Loading the Numbers
Reading the published files. Nothing is shown until they arrive.
TidelineSame pages and numbers in both modes. Easy uses everyday words; Pro has the full detail.
TidelineWhat this page answers: when an app raised its fee in the past, did people stop swapping?
UnderpoweredNot Enough Data Yet
Elasticity −0.06, 95% interval −1.64 to 1.51, from 21 of 28 fee changes in the 408 days of chain data, 1 Oct to 12 Nov. Fee-tagged (token_diff) swaps only.
At −1, a higher fee leaves revenue flat. Below −1 it lowers revenue; above −1 it raises it.
Pass-through 0.33: the share of each fee change users pay (1 means all of it). Interval −0.43 to 1.09, also underpowered.
We looked at 21 times an app changed its fee (out of 28 we found in 408 days of chain data, 1 Oct to 12 Nov) and checked whether people swapped less afterwards. Our best guess: they barely reacted (−0.06, what economists call ). But the data is thin. The real answer could be anywhere from −1.64 (people leave fast) to 1.51 (people swap even more). The line that matters is −1: past it, raising the fee loses money. Our range sits on both sides of that line, so we can’t say yet. Apps seem to pass about a third of a fee change on to users (0.33, ), though it could be anywhere from −0.43 to 1.09.
Where the Answer SitsOur Best Guess and Its Range
Dot: the estimate. Bar: the 95% interval. Pale band: effects too small to tell from chance.Dot: our best guess. Bar: where the real answer probably is. Grey band: too small to tell apart from luck.
0 fees do not change volume. −1 volume falls as fast as the fee rises, so revenue stays flat. Left of −1, a higher fee loses revenue.0: fees don’t change how much people swap. −1: swapping drops as fast as fees rise, so income stays the same. Left of −1: raising the fee loses money.
Checks
We tried 9 other ways of measuring this, plus random dates with no fee change. The random dates looked no different from the real fee changes. Every way gave a best guess between −0.18 and 0.06, and 8 of 9 ranges cross −1. So the methods mostly agree.
Placebo Dates
10,000 random draws of dates with no fee change gave p = 0.91 (lowest possible 0.0001).
Drop the largest frontend (1click-lifi)
−0.00, interval −1.67 to 1.66, 20 fee changes.
Fee shift 2 bps instead of 5
−0.04, interval −1.05 to 0.98, 21 fee changes.
Fee shift 10 bps instead of 5
−0.09, interval −2.48 to 2.29, 21 fee changes.
Leave out events whose windows touch 28 to 30 Oct (the move to the treasury)
−0.06, interval −1.64 to 1.51, 21 fee changes.
CoW Swap on DefiLlama as the control
0.06, interval −0.95 to 1.08, 21 fee changes.
1inch on DefiLlama as the control
−0.03, interval −1.07 to 1.01, 21 fee changes.
0x on DefiLlama as the control
0.03, interval −1.01 to 1.08, 21 fee changes.
KyberSwap on DefiLlama as the control
0.03, interval −1.01 to 1.06, 21 fee changes.
LI.FI on DefiLlama as the control
−0.18, interval −1.24 to 0.88, 21 fee changes.
Placebo Dates
10,000 random draws of dates with no fee change gave p = 0.91 (lowest possible 0.0001).
Drop the largest frontend (1click-lifi)
−0.00, interval −1.67 to 1.66, 20 fee changes.
Fee shift 2 bps instead of 5
−0.04, interval −1.05 to 0.98, 21 fee changes.
Fee shift 10 bps instead of 5
−0.09, interval −2.48 to 2.29, 21 fee changes.
Leave out events whose windows touch 28 to 30 Oct (the move to the treasury)
−0.06, interval −1.64 to 1.51, 21 fee changes.
CoW Swap on DefiLlama as the control
0.06, interval −0.95 to 1.08, 21 fee changes.
1inch on DefiLlama as the control
−0.03, interval −1.07 to 1.01, 21 fee changes.
0x on DefiLlama as the control
0.03, interval −1.01 to 1.08, 21 fee changes.
KyberSwap on DefiLlama as the control
0.03, interval −1.01 to 1.06, 21 fee changes.
LI.FI on DefiLlama as the control
−0.18, interval −1.24 to 0.88, 21 fee changes.
28 Frozen Fee Changes28 Fee Changes We Studied
Each tick is one of the 28 fee changes that pass the rules in PREREG.md (21 usable. Left out: 6 with under $50,000 in the 7 days before; 1 because a window has under 4 days, runs past the data, lacks a control day, or another change of its own touches it).Each line is one time an app changed its fee (21 had enough data to use).
Tideline is research, not financial advice.Follow Us on Telegram (opens in a new tab)