defi-research

Going to give a *sneak peak* about this paper w/ @ksk and others to FC before Twitter!

There's been a surge of interest in intent-based systems like @uniswapX or CoW Swap — their claim has been to help improve the acquisition of off-chain price data on-chain

But do they work?

Ph.D. student @berkeley_eecs. Research @gauntletnetwork. Geometry, dynamics, markets & intelligence.

The largest onchain marketplace. Buy and sell crypto on Ethereum and 11+ other chains.

In general, there is often the advice that you get in auction mechanism design of, "it is better for revenue and/or welfare to have more bidders in an auction than to have a better reserve prices/parameters" 

In fact, this is the content of the Bulow-Klemperer theorem!

ヽ(⌐■_■)ノ♪♬ @ @gauntlet/🤖Ventures/tldr/Aera \\ ¯\_(ツ)_/¯ Struggling with the nature of randomness \\ main: @gaa \\ /red-bull /defi-research maxi

But does "more solvers (bidders) = better welfare for the user" for intents? In our paper, we find that this is *paradoxically* not always true! In fact, a user using an auction like @uniswapX may actually want to *purposely limit the number of solvers* who find prices for them

Does this mean intents are broken? Nah

But it does say that you need to be careful when you design your intents — if you aren't cognizant of the microstructure you create, you might actually make the intent market worse than the fully on-chain market (e.g. a Uniswap V3 CFMM, in the case of UniswapX)

So why would a user want *fewer* solvers competing?

A problem arises when solvers incur
a) costs in order to be able to participate in the intent market
b) amount they have to spend to get best execution prices off-chain

A solver in an intent market faces the 1st cost, they face *costly entry* whereas the 2nd cost is *costly effort*

Costly entry: Have to buy a lot of servers and colocate w/ both ETH validators + Binance in order to effectively bid in intent auctions

Costly effort: To get the best possible price on Binance, I need to spend money to ensure that my slippage/impact costs are low

In some cases, you can't *avoid* costly effort

Concrete example — Consider two scenarios for a user:
1) Buy $1,000,000 of ETH
2) Buy $1,000,000 of WIF

1st: Many solvers are long ETH and don't have to buy ETH from an exchange to fulfill (low costs)

2nd Case: Few MMs hold WIF so they have to buy it — if they all do it at the same time they face congestion costs

This example shows that intent solvers face congestion costs due to inventory constraints — if they hold inventory of an asset, they don't have compete with other solvers for procurement but if they don't, then they all simultaneously hit the same off-chain venue and cause the execution price to get worse

This notion of 'congestion cost' due to lack of inventory gets passed on to the user: the solvers bid less aggressively

This type of 'collusive' behavior where all bidders bid below their true value for an item is known as 'bid shading' (a common problem in ad auctions)

This notion of 'congestion cost' due to lack of inventory gets passed on to the user: the solvers bid less aggressively

This type of 'collusive' behavior where all bidders bid below their true value for an item is known as 'bid shading' (a common problem in ad auctions)

https://en.wikipedia.org/wiki/Bid_shading

The difference between intent systems (eg @anoma, @uniswap, @flashbots, etc.) and ad auctions, however, is that ad auctions have 
*explicit* collusion

In intent systems, the collusion is 'implicit' — its because solvers congest each other, worsening welfare

The difference between intent systems (eg @anoma, @uniswap, @flashbots, etc.) and ad auctions, however, is that ad auctions have 
*explicit* collusion

In intent systems, the collusion is 'implicit' — its because solvers congest each other, worsening welfare

https://pubsonline.informs.org/doi/10.1287/mnsc.2019.3457

How bad can congestion be? We show that if the congestion costs (e.g. slippage + price impact on off-chain venues) are sufficiently high then you can have oligopoly (e.g. it is better for a user want fewer solvers participating to get better prices)

As @sui414 and @angelfish have shown, this occurs in practice!

But defining oligopoly is a bit delicate: even if solvers congest themselves such that fewer than the maximum amount participate in equilibrium, is this _really_ worse for the user?

To understand this more carefully, we need to define oligopoly formally

Oligopoly is defined as two contemporaneous things:
1) If there are n possible bidders, then only k = o(n) bidders participate in equilibrium 
- i.e. k/n → 0 as n → ∞ (% of solvers that participate goes to 0)
2) Users get worse prices as n increases (so they get better prices with fewer bidders)

We show that oligopoly exists, for many types of costs and prices, if solvers simultaneously face costly entry and costly effort — which means that intent markets may benefit from *restricting entry* (e.g. being *less* decentralized)

This is kind of surprising! But read the paper for more (+ ask questions!)

In general, there is often the advice that you get in auction mechanism design of, "it is better for revenue and/or welfare to have more bidders in an auction than to have a better reserve prices/parameters" 

In fact, this is the content of the Bulow-Klemperer theorem! https://www.cs.tau.ac.il/~fiat/mdsem12/amd05.pdf

It depends, but the problem with partial fills is you make the UX worse (the user has to wait / there's a higher opportunity cost of capital for the solver), so its not a free lunch

Really interesting findings. I wonder how much the ability for solvers to commit to partial fills would help here with the congestion costs?

In a fully permissionless intent system (so nothing that is live right now) where they are slashed if they don't deliver on the intent within some time period, they have to procure _some_ inventory to avoid getting slashed. The real problem is that if they all do this simultaneously they get worse prices

Would like to know how solvers incur congestion fees due to the lack of inventory, even before they secure the exclusive rights to the orderflow by winning the auction. Don’t they merely simulate the execution such that congestion fees don’t price in?

Awesome! Thanks for sharing here.

5000 $degen.

Tag @unjumble to bookmark, see bookmarks onframe.

iOS app for personalised updates shipping soon.

Follow /unjumble, DM @uncledavo if you need anything.

Interesting that there are congestion costs when solvers source inventory from the same off-chain markets. Did you also consider congestion cost from multiple solvers competing to fill an order on-chain in the same block?

Would you be open to share how simulations where made?

This is cool and makes total sense! 

In practice(cowswap) inventory procurement probably happens post-auction-win. So no congestion 

Winners are responsible for submitting solutions on-chain. They can force the solution to post end-of-block giving their off-chain procurement time to settle. Timing games are fun