Finally caught up on crypto twitter π A few thoughts on this AMM debate:
1) The math itself seems plausible based on volatility harvesting results (though I haven't checked the details). The interpretation is a little optimistic though imo. 1/7
2) Kelly criterion (max E[log V]) makes sense in context of an entire portfolio and effectively heavily penalizes possibilities of portfolio wipeouts. But when only modeling a component of a portfolio, it makes less sense. 2/7
To illustrate, a Kelly optimal portfolio could very well be: keep x% in safe assets w/ the rest spread over very risky bets. You wouldn't want to Kelly maximize each of the very risky bets. In effect, the safe x% removes the heavily penalized wipeout possibilities itself. 3/7
3) The results assume nice GBM. When you have less well-behaved processes, like jumps, the story may be very different. I suspect jumps are where AMM rebalancing costs are especially bad b/c the AMM offers liquidity at all prices spanning the jump distance. 4/7
4) Kelly optimality results (when applicable--see above) are on infinite time horizon. The absence of other risks from the model (e.g., stablecoin failure risks) should lend some skepticism to interpretations. 5/7
e.g. if either assets in the AMM pool were to fail, the entire pool -> 0, which makes it (in this way at least) *more* fragile than other strategies. This doesn't invalidate the results per se, just means be skeptical of asymptotic model results. 6/7
Altogether, I do think the work is interesting, and I tend to like AMM, ergodicity economics, and rebalancing ideas. Though I don't think it's robust enough to be making investment decisions based on it π 7/7
β’ β’ β’
Missing some Tweet in this thread? You can try to
force a refresh
For non-custodial stablecoins, it's more complex. Where they rely on central governors, they may have fiduciary risks. But where non-custodial stablecoins aim to align incentives of agents decentrally and remove custodial and fiduciary points, then the risks are very different
Non-custodial risks are more like blockchain and market manipulations, which are issues outside of any given stablecoin. Makes more sense to pursue such manipulators for attacks, hacks, frauds as opposed to the smart contract coder (unless they're the same!)
#CropRotation: Since the new governance change, the most profitable $COMP farming is now to borrow and re-deposit #Dai. With multiple layers of borrow-redeposit, you can get >40% yield still. The $BAT positions have massively unwound (w/o incident!) and replaced by Dai
As with $BAT before, this leads to concerns about deleveraging effects. Demand for Dai is high with price close to $1.02 and Dai interest rates are already low. Dai issuance is struggling to keep up with COMP farming demand in this otherwise 'normal' time