PART 1 OF 4

What is a Prediction Market and how do they work?

Live: Mississauga's mayoral election

Live embed from embed.polymarket.com, "Mississauga Mayoral Election Winner". Needs an internet connection when presenting; the next slide is the static fallback.

Betting on events is nothing new: a short history

Rhode and Strumpf, "Historical Presidential Betting Markets", JEP 18(2), 2004 (1868 to 1940, 1916, and the decline: polls, crackdowns, pari-mutuel racing legal in New York from 1939). British betting shops legal from 1961; US Wire Act, 1961. Forsythe, Nelson, Neumann and Wright (1992) on the IEM. Intrade: CFTC complaint 26 November 2012, trading suspended 10 March 2013. CFTC order against Polymarket, January 2022. BNN Bloomberg, 12 September 2024 (Kalshi). Wealthsimple newsroom, 18 June 2026. Nevada licensed bookmaking in 1949; the federal Professional and Amateur Sports Protection Act, 1992; Murphy v. NCAA, 14 May 2018; Bill C-218, in force 27 August 2021; AGCO and iGaming Ontario, 4 April 2022.

Who will be Elected Mayor of Mississauga in 2026?

Every Prediction Contract starts with a Question

The Basic Components, Part I

The Basics, Part II

The Basics, Part III

1 · CONTRACT CREATION

• User deposits $1
• Gets a YES and a NO
• The dollar stays at the exchange

2 · SELL THE OTHER SIDE

• Sell the NO for 25¢
• Sell YES at the same time? get 75¢.
• Sell both at "spread" (26/76) \(\to\) earn 2¢

3 · SETTLEMENT

The $1 goes to the winning side

The losing side is worth nothing

YES + NO = $1, always. Buying NO is the same as selling YES.

Every YES in the market has a NO behind it, and a dollar at the exchange behind both.

Polymarket splits a deposited dollar into one YES and one NO token and merges them back on demand; Kalshi creates a contract when a YES order and a NO order match and holds the $1 until settlement. Illustrative prices, before fees.

How a contract comes into being (Polymarket)

Because there is always $1 for a contract to exist, settlement is guaranteed

Polymarket and Kalshi run order books

The Basics, Part IV

PART 2 OF 4

Understanding the price

The Polls are tight, why isn't the market?

https://www.poliwave.com/ca/tor/polls

Happily, all you UofT alumni/ae know that
% of vote \(\not=\) probability of winning

THE QUESTION THE MARKET ANSWERS

Will the lead still be above zero on 26 October?

today: +3.2

Chow ahead: 70%

Bradford ahead: 30%

−10

0

+10

+20

Chow's lead on election day, in points

RULE OF THUMB

• margin of error on the lead is 1.8 per candidate \(=2\times1.8=3.6\)
• standard error of the lead \(\frac{3.6}{1.96}=1.84\)
• compute \(\Phi\left(\frac{\text{lead}}{\text{std error}}\right)=\Phi\left(\frac{3.2}{1.8}\right)\approx 96\%\)

• Polls: the lead is small, but on sampling error alone Chow wins 96% of the time.
• Polymarket: Chow 70%. The market sees more risk than sampling error.
• That 70% implies polls miss the lead by about 6 points, three times the margin of error: turnout, who answers polls, late swings.

Polls fromhttps://www.poliwave.com/ca/tor/pollsö Chance of winning = Φ(lead ÷ typical miss). Sampling error alone: the margin of error on the lead is 2 × 1.8 = 3.6 points, a standard deviation of 3.6 ÷ 1.96 = 1.84, so Φ(3.2 ÷ 1.84) ≈ 96%. Polymarket, [date and time]: Chow 70%. For a lead of 3.2 points, a 70% chance implies a typical miss of 3.2 ÷ 0.52 ≈ 6 points, three times the sampling error. In US elections, final polls miss by about twice their stated margin of error (Shirani-Mehr, Rothschild, Goel and Gelman, Journal of the American Statistical Association, 2018).

How good are prediction markets at predicting?
(paper by Akey, Grégoire, Harvie, and Martineau (2026))

Long shots: Bürgi, Deng and Whelan (2026), "Makers and Takers", CEPR DP 20631; Akey et al. (2026), Table 5. Optimism: Bartlett and O'Hara (2026), "Adverse Selection in Prediction Markets: Evidence from Kalshi", 41.6 million trades; 67.5% of single-name volume settles NO. Zelensky market: Decrypt, 9 July 2025; settled by the UMA oracle after a disputed first ruling. Netflix market, 20 January 2026: Bartlett and O'Hara (2026); makers lost $75,754.

market-implied probability of an event was 65%

How good are prediction markets at predicting?
(paper by Akey, Grégoire, Harvie, and Martineau (2026))

PART 3 OF 4

What are the issues?

 No crowd, no wisdom - a market without liquidity is not a market 

Leanne Caron: 51%

bid 47¢, ask 55¢

Example first seen in talk by Professor Charles Martineau.
Polymarket, "Guelph, ON Mayoral Election Winner", API as viewed 19 October 2026, 9 p.m. ET: last trades Caron 48¢, Chew 27¢; 24-hour volume $21. Mississauga and Toronto quotes: Polymarket, 14 and 19 October 2026. Bellwether (Hall and Paschal, Stanford): fewer than 1% of listed markets carry meaningful prices.

The value of the info is only as good as the market that delivers it

Markets need arbitrage

• Yesses (and Nos) must sum to 100 (or more)

• Yes+No per contract must sum to 100 (or more)

PredictIt, "Who will win the 2020 Democratic nomination?": bids summed to 110 to 125¢ for a year. Zitzewitz (2026), Figure 3.

There are well-known behavioral biases: long-shot

Kalshi: YES on a person or a company settles YES 4.5 points less often than its price says. Bartlett and O'Hara (2026), Figure 6.

Long shots

People overpay for long shots: YES at 5¢ happens less than 5% of the time. Under 10¢, buyers lose 60%+

Optimism

People prefer buying yes and overpay for yes

The spread

Quotes at 69¢ and 71¢: a buyer pays 71¢ for a 70% chance \(\to\) people lose on average

Long shots: Bürgi, Deng and Whelan (2026), "Makers and Takers", CEPR DP 20631; Akey et al. (2026), Table 5. Optimism: Bartlett and O'Hara (2026), "Adverse Selection in Prediction Markets: Evidence from Kalshi", 41.6 million trades; 67.5% of single-name volume settles NO. Zelensky market: Decrypt, 9 July 2025; settled by the UMA oracle after a disputed first ruling. Netflix market, 20 January 2026: Bartlett and O'Hara (2026); makers lost $75,754.

A skill test on 1.7 million Polymarket accounts: a spike of skilled accounts on the left, a spike of reliably wrong ones on the right, and a flat middle that trades at random. Gómez-Cram, Guo, Jensen and Kung (2026), Figure 2.

3% of accounts show skill

70% of the volume comes from the 97% who show none

46% of the skilled are still skilled next period. Fund managers: 10%

The winners post limit orders. The top 0.1% supply 47% of their volume as makers; the bottom 95% supply 17%. (Akey, Grégoire, Harvie and Martineau); a small minority makes the price right, and the crowd pays for it.

Gómez-Cram, Guo, Jensen and Kung (2026), "Prediction Market Accuracy: Crowd Wisdom or Informed Minority?", SSRN 6617059: skilled 3.2% of accounts; the crowd is 97% of accounts and about 70% of volume; 46% of the skilled stay skilled out of sample against 10% for mutual-fund managers; insider-like accounts are under 1% of volume. Akey, Grégoire, Harvie and Martineau (2026), SSRN 6443103: maker shares by PnL group; the most profitable users' contracts resolve 20 points better than their prices, the worst 5% of losers' 13 to 18 points worse.

The problem: people lack skill (but think they don't)

69%

of users lose money

The typical user loses about $2

Half

of all gains go to the top 0.1%

The top 1% take 77%

But should we care? Prediction Markets are Person to person, no bookie .... zero-sum before fees.

In practice the money flows to "professionals" like bots: $28 million traded, $135,000 profit. Half a cent per dollar \(\to\) profits derived from scale

Akey, Grégoire, Harvie and Martineau (2026), "Who Wins and Who Loses in Prediction Markets? Evidence from Polymarket", SSRN 6443103, version 18 October 2026: 2.4 million users, $67 billion, November 2022 to March 2026; median loss about $2; top 0.1% of winners take 51% of gains, top 1% 77%; top 1% of losers bear 68% of losses. The largest single accounts (a gain over $22 million, a loss over $10 million) likely hold the other leg elsewhere; the paper does not see off-platform trades. Account profile: polymarket.com, as viewed 19 October 2026.

The result: people lose on average

A well-known insight from gambling (Jonathan Cohen, Losing Big: America's Reckless Bet on Sports Gambling): 60% of people contribute less than 1% of profits to gambling platforms, 3% contribute over 80%.

PART 4 OF 4

So is this just old vice dressed as new?

What's so special about markets?

HAYEK (1945)

Knowledge is decentralized. Prices centralize this info.

IOWA ELECTRONIC MARKETS, 1988 TO 2004

74%

of the time the market was closer to the final vote than the polls

The only other way to get centralized knowledge is to run surveys or surveillance

• Surveys need to know whom to ask.
• A market lets people who know come to you.
• A planner gathers the information, then decides. A market lets the decision follow the price: the farmer reads the futures price before planting.

Hayek, F. A. (1945), "The Use of Knowledge in Society", American Economic Review 35(4), 519 to 530. Berg, Nelson and Rietz (2008), "Prediction Market Accuracy in the Long Run", International Journal of Forecasting 24(2), 283 to 298: five US presidential elections, 964 national polls.

Forecasts improve decisions

Weather

19.7 inches of snow in Central Park, February 2026. Utilities hedge heating days on CME; Kalshi lists city snow and temperature.

Housing

Thousands of realtors know where the market is heading; housing starts on the ground come faster than official data; allows families to plan

Interest rates

A floating-rate borrower buys "the Bank hikes". It pays when the mortgage bill rises.

Politics

Alberta separation, tariffs, a change of government: risks no insurer sells, now with a price.

Sports

A World Cup exit knocks the loser's stock market down the next day. Even a game is an economic event.

Most of the trading is speculation. The part that is real can still matter.

If you have a gigantic market and most is doodling, it can still provide real utility to 1%

Final Thoughts

Can't talk about anything without talking about AI...

Bot traders

• Bots already trade on Polymarket and Kalshi.
• They take advantage of any inefficiencies, trade across markets and beat you every time
• They can read public news fast.
• They can do maths faster and better than humans.

AI as a forecaster

• Language models rival, but do not yet beat, the best human forecasters in tournaments.
• Humans can beat AI where judgement is required, but that set is shrinking

AGI & trading

• We will likely see market manipulation attempts by AI agent swarms
• AI agents may add significantly to price discovery by using public info, and the market provides incentives to have bots collect the info
• AI bots still cannot aggregate the decentralized info that is in people's heads
• It remains to be seen if AI agent trading will generally improve markets

Bots as traders: Gómez-Cram et al. (2026) on fast public-information traders; Bartlett and O'Hara (2026) on one-sided flow in thin single-name markets.

Reading about prediction markets

Andreas Park, "Prediction Markets Are Good, Actually", Maclean's, 10 June 2026, as opened on 18 October 2026. Banner at the bottom: OLG, Ontario's provincial lottery corporation, Lotto 6/49.

... and the province does not miss its chance

Same article, same evening: a full-screen OLG ad for Cash Pop, "Play in-store today or on OLG.ca". Provincial lottery corporations lobby for tighter limits on prediction markets (Globe and Mail, 2 September 2026).

The public debate (read or listen at home)

• Andreas Park (University of Toronto), "Prediction Markets Are Good, Actually", Maclean's, 10 June 2026. Regulate them as securities, with limits on engagement-driven design.
• Charles Martineau (University of Toronto) and Marius Zoican (University of Calgary), "Put plainly, prediction markets are not for you", The Globe and Mail, 25 September 2026. Efficient prices mean an expected loss after fees.
• Capitalisn't, "Are Betting Apps Engineered for Addiction?", Bethany McLean and Luigi Zingales with Jonathan D. Cohen, Chicago Booth Review, 9 February 2026. Sports betting apps, prediction markets, and where investing ends and gambling begins.
• Jonathan D. Cohen, Losing Big: America's Reckless Bet on Sports Gambling, Columbia Global Reports, 2025.

Opinion pieces, a podcast and a book. Opened 18 October 2026.