The question
When a PT reaches maturity, its holders can redeem it for the underlying, worth one unit of the accounting asset per PT. Some of them buy the next maturity on the same asset. Does that show in the next maturity's implied rate?
Method
Data: Pendle's daily implied APY for every maturity this service has recorded, from April 2023 to October 2026: 475 maturities with readings.
An event is a shorter maturity of an asset expiring on a chain while the nearest longer maturity of the same asset on the same chain has a reading seven days before the expiry and lives at least 30 days after it. For each event, the change in the longer market's implied APY from seven days before the expiry to fourteen days after it, 21 days in all.
Control: 21-day windows of markets with no sibling maturity expiring within 14 days of the window and at least 30 days of life left after the window's seventh day, one window per market every seven days, grouped by the market's age, because a new market's rate moves differently from an old one's. Each event's change is compared with the control of its own age group.
Result
- Around an expiry (110 events): a median change of −0.75 points after the age control (−0.76 before it); the rate rose in 29 of the 110 events, 26%.
- Control (759 to 1,766 windows in each age group): a median change between −0.05 and +0.01 points; the rate rose in 47% to 54% of the windows.
- Where the longer market was already 45 days old or more (7 events): a median of −0.64 points; the rate rose in 1 of the 7.
- One asset in detail: USD.AI on Arbitrum. Two of its rolls count as events, in February and June 2026, and the next maturity of USDai and of sUSDai fell in all four cases: by 2.75 and 3.90 points in February, by 1.16 and 0.64 points in June. Its November 2025 roll is not counted: the next maturity opened six days before the expiry, so there is no reading a week before it.
A possible mechanism
Holders who redeem a matured PT and buy the next one add buyers of PT at the same moment, and buying PT raises its price and lowers its implied rate. The data are consistent with that; they do not prove it.
Limits
- The events are not independent: several share an asset, a period, or both.
- 110 events over three and a half years is a small sample, and 7 in the older-market group is very small.
- The control windows overlap (21 days long, one every seven days), so they are not independent cases either.
- The figures describe the past. They are a base rate for what has happened, not a forecast for any market.
- This is the run of 6 October 2026 of the study's code in the author's service, which reruns on the same data (python3 -m pendle.studies roll; the code is not public yet). An earlier run on 26 September 2026, by a script with a looser rule for the reading a week before, found 121 events and −0.83 points.