But long-term structural patterns take far more capital to arbitrage and your position can be deep under water for years at a time.
There just isn't that much patient capital.
This isn't necessarily true. There are demonstrable, persistent calendar effects that can be reliably traded. Such trading can be thought of as a service being provided to market participants for a fee (e.g. balance sheet as a service).
One example of a calendar effect that is likely not easily monetized is the 'Christmas effect' for VIX [0] due to decreases in trading activity.
The treasury market reflects demand for 'risk-free' securities, which functionally serve as bank accounts for corporate entities without access to bank reserves on the Fed's balance sheet. When market participants expect rate cuts in the future, long term assets with a locked in yield become more attractive to hold relative to short tenors, as the short dated treasuries need to be rolled more frequently.
Additionally, it may be that the curve inversion itself causes greater expense in short term funding (e.g. overnight), leading to potential contractions in economic activity.
There's no arbitrage in the strict sense, that's clear.
But there's none in the statistical sense, either. Yield curve inversion doesn't happen all that often, how are you going to confidently bet that what happened the last few times will happen again this time, given that n is so low and there's always the issue of regime change?
Expectations of Fed rate hikes and cuts are reflected in the fed funds futures market, where participants estimate future values of the fed funds rate. Currently, the market implies a 94% chance of at least one rate cut by December [0].
[0] https://www.cmegroup.com/trading/interest-rates/countdown-to...
[1] https://www.google.fi/amp/s/amp.ft.com/content/855d23e2-83b7...