Why knowing the future doesn't always help predict markets | The Economist

Study Methodology

• Conducted by Elm Wealth, an investment firm, involving 15 trading days of simulated trading using 15 historical Wall Street Journal front pages (2008–2022).
• Participants were provided with "tomorrow's news" (future headlines) and allowed to trade stocks and bonds with leverage up to 50x.
• Each participant started with a baseline of $50, with trades settled daily against actual historical market moves.

Layperson Performance

• Only approximately 50% of participants ended the 15 days with more money than they started.
• One in six participants lost their entire stake ("went completely bust").
• The average return was negligible, with the median participant finishing with approximately $51.
• Directional accuracy was low, with participants correctly guessing market movement direction only 51% of the time.

AI Performance

• Artificial intelligence models outperformed humans slightly in directional prediction, achieving roughly 60% accuracy.
• AI models failed similarly to humans in bet sizing, taking excessive risk and failing to calibrate leverage based on confidence levels.
• Some AI models incurred losses despite the advantage of having perfect future information.

Professional Macro Trader Performance

• Five recruited expert macro traders all finished with profits.
• The group's average return was more than double their starting capital.
• The decisive factor in their success was not superior prediction accuracy, but their ability to dynamically adjust bet sizing based on confidence levels.
• Professionals employed a strategy of avoiding bets entirely on low-confidence days and aggressively increasing leverage on high-confidence days.

Key Structural Findings

• Market Pricing: Markets often move based on expectations rather than raw data; positive news can lead to price declines if the data falls short of prior market consensus (e.g., 100k jobs added when 150k was expected).
• Bet Sizing Deficits: Both humans and AI struggle to scale bet sizes according to probability or confidence, often over-leveraging on uncertain outcomes.
• Rationality vs. Accuracy: Markets can be "rational and completely wrong"; traders can profit by betting on market momentum even if the underlying prediction of economic reality is inaccurate.

Investment Implications

• The most critical question in investment is identified as "how to size my bet" rather than "what to buy."
• Successful investing requires the ability to quantify confidence and adjust leverage accordingly, a skill predominantly held by professional macro traders.
• Most investors devote excessive focus to asset selection while neglecting the magnitude of their exposure relative to conviction.