Thursday links: Election betting, the true purpose of prediction markets, Druck does perps

The pre-history of US election betting

Kalshi CEO Tarek Mansour says next week’s midterms will be the “first prediction-market election.” But a report from the Council on Foreign Relations reminds us that Americans have been betting on elections from the very start.

In 1796, for example, Federalist partisan Joab Griswold wrote that he had bet “more pounds than I am worth pence” on William Cooper winning reelection to Congress by 1,000 votes.

Cooper lost by 18 votes, which must have bankrupted Griswold.

(Note: His exact relation to Clark Griswold was impossible to verify at press time.)

By the 1880s, election betting had been "institutionalized," with specialized bookmakers setting odds from offices on Wall Street. Betting volumes soared. One study estimates there was $6 of betting for every vote cast in the general election of 1888 (in today’s money).

In 1896, election betting is believed to have surpassed the volume of trading on the New York Stock Exchange.

In the early 1900s, the Tammany Hall political machine maintained a war chest of funds dedicated to manipulating betting odds in favor of its preferred candidates. An historian described the practice as a form of “indirect propaganda.” In 1904, the New York Times accused Republicans of manipulating betting odds in favor of Teddy Roosevelt.

Election betting peaked in 1916 when an estimated $10 million was wagered — more than double what the campaigns themselves spent that year. Major newspapers published the odds almost daily in the run-up to Election Day.

But when George Gallup correctly predicted Franklin D. Roosevelt’s 1936 landslide, it was the beginning of a quick end for election betting. Newspapers replaced betting odds with Gallup’s polling results.

The final blow came in 1939 when New York legalized betting on horse races, which proved to be a much more popular way to lose money. By the 1944 election, New York’s election bookmakers had all shut down.

Now, they’re back! Prediction markets have already brokered more than $1.4 billion of bets on next week’s US midterms.

They might just be getting started, too. Betting on the 2028 general election is expected to surpass $10 billion.

Don’t bet and vote

In 1874, New York amended its state constitution to bar anyone who had bet on an election from voting in it: “No person who…shall make or become directly or indirectly interested in any bet or wager depending upon the result of any election, shall vote at such election.”

It’s still on the books!

Article II, Section 3 of the New York Constitution — revised as recently as 2002 — retains that exact wording.

If you live in New York (and Wisconsin, too), you can bet on an election or you can vote in it. You cannot do both.

Not legally, at least.

I’m guessing no one will cross-check your Polymarket account against the voter rolls in New York and Wisconsin.

They might, though!

In 1864, Abraham Lincoln's campaign reportedly tried to disqualify Democrat voters by luring them into making minor wagers on the election. Lincoln won by 10 percentage points in the end, so the scheme turned out to be unnecessary.

This year, though? There is only one competitive race next week where the gambit might work: The Cook Political Report rates the NY-17 congressional vote as a toss-up.

So, if you’re a supporter of either Michael Lawler or Cait Conley, consider luring someone on the other side into a bet.

(On Polymarket or Kalshi, of course. You don’t want to disqualify your own vote, too.)

The real purpose of prediction markets?

A new paper finds that the best way to teach an AI agent to forecast real-world events is by training it on prediction markets.

Researchers built an “agentic forecasting environment” with more than 2,100 resolved questions from Polymarket and gave the agents access to the internet to research them.

The innovation here is the researchers’ filtering system, which prevents agents from accessing information published after a cut-off date set at least seven days before each real-world market was resolved.

This forced the agents to “learn to forecast by learning to search”  — while searching only the information that would have been available at the time.

It worked. After specialized training, the open-weight Qwen models the researchers trained were as much as 40% more accurate than the same models before training.

The trained models also outperformed every frontier model the researchers tested. The models did best on the hardest questions. And they made their predictions using just 5% of what it cost the frontier models to make theirs.

All of which makes me think prediction markets might be training themselves out of a job.

Just as the real purpose of the web has turned out to be training LLMs to think, the real purpose of prediction markets might turn out to be training them to predict.

Druck sees the future of trading and it’s tokenized

Stanley Druckenmiller is most famous for breaking the European Exchange Rate Mechanism by shorting more of the British pound than the Bank of England was able to buy. That was in 1992.

In March 1999, he lost $600 million shorting $200 million of dotcom stocks. By the end of the year, he had made it all back, and more.

In March 2000, Druckenmiller lost $3 billion on a $6 billion long he bought at the top of the market.

"You asked me what I learned,” he said later. “I didn't learn anything. I already knew that I wasn't supposed to do that. I was just an emotional basket case and couldn't help myself.” 

In 2008, he made a killing on the mortgage securities he’d been shorting since 2006.

Over 30 years, his hedge fund, Duquesne Capital Management, posted an annualized return of 30.2% — one of the very best track records of all time.

He’s still trading.

In 2023, he made something like $500 million betting on Nvidia (just before that became an obvious thing to do).

In 2026, he’s buying into Hyperliquid.

Earlier this year, Druckenmiller bought PURR, a digital asset treasury company that owns HYPE tokens.

This morning he announced an investment in trade[XYZ]:

What makes a 72-year-old trading legend think perpetual futures traded on blockchains will enhance his investment process?

He might just want to trade on weekends.

On Friday, Oct. 16, 1987, Druckenmiller made “one of the most tragic mistakes” of his trading career when he went 130% long US equities thinking they were due for an oversold bounce.

He quickly realized his mistake: “Over the weekend, after studying trading charts,” he said later, “I knew I was wrong.”

He resolved to sell everything and flip short as soon as he could, which meant waiting for the stock market to open on Monday — Black Monday, when the Dow Jones set a record by falling 22.6% in a day.

It turned out OK: “On Monday morning, I took advantage of a very brief rally to sell the equities that I had bought on Friday. I finished the week with a profit due to those early Monday morning trades.”

He would have had a much larger one if there were perps to trade.

— Byron Gilliam

_{Brought to you by:}

JLTXX, our second tokenized money market fund, brings a registered government money market fund structure to public Ethereum for eligible qualified investors.

Backed by our initial $100MM investment, JLTXX is designed to invest in a manner that supports stablecoin issuers under the GENIUS Act.

All of finance is coming onchain and the institutional heavyweights are headed to DAS London this November 10-11.

Grab your tickets for unmatched access to the leaders bringing institutional finance onchain.