Prediction markets for distant events are less accurate than those for nearby events. The standard explanation is opportunity cost: money locked in a long-horizon prediction contract earns nothing while it waits. Traders avoid long-horizon markets, liquidity drops, and prices drift from the correct probability. This is the long-horizon problem, and it is widely considered severe.
Fountain (arXiv:2602.21091) finds two surprises. First, the problem is smaller than expected. In agent-based simulations with LLM-driven traders, the accuracy difference between 4-day and 2-year horizons is 0.72 percentage points — far less than prior theoretical and empirical estimates suggested. The market's price is already close to correct even with reduced participation.
Second, paying interest on open positions — making the prediction contract also function as a yield-bearing instrument — eliminates 83% of this already-small accuracy gap. But not by making the price more correct. By making more traders show up. Participation triples from 17% to 62% of available wealth. The mechanism is not correction of bias but mobilization of capital.
This is a diagnostic inversion. The long-horizon problem was framed as a pricing problem — the market gets the probability wrong because not enough information is incorporated. The fix reveals it was a participation problem — the market had access to the right information but not enough capital to express it. The price wasn't wrong because traders were misinformed. It was imprecise because too few traders bothered.
The general point: in information-aggregation systems, the bottleneck is often not the quality of individual signals but the fraction of signals that enter the system. Small incentive changes that increase participation can improve aggregate accuracy more than large improvements to individual precision. Getting more opinions matters more than getting better opinions — when the opinions are already approximately correct.