Archive / Crypto & DeFi / E803
Fixed yields dip around maturity, but the 24-hour trade does not pay
Across 90 maturities the next market's fixed yield fell 0.64 points, yet a 24-hour principal-token trade netted about -5.4 bp, not distinguishable from zero, and a yield-token trade lost 166 bp.
On an on-chain fixed-yield protocol, each market expires on a set date and holders who want to stay invested must move into the next market. People argue that this forced roll pushes the next market's fixed yield down, so buying its principal token (the bond-like leg) just before expiry should pay. Across 90 maturities the next market's yield did fall, by 0.64 percentage points, but the fall did not grow with the size of the roll. A 24-hour trade across the maturity was about -5.4 bp after costs, not distinguishable from zero; the yield-token trade lost 166 bp.
When a fixed-yield market expires, holders must roll into the next market, and that forced demand pushes the next market's yield down, so buying its principal token around the roll pays.
Why people believe it
A principal token is bought at a discount and redeems at face value on a fixed date, so its price implies a fixed yield. When the market expires, everyone who wants to keep that fixed exposure has to buy the next market's principal token on the same dates. People argue this one-way demand bids the next token up, and its implied yield down, in a predictable window.
Why it would persist The argument, as stated before testing
People argue that the expiry dates are public months ahead but the people rolling are often passive depositors who move on or near the date, and that the markets are shallow enough for their buying to move price.
For every expired market that had a later market on the same underlying asset, we measured the later market's implied yield from 4 to 5 days before expiry to the 24 hours around it, and subtracted what the same market did on a placebo date 14 days earlier. We then tested two trades: buying the later principal token 12 hours before expiry and selling 12 hours after, and, if the yield had dropped, buying the matching yield token (the leg that receives the yield) and selling 24 hours later. Results are in bp (1 bp = 0.01%). Fees, price impact and gas were charged on both.
At least 60 events from at least 15 underlying assets. Mean yield change at most -0.20 percentage points with the confidence-interval upper bound below zero, and the top-roll-size tercile below the bottom tercile. And either the principal-token trade nets at least +8 bp or the yield-token trade nets at least +150 bp, with a confidence-interval lower bound above zero.
The yield-change interval includes zero, or both trade intervals include zero.
Trade A: twice the market's fee rate times the years to maturity, plus 5 bp of price impact per side and $20 of gas on a $25,000 trade (8 bp). Trade B: 1% for the round trip.
Placebo date 14 days before expiry for the yield change. Confidence intervals are bootstrapped over underlying assets (2,000 resamples), because events on the same asset are not independent.
Both trade confidence intervals include zero (principal token -10.8 to +1.3 bp, yield token -457 to +112 bp), and the yield change also missed the roll-size pass condition: the yield move is real but smaller than costs inside the window.
- The successor market's implied yield fell 0.64 percentage points around expiry compared with 4 to 5 days before, net of the placebo date: median -0.42, 90 events from 41 underlying assets, interval -1.09 to -0.17. The fall is a real feature of the data; 61 of the 90 events (68%) were negative.
- The fall did not grow with the size of the roll. By roll-size tercile of 30 events each: smallest -0.62, middle -0.96, largest -0.35 percentage points. The pre-set condition needed the largest tercile to fall more than the smallest.
- The 24-hour principal-token trade, 12 hours before to 12 hours after expiry, netted -5.4 bp on average over 117 events from 58 underlying assets (median -13.2 bp; interval -10.8 to +1.3 bp) after fees, price impact and gas. The pass threshold was +8 bp.
- The yield-token trade, run only on the 48 events (30 underlying assets) where the yield had fallen by at least 0.30 points, netted -166 bp on average after a 1% round-trip cost (median -220 bp; interval -457 to +112 bp). The pass threshold was +150 bp. A positive mean would have meant the yield fall partly reversed within 24 hours; it did not.
The successor's yield falls around expiry
Change in the successor market's implied yield around expiry, relative to 4 to 5 days before and net of a placebo date 14 days earlier. The interval excludes zero, so the effect is real in this sample; the next chart shows it does not scale with roll size.
Chart data (percentage points of yield)
| Mean | 95% low | 95% high | |
|---|---|---|---|
| All events (90) | -0.64 | -1.09 | -0.17 |
A bigger roll does not mean a bigger fall
Mean yield change by roll size (expired market's value locked relative to the successor's), 30 events per group. If forced demand drove the fall, the right bar would be the longest. It is the shortest.
Chart data (percentage points of yield)
| Mean yield change | |
|---|---|
| Smallest rolls | -0.62 |
| Middle | -0.96 |
| Largest rolls | -0.35 |
The 24-hour principal-token trade after costs
Net return of buying the successor principal token 12 hours before expiry and selling 12 hours after, 117 events. The interval straddles zero and the mean sits well below the +8 bp pass line.
Chart data (bp)
| Mean | 95% low | 95% high | |
|---|---|---|---|
| Net of fees, impact and gas | -5.4 | -10.8 | 1.3 |
Why it failed Interpretation · not established
The yield does move around maturity, by about 0.6 percentage points. On a token with roughly a quarter of a year left, that is about 10 to 20 bp of price, a reading that was not measured separately. Timing within the days around expiry was not measured, and no multi-day version was tested, so no claim is made about one. Inside the 24-hour window the trade did not beat the fees, price impact and gas of getting in and out. The lack of a roll-size gradient suggests the fall is not simply forced roll demand, although this test cannot say what else drives it. The yield-token trade assumed the move would partly reverse within a day and it did not.
Lesson
A visible, statistically clear price move can still be smaller than the cost of trading it.
Scope of this result
Covers Ethereum markets of one fixed-yield protocol with a successor and at least $500k of value locked, expiries 2023-06 to 2026-10, one trade size ($25,000) and a 24-hour window around expiry.
A rejected experiment applies to the tested conditions and does not prove an idea can never work. bp = basis point, 0.01%.
1. Test design
Pre-registeredDataUniverseSignalExecutionCostsControlsSample splitPass criterionKill criterionPower2. Result
ObservationsInstrumentsNet mean95% intervalMedianWorstFinding 1Finding 2Finding 3Finding 43. Interpretation
Why it failedLessonScope4. Reproducibility
StepsGrade