("79% of Kalshi’s recent volumes have been related to sports") where auctions as infrequent as hourly fundamentally changes that offering
Medical prediction markets, like in dath ilan, might prefer a 5-second reaction time to 1-minute.
Curated and popular this week
