When Prediction Markets Can (and Can't) Help Us Make Smart Choices
If you’ve been hanging around The Audit for long enough, you’ll know that I’m always on the lookout for cool datasets. In particular, I love data that can be used to represent public sentiment in creative ways. (For some dark and poorly-understood reason, I hate opinion polls.)
If you ever need to make financial, educational, employment (or retirement) decisions, then you should share at least some of my enthusiasm.
You might already have joined me on my travels through Reddit comments sentiment analysis, setting AI models loose on detailed demographic data, reverse-engineering insider trading activities, or my recent post on using Canadian Election Study survey data in eccentric ways. The common thread of all of those efforts involved looking for new ways to understand Canadians through creative data analysis.
So a new API provided by Delphi Markets caught my attention. The API allows you to use an AI model to analyze the complete datasets of multiple prediction markets - like Polymarket and Kalshi - with a single prompt. That gives me a lot of analytic power for assessing the betting decisions of thousands of people.
WARNING: As I’ve previously written, it’s important to remember that placing bets on a prediction market platform from within Canada could be treated as a criminal offence. Enforcement has so far been spotty, but it’s been argued that the practice could expose you to prosecution for either binary options trading or unlicensed gambling.
At any rate, the goal of analyzing bets taken by other people would be to identify trending market sentiment as it’s applied to Canadian economic, social, or political events. More specifically, I’d like to be able to measure the accuracy of prediction market sentiment against the stuff produced by high-cost government consultants and officials.
First though, let’s see if we can use historical market activity to measure how successful prediction market predictions have actually been. There have definitely been some spectacular failures:
2025 Canadian federal election (the Liberal party comeback)
Prediction markets at one point heavily favored the Conservatives and Pierre Poilievre. On several large markets, Liberal victory odds reportedly fell into roughly the 10–25% range before rebounding sharply after the leadership vote and tariff-related political shifts.
The eventual Liberal win therefore represented a pretty obvious “market miss” relative to the earlier consensus. This is particularly telling because:
Millions of real dollars were invested by users of various prediction markets, so this wasn’t just some random online poll.
The market clearly underestimated the party’s leadership replacement effects, widespread anti-U.S. nationalism, and voter consolidation dynamics.
Carney becoming Prime Minister quickly
Markets initially treated Mark Carney entering frontline Canadian politics as unlikely or speculative. But once leadership transition pressure accelerated, the repricing happened very quickly. Traders who entered early on “Carney PM” or “next Liberal leader” style contracts likely captured substantial returns, but most early players simply missed it.
Early resignation pressure on Justin Trudeau
For much of 2024, prediction markets still assigned meaningful probability to Trudeau remaining Liberal leader through the next election cycle. Like most of us, they obviously missed growing internal caucus pressure, donor fatigue, and the panic caused by some awful electoral math.
Canadian housing / Bank of Canada rate timing markets
There were several periods where markets overestimated “higher for longer” persistence. When growth slowed faster than expected, most betting contracts - even in markets with six- and seven-figure cross-platform liquidity - were caught incorrectly anticipating delayed easing or economic recovery.
Immigration-cap reversal markets
Markets (and “expert” political commentary) initially underestimated how aggressively the government would reduce caps on international students, temporary residents, and immigration-growth. Dominant market sentiment instead voted against visa restrictions, population-growth slowdown, and sharp policy tightening.
No one is claiming that these large, skin-in-the-game markets are supposed to be perfect, but they are expected to deliver predictions that outperform opinion polls or common knowledge.
Having said that, in many of those same predictions, markets were quicker to correct to changing conditions than many pundits and experts. But in the context of Canadian events, those benefits will mainly show up at the margins.
However, information from prediction markets could be particularly valuable where information is fragmented or politically distorted, official forecasts are unreliable or slow, and existing incentives reward optimism or opacity. Here are some real possibilities:
Immigration and refugee system backlogs
Here’s a sector that’s long suffered from chronic forecasting failures, “politically constrained public discussion” (translation: “You’re a racist and a bigot for even using the word immigrant!”), and poor operational transparency.
Prediction markets - fueled by inside information - could potentially forecast:
refugee-claim inventory levels,
IRB processing times,
asylum approval rates,
temporary resident population levels,
or shelter utilization by claimants.
There are people both inside and outside government who know that information. Some of them could be willing to put their knowledge to work.
Housing completions and construction bottlenecks
Canadian housing policy - especially programs involving First Nations - is full of aspirational targets that frequently conflict with reality and seldom produce useful status reports.
Prediction markets could therefore be useful for:
housing-start targets,
condo completion rates,
municipal approval timelines,
infrastructure delivery,
or labor-shortage constraints.
These are markets where the expertise of developers, lenders, contractors, and municipalities, can be profitably tapped.
Major Canadian government IT projects
Canada has a…mixed…record on federal IT-projects. See Phoenix, ArriveCAN, PrescriptIT, and IRCC digitization delays for details. Large complex and unwieldy tech projects always attract unhappy (and chatty) contractors and engineers. And they can be much more reliable sources of information than senior management.
Housing stress interactions
CMHC and other official agencies do provide useful data and analysis covering the housing market. But it doesn’t always arrive as quickly and in much detail as we’d like. Lenders, brokers, insolvency trustees, and real-estate professionals on the other hand, all have access to localized information that’s not immediately visible through official statistics. Which could make prediction markets especially attractive destinations for any insiders willing to put some money on the line.
I think you get the idea here. You should feel free to use your favorite AI model to dip into the wisdom of the prediction markets yourself. But I hope to test it out for myself for at least a few topics in the near future.
For now, Canadians buying market bets on these platforms might be exposing themselves to legal risk. But the fact that so many Canadians seem willing to take that risk presents an opportunity for the rest of us.
