Candid Conversations for Financial Markets

Vincent Grégoire discusses prediction market mechanics, calibration accuracy, and profit concentration based on 588M transactions.

34 min Listen to the original Episode page
Prediction MarketsPolymarketKalshiMarket MicrostructureForecastingAsset PricingBehavioral FinanceRegulatory Compliance

Summary

Vincent Grégoire joins Laura Paglia to discuss the mechanics and empirical performance of prediction markets. Grégoire explains that binary contracts on platforms like Polymarket price outcomes as risk-neutral probabilities, with liquid markets generally showing strong calibration where a 70-cent price corresponds to a 70% win rate. He presents research analyzing 588 million transactions, finding that price discovery occurs primarily through limit order quotes rather than executed trades. The study reveals extreme profit concentration, with the top 1% of accounts capturing 76.5% of gains. Grégoire notes that while insider trading exists, it accounts for less than 0.1% of profits, and prediction markets offer superior real-time information compared to static analyst reports.

Episode sections

  1. 0:20 – 6:18

    Introduction and Prediction Market Mechanics

    Laura Paglia introduces guest Vincent Grégoire and the topic of prediction markets. Grégoire explains how binary contracts work on platforms like Polymarket, where a $1 position is split into 'Yes' and 'No' tokens that pay out $1 upon resolution, effectively pricing outcomes as risk-neutral probabilities.

  2. 6:18 – 7:39

    Calibration and Forecast Accuracy

    Grégoire discusses the calibration of prediction markets, noting that liquid markets are generally well-calibrated, meaning a 70-cent price corresponds to a 70% win rate. He highlights exceptions in weather and financial contracts due to risk correlation and liquidity issues, and notes poor calibration at extreme prices (near 0 or 1) due to long-shot bias and time value of money.

  3. 7:39 – 15:23

    Research Methodology and Data Scope

    Paglia outlines Grégoire's study 'Who Wins and Who Loses In Prediction Markets?', which analyzed 588 million transactions worth $67 billion across 2.48 million blockchain addresses from November 2022 to March 2026. The study aimed to identify profitable participants and assess forecast utility.

  4. 15:23 – 19:37

    Price Discovery via Quotes vs. Trades

    Grégoire explains his second study, 'Trades, Quotes, and Price Discovery in Prediction Markets', which utilized quote data to understand price formation. He found that most price discovery occurs through quotes (limit orders) rather than executed trades, mirroring traditional equity markets where market makers incorporate information.

  5. 19:37 – 26:05

    Profit Concentration and Top Traders

    The research found that the top 1% of profitable accounts captured 76.5% of gains. These winners largely acted as liquidity providers, earning small profits on high volume, or made large, correct bets on specific events like the 2024 US Presidential Election. Insider trading cases were negligible in total profit impact.

  6. 26:05 – 31:25

    Comparative Utility of Prediction Markets

    Grégoire notes that while his studies didn't directly compare prediction markets to surveys, existing literature suggests prediction markets offer superior real-time information for earnings and macroeconomic announcements compared to static analyst reports or futures, which are updated less frequently.

  7. 31:25 – 34:20

    Market Growth and Platform Shifts

    Grégoire expresses surprise at the explosive growth of prediction markets, noting that adding four months of data to his sample quadrupled the number of trades. He mentions Polymarket's dominance during the sample period and its subsequent overtaking by Kalshi in the US, despite regulatory differences.

Claims worth verifying

Polymarket lists approximately 100,000 new markets every month.

Unidentified speaker (speaker_1) ·7:22

How to verify: Check Polymarket's official website or API for the number of active or newly created markets over a 30-day period. For recent events, verify 2026-08-12 through 2026-11-12. Use another period only when the speaker explicitly names it.

The study 'Who Wins and Who Loses In Prediction Markets?' analyzed 588 million transactions worth $67 billion across 2.48 million blockchain addresses from November 22, 2022, to March 2026.

Laura Paglia ·7:39

How to verify: Verify against the published paper 'Who Wins and Who Loses In Prediction Markets?' by Grégoire et al., checking the methodology section for sample size and date range. For recent events, verify 2026-08-12 through 2026-11-12. Use another period only when the speaker explicitly names it.

The top 1% of profitable trading accounts captured 76.5% of the gains in the studied sample.

Unidentified speaker (UNKNOWN) ·19:37

How to verify: Verify against the published paper 'Who Wins and Who Loses In Prediction Markets?' by Grégoire et al., checking the results section on profit distribution. For recent events, verify 2026-08-12 through 2026-11-12. Use another period only when the speaker explicitly names it.

The largest documented insider trading case on Polymarket involved a Google employee (Alpha Raccoon) who ranked 115th in profitability.

Unidentified speaker (speaker_1) ·24:41

How to verify: Search for SEC enforcement actions or news reports regarding 'Alpha Raccoon' and insider trading on Polymarket. For recent events, verify 2026-08-12 through 2026-11-12. Use another period only when the speaker explicitly names it.

Primary source

Two documented insider trading cases accounted for less than 0.1% of total profits in the sample.

Unidentified speaker (speaker_1) ·25:57

How to verify: Verify against the published paper 'Who Wins and Who Loses In Prediction Markets?' by Grégoire et al., checking the section on insider trading impact. For recent events, verify 2026-08-12 through 2026-11-12. Use another period only when the speaker explicitly names it.

Adding four months of data (November 2025 to March 2026) to the study sample quadrupled the number of trades.

Unidentified speaker (speaker_1) ·32:19

How to verify: Verify against the published paper 'Who Wins and Who Loses In Prediction Markets?' by Grégoire et al., checking the data update notes or appendix. For recent events, verify 2026-08-12 through 2026-11-12. Use another period only when the speaker explicitly names it.

Kalshi overtook Polymarket as the main US prediction market platform after the study period.

Unidentified speaker (speaker_1) ·33:25

How to verify: Check recent market share reports from prediction market aggregators or news articles comparing Polymarket and Kalshi trading volumes post-March 2026. For recent events, verify 2026-08-12 through 2026-11-12. Use another period only when the speaker explicitly names it.

Weather contracts are less well-calibrated than sports or politics contracts.

Unidentified speaker (speaker_1) ·10:47

How to verify: Verify against the published paper 'Who Wins and Who Loses In Prediction Markets?' by Grégoire et al., checking the calibration results by category. For recent events, verify 2026-08-12 through 2026-11-12. Use another period only when the speaker explicitly names it.

Price discovery occurs primarily through quotes, not trades, in prediction markets.

Unidentified speaker (UNKNOWN) ·15:51

How to verify: Verify against the published paper 'Trades, Quotes, and Price Discovery in Prediction Markets' by Grégoire et al., checking the price discovery analysis section. For recent events, verify 2026-08-12 through 2026-11-12. Use another period only when the speaker explicitly names it.

Prediction markets offer more frequent updates than analyst reports for earnings announcements.

Unidentified speaker (speaker_1) ·27:55

How to verify: Verify against literature cited by Grégoire comparing prediction market prices to analyst forecast revisions for earnings events. For recent events, verify 2026-08-12 through 2026-11-12. Use another period only when the speaker explicitly names it.

Do your own research

  • Verify the calibration of specific prediction market contracts against actual outcomes using historical resolution data.
  • Review the methodology of Grégoire's study 'Who Wins and Who Loses In Prediction Markets?' for sample period biases (Nov 2022-Mar 2026).
  • Compare real-time prediction market prices against CME futures and analyst consensus for major macroeconomic events to assess informational advantage.
  • Investigate the regulatory status of platforms like Polymarket and Kalshi in your jurisdiction to ensure legal access.
  • Check for recent insider trading prosecutions in prediction markets to assess the current risk of regulatory enforcement.
  • Analyze the liquidity depth of limit orders on target contracts to understand the cost of entering and exiting positions.
  • Review the specific terms and fee structures of prediction market platforms to calculate net profitability for high-frequency strategies.
  • Look for counter-arguments in academic literature regarding the 'long-shot bias' and its impact on extreme price calibration.
Full transcriptexpand
0:02
Laura Paglia: Welcome to Candid Conversations for Financial Markets podcast series. I'm your host, Laura Paglia, President and CEO. Our podcasts are designed to provide insights, transparency, and provocative yet achievable proposals for matters that move financial markets in Canada.
0:20
Laura Paglia: Prediction markets let people trade contracts on outcomes of real-world real world events from elections and sporting.
0:27
Laura Paglia: results to economic announcements. Their prices are increasingly treated as forecasts. How good are those forecasts, and who who is earning the money? Our guest, Vincent Gregoire, a professor of finance at HEC Montreal, and holds the Canada Research Chair in Finance and
0:46
Laura Paglia: Technology. He has earned a doctorate in finance at the University of British Columbia.
0:51
Laura Paglia: and master's degrees in financial engineering and electrical engineering at the University of Laval. Before joining HEC Montreal, he taught at the University of Melbourne. He has also worked as a chief data scientist. His research on market, microstructure, information, and financial technology appears in
1:11
Laura Paglia: leading academic journals. He is a frequent research. presenter and media commenter. We are delighted to have him with us. Welcome, Professor Gregois.
1:22
Unidentified speaker (speaker_1): Thank you. My pleasure to be
1:23
Unidentified speaker (UNKNOWN): here.
1:24
Laura Paglia: So, Professor, for someone who's never used a prediction market, could you walk us through a very simple contract? What a participant buys, how it's paid out, and why a price of $70. per 70 cents is often described as a 70% probability.
1:45
Unidentified speaker (speaker_1): Yes. So the way the contracts are set up varies from market to market. I'll take the polymarket example. I think it's the most straightforward to understand. Uh So a market is a yesno question, where uh the example that I like to use being from Montreal is uh
2:05
Unidentified speaker (speaker_1): like an example market would be will the Montreal Canadians win the Stanley Cup in 2027? And then uh that's the question. It's a binary uh question. It's either yes or no, they would they'll they'll win or not. Um and if you trade on this market, uh there's two tokens or two contracts that are
2:25
Unidentified speaker (speaker_1): complementary. So when a contract, when a pair of contracts is created, the platform takes one dollar and creates two contracts. They call them token on polymarket, but they can see as a kind of two contracts, one for the yes and one for the no. And whoever is holding the
2:45
Unidentified speaker (speaker_1): winning contract at resolution gets the dollar back.
2:49
Unidentified speaker (UNKNOWN): Okay.
2:50
Unidentified speaker (speaker_1): So the two.
2:51
Unidentified speaker (speaker_1): contracts together are always worth one dollar. They're always fully collateralized, at least on the main platform like Polymarket and Calci in the US. These will be fully collateralized contracts.
3:09
Unidentified speaker (UNKNOWN): um
3:10
Unidentified speaker (speaker_1): uh financial assets on electronic platforms using a uh limit order book uh system so you you can put quotes or use market orders to uh buy the outside uh whatever liquidity is there uh and that's pretty much it so it's it's uh
3:29
Unidentified speaker (speaker_1): like pretty pretty simple. Uh there there's uh a few details to understand about like how the resolution works. You have to understand the details of your contract, but that's that's pretty much it. Uh so for example, if you are buying a um uh contract for a yes at 70 cents, which is uh kind of the example that you had.
3:49
Unidentified speaker (speaker_1): Uh most often we we the commentators and the platforms will present that as having uh as you placing a 70% chance on on that. that outcome. Because if you're willing to pay 70 cents to buy an asset that's uh paying
4:09
Unidentified speaker (speaker_1): one dollar, uh then it's as if you're uh placing the uh probability of 70% on that asset paying off. Uh it's kind of true.
4:22
Unidentified speaker (UNKNOWN): Uh
4:23
Unidentified speaker (speaker_1): so we know for those uh who has uh have worked in derivative. derivatives or or were did the finance courses on derivatives, we know that there are uh details, right? Some states of the world, some outcomes are more valuable than others. So typically it's uh we we would kind of adjust prices uh for that
4:42
Unidentified speaker (speaker_1): based on on that value. Um so the important thing to understand here, when we're quoting a probability here, we're assuming that it's a risk neutral probability, which in most cases uh kind of applies if these uh markets, if these uh
5:02
Unidentified speaker (speaker_1): markets aren't uh what we call priced in the economy, but that that would be like uh like a hockey match or a hockey team winning. It doesn't have any uh like whether the Montreal Canadians win or lose won't really correlate with the economy of Canada. So so the the like the in in those
5:22
Unidentified speaker (speaker_1): cases, the uh uh risk neutrality assumption is is is fine. The other kind of thing that uh is also another approximation here is that uh well the money's money is locked up, right? So the there it it also this assumption doesn't take into account the time value of money. So the the interest rate that you
5:42
Unidentified speaker (speaker_1): could be earning on these uh uh on the cat capital that's being locked up by you having this position because it is fully collateralized. Platforms have some ways to compensate from that depending on how you access them. Some will pay some interest rate on your holding position. Uh so so that there's slight details there. Uh
6:02
Unidentified speaker (speaker_1): but also overall, yeah, that's why we called that the a probability.
6:08
Laura Paglia: And most people, thank you very much. Most people think the mistake is thinking that that is a 70% chance
6:18
Laura Paglia: of that outcome occurring. And it's not quite that simple.
6:24
Unidentified speaker (speaker_1): Yes. Well, I mean, this is a market price. right? So we can uh often in infer that as this is what the market thinks, right? And then you you you have to look at uh what's the market, who's the market, who's sending the prices in that market. Uh yes, we we we do see that on
6:44
Unidentified speaker (speaker_1): the on the liquid markets, the the the most liquid ones, they tend to be uh fairly well what we call calibrated. meaning that a a token that sells at 70 cents, uh, if we bunch them up and look at them over time, those that sell at 70 cents on average
7:01
Unidentified speaker (UNKNOWN): uh
7:02
Unidentified speaker (speaker_1): win 70% of the time, roughly. Obviously, it's more precise, the the shortest the horizon is, it varies from market types to market types, but the overall these markets, especially the liquid ones. uh and and I'm saying the liquid ones because for example, just polymarket lists over like I think the
7:22
Unidentified speaker (speaker_1): latest numbers is about 100,000 new markets every uh every month. So it's uh not all of them are are quite liquid, but uh for those that are, they tend to be fairly uh fairly well calibrated.
7:39
Laura Paglia: Okay, thank you. Um your
7:42
Laura Paglia: work, Who Wins and Who Loses in Prediction Markets, Evidence from Polymarket. You, along with your co-authors, Pat Ackley, Nicholas Harvey, and Charles Martinal, examined approximately 588 million transactions representing $67 billion in trading in US dollars across.
8:03
Laura Paglia: 2.48 million blockchain addresses over a period of approximately three and a half years from November 22 to March 2026.
8:14
Laura Paglia: You looked at the actual trading records rather than the surveys, investigating who was making money, who was losing money, how participants traded, and to the point that you just made, to what extent are market prices providing useful forecasts?
8:33
Laura Paglia: The paper finds, as you've stated, that the prices generally follow up with how events actually occur in certain instances. You've explained Explain to us the most liquid or well calibrated. Can you walk us through? the strongest evidence that your research
8:53
Laura Paglia: produced that markets provide useful forecasts and where your research shows they fall short?
9:03
Unidentified speaker (speaker_1): Yes. So
9:05
Unidentified speaker (speaker_1): we as I explained earlier, the main test that we did for that was to what we call a calibration test. So we've looked at the whole the whole history of contracts that we had on that platform, looking at the prices they traded at uh one day, seven days, two weeks,
9:25
Unidentified speaker (speaker_1): one month
9:26
Unidentified speaker (speaker_1): before they expired. Uh, and looked at how well these prices align with the uh outcome uh for these markets. So we grouped all of these observations at different horizons. We grouped them by uh by price. So those that are roughly 70 cents would go together. So those that are
9:46
Unidentified speaker (speaker_1): roughly 25 cents would go would go together. And so we did that for all the price points and looked at how well they uh they matched up uh with the in terms of realization frequency and winning frequency. And and we find that the outcome is actually
10:06
Unidentified speaker (speaker_1): very close to the price. So the price match up in aggregate. It does vary from uh category to category. So we we've looked at uh all the markets on polymarket. Polymarket uh have contracts on sports, politics, finance, uh geopolitics.
10:27
Unidentified speaker (speaker_1): uh, the on crypto, uh, on cultural events as uh uh as well as uh weather. So for most uh of them these uh would be fairly accurate but the dynamics are quite different from one category to to another um the some that are are
10:47
Unidentified speaker (speaker_1): less well calibrated uh for example the weather contracts are much well calibrated. Um, to us, it's not clear whether the it's the prices that are are wrong. Uh, or as I explained earlier, uh, when we treat these prices as probabilities, we're treating them as uh as basically
11:07
Unidentified speaker (speaker_1): we're as a risk neutral probability. Uh but for weather contract and some uh financial-based financial base contracts, we know that the these are uh these are well correlated with economic out important economic outcomes. So so whether whether if it rains a lot, if the weather's good or bad for uh crops or
11:27
Unidentified speaker (speaker_1): uh uh or whatever, we know these uh probably don't line up well with the risk neutrality assumption uh that's behind uh these. So it it is a uh potential explanation as to why they they show up as less uh well calibrated. They are also a bit less liquid
11:47
Unidentified speaker (speaker_1): uh than other other markets. So that that could be another uh explanation. Um another kind of part of contracts that are tend to be well uh uh less well calibrated would be the ones that are um at extreme prices. So because these contracts
12:07
Unidentified speaker (speaker_1): they are bounded by uh b b to prices between zero and one dollars because the the contract will pay one dollar to the winner uh at the end uh they they they there's kind of natural cutoffs there. Uh we see a lot of trades and a lot of activity in the extreme prices.
12:27
Unidentified speaker (speaker_1): range. So close to one cents or even below, or uh like 99 cents or or or above. Um and these tend to be slightly well calibrated. Obviously, it's hard to measure. Uh, there's many potential explanations uh for why these are less well
12:47
Unidentified speaker (speaker_1): calibrated. Uh one, uh, there's a long uh standing belief. We don't have any still conclusive evidence uh on that, that that there's a long shot bias among uh retail investors. So they they like uh to to to they're willing to pay a bit more to to to buy a kind of
13:07
Unidentified speaker (speaker_1): lottery type
13:08
Unidentified speaker (UNKNOWN): uh
13:09
Unidentified speaker (speaker_1): Um uh
13:11
Unidentified speaker (speaker_1): kind of tickets or or assets uh we know these types of assets are also for those providing liquidity and taking the other side, they are highly, highly leveraged on the other side. So they are also, there's probably risk management uh uh ideas there. Um the other thing to consider
13:31
Unidentified speaker (speaker_1): when we get into these prices is as I I said, these are fully collateralized. So the capital is locked up when you when you are holding. a token. Uh so if so if say if I want to buy a la long shot, right? So a contract that's priced around one or two cents, uh, the
13:49
Unidentified speaker (speaker_1): counterparty uh has to buy the other position, right? Which is a no at 98 or 99 cents. Uh like the abstracting away from the trade trading fees, right? Because the uh these are cut like create these positions are created by in pairs, right? So so that there's always one dollar
14:09
Unidentified speaker (speaker_1): backing a yes and a no. Um so if I'm providing liquidity or even taking that that uh position, uh depending on the horizon, if I'm taking a 99 cents uh position to with a potential gain of one cent. Uh well I I have to start
14:29
Unidentified speaker (speaker_1): thinking about the uh the the time value of money, about the the interest or the other use I could have for my capital. So that also can distort prices uh slightly when you get to the uh to these uh uh more extreme uh prices. Uh so so that's not something that's fully
14:49
Unidentified speaker (speaker_1): resolved yet in the in the in the academic literature. We I think it's fairly recent literature. We are now starting to look at the data and looking at all of these little the little details. But there's a uh uh a lot of little details like this that kind of are uh
15:09
Unidentified speaker (speaker_1): very interesting and that we need to understand. I'd say overall like like broadly these markets seem to be very kind of well calibrated, uh, especially for the most liquided ones.
15:23
Laura Paglia: Thank you. That's that's very helpful. In your recent study uh released just last month, trades quotes, and price discovery and prediction markets, um, you and your fellow researchers examined
15:37
Unidentified speaker (UNKNOWN): um how
15:38
Laura Paglia: the prices are formed in polymarket.
15:42
Unidentified speaker (UNKNOWN): Um
15:43
Laura Paglia: and uh one of the findings was that the prices frequently change without anyone completing a trade,
15:51
Unidentified speaker (UNKNOWN): uh,
15:51
Laura Paglia: or that trading can be particularly difficult when an outcome is uncertain. So, to your point, uh, that you uh explained that market prices, some market prices are more reliable. forecasts than others.
16:07
Laura Paglia: What does it tell us about the fact that
16:13
Laura Paglia: most pricing changes, according to the research, happen during the minutes when nobody's completing a transaction?
16:22
Laura Paglia: What does that tell us about how information is getting incorporated into the prices? And does it tell us anything further about the reliability of forecasts?
16:36
Unidentified speaker (speaker_1): Yeah. So the reason we started looking at that specifically
16:42
Unidentified speaker (speaker_1): is that in our first paper, as you as you mentioned, And uh we were looking only at trades. And that's what most of the literature, uh the very recent literature, on prediction market, uh uh has looked at because that's what polymarket makes easily available by selling the trades on the blockchain. So most of the research about
17:02
Unidentified speaker (speaker_1): price discovery, or these prices are set, have been looking only at trades. Uh so we got our hands on some Unquotes quotes data for a shorter period. Uh so for us, it was really important to uh start. understanding how does price discovery occurs,
17:19
Unidentified speaker (UNKNOWN): uh
17:19
Unidentified speaker (speaker_1): to see how we should think about and interpret all of these other studies, including ours, that have looked only at trades so far.
17:27
Unidentified speaker (UNKNOWN): Uh
17:27
Unidentified speaker (speaker_1): because even in our sample, it's a very short period. We we we wanna we wanted to know how how far can we learn and about these markets by looking only at the trades. So we we use these quotes to inform us. And the reason we we looked at that as well is that we know from equity markets,
17:47
Unidentified speaker (speaker_1): from like modern financial markets, that most of the prices discovery or the information into prices enters through quotes, through the professional market makers and liquidity providers that are uh uh active on these markets. And our process. prior was that that's what we'd observe as well on on
18:07
Unidentified speaker (speaker_1): polymarket, even though uh as far as we know, there's no uh official market makers or or even like uh professional trading firms that are uh publicly active on on poly Polymarket. market.
18:22
Unidentified speaker (UNKNOWN): Uh
18:23
Unidentified speaker (speaker_1): we know that it there are some people providing liquidity because it is a liquid limit order book system. So you have to have liquidity providers. And from what we know from traditional financial markets, is that to be a successful liquidity provider, you need to be good at it. And by being good, meaning like you
18:44
Unidentified speaker (speaker_1): are good at incorporating information and uh adjusting your prices quickly. So our our our prior intuition. was that we'd find uh a lot of the price discovery occurring to these quotes. Uh we wanted to measure when are trades enough, uh, so that the other researchers, us included, when we
19:04
Unidentified speaker (speaker_1): look at uh uh longer time periods and and uh broader markets, that we can uh have a sense of when can we trust trades, when are trades enough. Uh so that's for from the researcher perspective but the it was also to just to understand are are these behaving like we what we see in in equity
19:24
Unidentified speaker (speaker_1): markets and um the numbers or the ratio might not be exactly the same but for from a journal kind of general practical perspective That these they seem to be behaving in the same way
19:36
Laura Paglia: okay all right thank
19:37
Unidentified speaker (UNKNOWN): you um
19:38
Laura Paglia: In the paper, Who Wins and Who Loses in Prediction Markets, the finding was that the top 1% of profitable trading accounts captured about 76.5% of the gains.
19:54
Laura Paglia: According to your research, what did these accounts actually do differently? Was it that the events that they traded or how they placed their orders? Can you tell why?
20:04
Unidentified speaker (speaker_1): yes so there's uh like we've looked closely at the the top 100 and uh a bit less closely but still closely at the top 1000 uh winners uh in in that sample to like understand what they were doing uh specifically uh differently from from others um and and just
20:24
Unidentified speaker (speaker_1): before i get as to how what they do right what one thing that we didn't even know starting with was to we didn't have any prior expectation as to what numbers should there be. So we we we we we because we we don't have that day type of data on traditional financial markets. So we don't
20:44
Unidentified speaker (speaker_1): know if I'd say look at the TSX or the uh or like more closely maybe the the uh the derivatives exchange from the from the TSX, right? These would be um these are derivatives contract, there's no value creation, it's just risk transfer between between traders.
21:03
Unidentified speaker (speaker_1): Our prior would be that there would be some profit concentration in favor of market makers. They they have to they are providing a service by providing liquidity. We'd expect them to be picking up the the trading gains. And because there's no value creation, uh like there's no uh capital appreciation, like you'd see in the equity markets and
21:23
Unidentified speaker (speaker_1): divorce markets, these are pure kind of uh trading markets. Uh, we'd expect to find some concentration
21:31
Unidentified speaker (UNKNOWN): uh uh
21:32
Unidentified speaker (speaker_1): of of profits, right? The in favor of the market makers. Uh that being said, we don't have that data. Uh, as researchers, we don't. we don't we've never been able to uh to observe that data on uh on equity or divided markets. Uh so that's why we've looked at that specifically on polymarket. That was the first thing we wanted to look at. Uh, because
21:51
Unidentified speaker (speaker_1): now we had a setting where we could look at that profit distribution. And what we found was in line uh with with uh the this uh this the story where uh a lot of the winners.
22:03
Unidentified speaker (speaker_1): uh the large winners they they win uh by trading a lot mostly by providing liquidity and making a small profit on average for every trade right so behave that so even though they're not kind of official market makers they're they're behaving like market makers liquidity providers so
22:23
Unidentified speaker (speaker_1): that's one part of the the winners among the winners we also have large concentrated traders that have made large bets. So the if we take the the main winner, uh the largest winner on on polymarket during our sample period, uh was an
22:43
Unidentified speaker (speaker_1): account that uh made large bets in tree markets for the uh presidential election. uh 2024 presidential election, and was right.
22:56
Unidentified speaker (speaker_1): We also see like users that have made large bets and were wrong and and they they they tend to be fall into our biggest losers. So so so we we do have uh some of these and we we have others that have uh say large put put in large uh bets on a few sports matches and things like that.
23:16
Unidentified speaker (speaker_1): Uh we don't think they they they don't it doesn't seem like these would be mostly kind of insiders or informed like traders that have bet bet like better information uh or I'd say privilege information uh b because they uh like on the presidential election I don't think anyone knew the
23:36
Unidentified speaker (speaker_1): outcome right the the before. Uh when when we're looking at these uh these large trades on sports events they tend to be uh even though they're concentrated there's uh oh most of them are on a multi on multiple matches on multiple teams even sometimes across multiple sports so to us it's looks
23:56
Unidentified speaker (speaker_1): more like either large gamblers or uh potentially bookies or other kind of users that have exposure to these to these matches and wanted to to offset their books, right? We don't know who they are. That's one of the limitation of of the polymarket. We see their crypto wallet addresses, but we don't we don't know who's behind
24:16
Unidentified speaker (speaker_1): these addresses.
24:17
Unidentified speaker (UNKNOWN): Uh
24:18
Unidentified speaker (speaker_1): we we did look at uh because the the there's always the Uh we always get the question as to whether there is insider trading. So like there's no obvious uh kind of patterns of suspicious patterns among our our top winners.
24:32
Laura Paglia: Yes.
24:33
Unidentified speaker (speaker_1): Uh we know the two largest um documented cases uh that have been prosecuted
24:40
Unidentified speaker (UNKNOWN): uh
24:41
Unidentified speaker (speaker_1): of uh insider or in like uh privileged trading on poly market. Uh
24:46
Unidentified speaker (UNKNOWN): the
24:47
Unidentified speaker (speaker_1): first one was a user for uh called Alpha Raccoon who who's a Google employee who had uh information about the end of year kind of uh ranking of search popular search terms for Google and bet on the market related to that. Uh so that that user, we have it in our sample. We
25:06
Unidentified speaker (speaker_1): see he's winning. Uh he got prosecuted for that. Uh, but it's it's not a large winner uh among ourselves. our sample. Like it is the largest documented uh insider trading case on poly market, and he is he ranks uh 115 on our um
25:24
Unidentified speaker (UNKNOWN): uh
25:24
Unidentified speaker (speaker_1): on our sample. There's another one uh that is more in the 300 or something uh ranks. That's it was a uh US uh soldier who had information about uh
25:37
Unidentified speaker (UNKNOWN): I
25:37
Unidentified speaker (speaker_1): think it was the Major. Maduro's uh capture uh and bet on on that event, he was also uh prosecuted and uh captured. But the two of them together, they account for less than 0.1% of the profit that we see in the it's just it's it's so they uh they are very small and they're not
25:57
Unidentified speaker (speaker_1): within the top 100 uh users uh that we see on the platform.
26:03
Laura Paglia: Okay.
26:04
Unidentified speaker (UNKNOWN): Um
26:05
Laura Paglia: when we take your studies together looking at whether prediction market
26:12
Laura Paglia: prices are truly informative, how those prices and how those prices are formed. Currently, beyond elections and sports,
26:24
Laura Paglia: does your research show that? prediction markets can provide useful information that surveys and expert forecasts or other other methods don't provide as effectively?
26:38
Unidentified speaker (speaker_1): Uh so that we haven't looked at specifically. There's a lot of active research uh looking at these because there's there uh taking your question, we have to decompose it into like that to me, there's two parts to that question
26:52
Unidentified speaker (UNKNOWN): the
26:52
Unidentified speaker (speaker_1): first the first one is what's useful information
26:56
Laura Paglia: yes
26:58
Unidentified speaker (speaker_1): so that's a a a matter of taste so that we haven't we haven't looked at at that uh and and then there's the the the there's the question of uh like compared to other ways that we had to obtain obtain forecast and what makes it hard for us to study, and that's why
27:18
Unidentified speaker (speaker_1): we haven't looked at that like in more detail so far in our studies, is that these have to be addressed on the kind of not market to by market, but at least like by by categories. So if you look at markets on crypto, right, the your benchmark and kind of what the what is gonna be the the comparative
27:37
Unidentified speaker (speaker_1): information is gonna be. different. Uh what I I we've seen so far in the in the literature, so not us but others that I've I've looked at,
27:48
Unidentified speaker (speaker_1): they if we look at uh prediction markets, say for earnings announcements versus analyst reports.
27:54
Unidentified speaker (UNKNOWN): Yes. Right.
27:55
Unidentified speaker (speaker_1): Prediction markets tend to uh at least the early evidence shows that the these prediction markets can offer additional information. which seems to be mostly due to the fact that analysts don't revise their forecast every day or every hour of the day. So as as time
28:10
Unidentified speaker (UNKNOWN): progresses,
28:10
Unidentified speaker (speaker_1): there's new information, these markets will uh reflect the uh the information. Um we've seen similar things for uh macroeconomic announcements. Uh so traditionally, we'd have either forecasts of experts for uh macroeconomics.
28:27
Unidentified speaker (speaker_1): uh announcements, or for some measures, say interest rate changes from the from the FOMC uh announcements for the Fed fund rates. Uh, people so like in the past have used the uh FedWatch tool that CME has developed to back out probabilities from the futures prices, right? So
28:47
Unidentified speaker (speaker_1): uh they seem to so far from what I've seen in the like uh very recent uh literature on that they seem to be on par with with with futures uh the main distinction with prediction markets is that uh there's a much broader
29:07
Unidentified speaker (speaker_1): range of contracts so uh then we'd see safe safer futures so it's easier to to uh so so we get more information there that we couldn't like some things that we couldn't infer necessarily uh easily from uh
29:24
Unidentified speaker (UNKNOWN): uh
29:24
Unidentified speaker (speaker_1): from futures such as inflation expectations and things like that. Um and the uh other kind of benefit from from these contracts is it's just the way they are designed, they are really easy to interpret. So if we're thinking about Fed like uh FOMC announcements and the uh
29:44
Unidentified speaker (speaker_1): Fed fund rate uh changes, uh traditionally we've kind of inferred them from the futures prices, right? But they they they they uh we have to bag them out, look at multiple maturity and look at the different change. Uh whereas the prediction markets will make it clear uh directly. directly. If we're thinking about earnings
30:04
Unidentified speaker (speaker_1): announcements, you can look at the stock price, but it won't tell you much, right? You can even
30:11
Unidentified speaker (speaker_1): if you want to hedge whether the announcement is going to be positive or negative, you can't really do that with the stock price. But if you're looking at the uh at these prediction markets, Market, these do allow to like bet on precise forms of information.
30:27
Unidentified speaker (speaker_1): uh, which means that the prices, as long as they're accurate, can are easier to interpret that if you're trying to just figure out uh will uh
30:38
Unidentified speaker (speaker_1): uh such and such company beat expectations or not, just by looking at the stock price, it's almost um uh impossible to infer, right? You've got way too many things uh going on. So so that's the uh so for some of them, yeah, the the early literature seems to point that they they are useful
30:58
Unidentified speaker (speaker_1): uh in terms of providing uh additional information from what we've had before.
31:04
Unidentified speaker (UNKNOWN): Um
31:05
Laura Paglia: thank you professor Gregoire. You've uh looked at these markets very extensively from the perspective of both how these prices move uh and what you've explained, uh you've uh you've been able to show and who who profits to the extent that we can determine that.
31:25
Laura Paglia: What findings surprised you the most?
31:30
Unidentified speaker (speaker_1): Huh, that's a good question. Uh I I'd say we weren't surprised that much by anything because we didn't have any prior. for most of them. We were kind of we were just curious trying to uh look at that. I I think the one thing that surprised us the most after starting working on this project is the
31:50
Unidentified speaker (speaker_1): growth that these uh platforms sustained. So the yes uh our our first research paper, the first version, versioned
31:58
Unidentified speaker (UNKNOWN): uh uh
31:59
Unidentified speaker (speaker_1): the the first draft that we had, we we were uh we had data. until the end of November 2025. So that was our initial sample. One point we figured well since we're gonna up uh revise the paper we'll update our sample so we added four months until uh March 2026 and uh
32:19
Unidentified speaker (speaker_1): when we did our sample and our number of trades quadrupled
32:23
Unidentified speaker (UNKNOWN): uh
32:24
Unidentified speaker (speaker_1): so uh by adding it's just a little four months to our our our sample, uh we kind of quadrupled our
32:31
Unidentified speaker (speaker_1): we figured okay that's the last version for that for for that paper because every uh extra month that we look at now it's kind of exploded and uh over the the summer we we had the world cup as well that brought
32:44
Unidentified speaker (UNKNOWN): in
32:44
Unidentified speaker (speaker_1): a lot of uh uh that here in Canada we're a bit uh insulated from that but I I've traveled a few times to the US in the this year and their prediction market advertisements, it's uh it's everywhere.
33:00
Unidentified speaker (speaker_1): So you they you you yeah you see that and they they they uh yeah and now uh polymarket uh
33:09
Unidentified speaker (speaker_1): during the time of our sample for our paper it was the largest uh platform. It's but it's not legal either. in Canada or in the US, right? That that uh polymarket international platform. So since then, at least in the US, it's was overtaken by Calci as the main
33:25
Unidentified speaker (speaker_1): platform.
33:26
Unidentified speaker (UNKNOWN): Uh
33:27
Unidentified speaker (speaker_1): so we would be interested to look at Calci. Offers it, unfortunately, we don't have the the public data that Calci makes available, uh, isn't as detailed as what we have on the for polymarket. It doesn't allow us to to trade individual users over time uh like we do. But to us, I I I think the most surprising
33:47
Unidentified speaker (speaker_1): wasn't so much in our result, it was more after starting working on this platform on these platforms is how much growth there was uh
33:56
Unidentified speaker (speaker_1): over over the last year. It's been a bit uh uh that was uh uh unanticipated.
34:01
Unidentified speaker (UNKNOWN): Yes.
34:02
Laura Paglia: that way the the market has spoken, so to speak.
34:06
Unidentified speaker (UNKNOWN): Um
34:06
Laura Paglia: Professor Grego, thank you very much for your extensive research. It's incredibly useful and it was uh very helpful to have you join us here today. Thank you.
34:17
Unidentified speaker (speaker_1): My pleasure.
34:18
Unidentified speaker (UNKNOWN): Thank
34:18
Unidentified speaker (speaker_1): you for having
34:19
Unidentified speaker (UNKNOWN): me.
This page was produced automatically: the audio was transcribed with nvidia/parakeet-tdt-0.6b-v3+sortformer-v2.1 and interpreted by Qwen3.6-35B-A3B-UD-IQ3_S.gguf. Transcription and interpretation can contain errors; always confirm against the original episode. Nothing here is investment advice.
All insights