When Knowing the Future Makes You Go Broke
Expensive Lessons from Haghani & White (2024)
Let’s Play a Game
I give you some money, a crystal ball, and a time machine.
The rules. Each round, you land on a random date between 2008 and 2022.1 You’re shown the front page of the Wall Street Journal from the next day (prices blacked out; it’s a game, not a cheat). You read. You bet on the S&P 500, on 30-year Treasuries, on both, or you skip the round. Maximum leverage allowed: 50x. You close your position at the next day’s close.
15 rounds. 45 minutes. Go.
The Results
118 finance students played this game with real money on the table (only $50, but you’ve got to start somewhere).
They know what a Fed decision does to the yield curve. They know why a strong dollar weighs on multinational earnings. They’ve spent years learning exactly what they needed for this game, so we should expect a strong performance.
And they delivered: Average return: 32%. Only 4.5% lost money.
Just kidding. Average return: 3.2%. 45% lost money. 1 in 6 went bust. With tomorrow’s front page in their hands.
Across the 2,000 trades placed by the 118 players, the directional hit rate was 51.5%. A coin flip.2 The information was there, they just couldn’t read it, on average (1 in 5 still managed to double their capital). Bad reading, fine. But how did they bet on it?
On 30% of trading days, players used leverage above 20x. And the correlation between bet size and actual hit rate across days was 0.628. At least they knew exactly when to bet big.
Just kidding, again. The correlation is zero. They had no idea about the quality of their own convictions, or what to do with them.
So the full picture is (on average): they couldn’t read the information they were handed (literally the future), and they bet big at random without even knowing they couldn’t read it.
It would have been interesting to run the same experiment with professional macro traders for comparison. But guess what, that’s exactly what the authors did.
Let’s Play a Game, Again
They invited five seasoned macro traders to play. The head of trading at a top-5 US bank. The founder of a top-10 macro hedge fund. Former seniors from Jane Street and from the Treasuries desk of a primary dealer. Let’s call them the veterans.
None of the veterans finished in the red. Average return: 130%. Median: 60%. No, I’m not kidding this time.
Two things separated the veterans from the students.
First, they read the paper better. 63% directional accuracy, against 51.5% for the students. The difference between a coin and an edge. That’s what you build over twenty years of watching how markets actually react to the news.3
Second, they knew when to skip. On roughly a third of trading opportunities, they did nothing. On the rest, they went big.
“Soros taught me that when you have tremendous conviction on a trade, you have to go for the jugular. It takes courage to be a pig.” — Stanley Druckenmiller
The 118 students had tomorrow’s paper and they burned it. The five veterans had the same paper and they made 130% in fifteen days.
Information is just an input. It tends to get stuck in the bottleneck between the screen and the chair.
Tomorrow, the input will be even more abundant. But the bottleneck won't widen on its own.
We are our own edge. Always have been.
Take care,
Flo
Want to try it for real? The game actually exists, you can play it here (relax, the money is fake). Let me know how it goes.
Source: When a Crystal Ball Isn’t Enough to Make You Rich, Haghani & White (2024)
To make the game interesting, the days were drawn from a set where one third are employment report days, one third are Fed announcement days, and the last third are picked purely at random.
I can already feel some of you wanting to quote Jim Simons: “We’re right 50.75 percent of the time... but we’re 100 percent right 50.75 percent of the time. You can make billions that way.” But we’re not in an algorithmic trading context here, so hold your horses.
The problem with taking seasoned macro traders and testing them on the 2008-2022 period is that they got seasoned during that same period. Run the same game on 1970-1985 and the gap would probably narrow.