n-way swaps

[ken-underscore]

Have you considered support for n-way swaps? In a world where many stablecoins coexist, this seems like it could be a powerful
tool for increasing the liquidity of stablecoins in the ecosystem. If a user is willing to accept the risk of holding
3 or more different stablecoins then they could benefit from providing liquidity in a way that allows others to swap between any of them.

I've been thinking about this in the context of an AMM-like interface for deploying liquidity onto cardano-swaps. By composing
two-way swaps we can get a uniswap v3 concentrated liquidity like experience with some added benefits like user specific
tick spacing. With n-way swaps, we can allow users to deploy liquidity to stablecoin "pools" that contain as many different stablecoins as they want and profit from the short term volatility between them.

Curious to hear your thoughts on n-way swaps!

[fallen-icarus]

Hey! I did at one point but I couldn't come up with a satisfactory implementation. My philosophy is to compose simple pieces to create more complex features and this seemed like something that should be handled by composing two-way swaps together.

Pooling them all also increase the overall risk: if one stablecoin's peg breaks the others can all be drained.

Did you come up with a simple implementation for this? I want to avoid hard-coding the price between them all to be 1:1 because they aren't actually always 1:1 convertible.

[fallen-icarus]

Three two-way swaps are a bit different because liquidity is trapped within each individual swap, ...

That's not how I see it. If I have 50 DJED -> 50 USDM and 50 USDM -> 50 USDA, what happens if you want to convert 10 DJED to 10 USDA? After composing them, the final state ends up as 60 DJED -> 40 USDM and 60 USDM -> 40 USDA. The liquidity moved between your two-way swaps. As long as the chain exists, you can convert any stablecoin to any other stablecoin with just two-way swaps. It feels like an n-way swap.

[ken-underscore]

Yeah, you can get much of the functionality by composing two-way swaps, but there is at least one important property I think you lose. In your example, the user is unable to set an explicit fee for converting from DJED to USDA using just 2 two-ways swaps, I don't think you always want it to be the sum of the fees of swapping from DJED to USDM and then USDM to USDA.

The important property I think you lose is when someone is looking to swap DJED in order to buy 100 USDM in your example. With the way the liquidity is currently fragmented even though you have 100 USDM available to trade, only part of that can be swapped for DJED, the other 50 USDM you have is locked in the USDM/USDA swap. If instead this was an n-way swap, your 100 USDM would have defined paths to swap for either DJED or USDA, so your liquidity would be able to satisfy the user looking to buy 100 USDM using DJED. N-ways swaps and aggregated stable pools like 3pool increase the efficiency of supplied liquidity by making it more flexible, removing the barrier to only trade along a single path.

[fallen-icarus]

It is an interesting push back, but you are assuming:

1. The collective fees for the multiple two-way swap hops would be higher than the AMM fee + slippage.
2. A supply mismatch won't self-correct due to the higher premium it generates.

TBH I'm not sure I agree with either assumption. If there is a lot of demand for stablecoin conversions, it will create more two-way swaps at cheaper prices. Consider that most LPs today have a fixed 0.3% fee. I can easily envision stablecoin<>stablecoin two-way swaps charging 0.01-0.05%. You would need 6 two-way swaps at the higher range to compare to one swap on an AMM today.

If there isn't enough DJED->USDM, that creates a lopsided premium which provides a higher APY for DJED->USDM suppliers. This should attract more supply or pull it from some of the other pairs.

Will the incentives be enough? IDK. Only one way to find out 🙂

To be clear, I think it is worth exploring how we could support an n-way swap. But I would rank it a low priority unless your assumptions prove to be true.

[ken-underscore]

Hmmmmm, I'm not sure I'm assuming either of those things. I think you are right in both of those cases, I think two-way swaps can outcompete AMM's on fees/slippage and I think natural incentives will combat supply mismatching. I just also think that n-way swaps have a significant impact on the efficiency of stablecoin liquidity over two-way swaps by unlocking them from a single trading path. I'm still wrapping my head around this concept myself, so I'm grateful for the discussion ☺️

[fallen-icarus]

Ah sorry if I misunderstood you 😅

I appreciate the discussion too!

I just also think that n-way swaps have a significant impact on the efficiency of stablecoin liquidity over two-way swaps by unlocking them from a single trading path.

I would need to think about this more. Aside from UTxO contention concerns, I never thought of using multiple two-way swaps as an efficiency disadvantage. 🤔

[fallen-icarus]

An idea I've been exploring is to use CIP-159 to create an AMM where the liquidity pool is inside an account address. I've actually already sketched it out where it would be batcherless, MEV-resistant, and have dynamic fees for LPs that scale with demand.

The basic idea would be to just use the constant-product formula for the two assets being traded, but it might be easy to extend this to n assets. I'm not sure it will work well for volatile trading pairs, but stablecoins are stable relative to each other so a concentrated liquidity pool could possibly work between several stablecoins at once. What do you think about this?

[fallen-icarus]

The AMM smart contracts can't see the exact balances at execution time. They only see the balance intervals the user sets when they build the transaction. The smart contract uses the bounds as the exact balances at execution time. So a sandwich attack isn't possible. If you insert a trade before mine, you move the actual balance but my bounds are hard-coded into my transaction. When my transaction then gets processed, it still uses my bounds as the asset balances which means it executes as if your transaction didn't exist. Users effectively hard-code their market prices by setting the appropriate interval bounds. It is just like my limit orders, you can't change my execution price to give me a worse price. The worst you can do is censor me.

When calculating an example fee of 3%, I assume the 2% comes from the user constructing a tx 1% above market price when the price actually drifts down 1%?

Not quite. As in my previous paragraph, the smart contracts can't see the exact balances of the assets at execution time so it is forced to use the interval bounds. So imagine I want to buy asset A with asset B. I'll set the lower bound for asset A to live_balance - 1% and the upper bound for asset B to live_balance + 1%. These are the numbers actually used in the constant-product formula for each asset. The AMM formula the has its own slippage base on these values of 1%. So it is:

1% constant-product slippage + 1% lower bounds margin + 1% upper bound margin = 3%

You have to set the safety margin of 1% on either side because buyers and sellers could get processed before you and could invalidate your interval bounds if the live balances move outside your range. If you think the market is trending in one direction, you could skew your bounds to one side like live_balance for asset A and live_balance + 2% for asset B.

The most interesting part about this design is the impact on impermanent loss. Studies have come out that showed most LPs of concentrated liquidity pools would have done better by just holding the asset. The problem is the fees used to offset impermanent loss don't scale with volatility. These surplus "fees" actually do scale with volatility (you have to widen your safety margin in your interval bounds) which means this could actually be the first AMM to adequately compensate for impermanent loss. You could swap out the constant-product formula for any other pricing strategy and the surplus fees work the same way; they scale with expected volatility.

[ken-underscore]

Yeah, I see why a sandwich attack on the user that is swapping isn't possible, but I don't see why it isn't possible to steal the surplus that is intended for the LP's as fees. If a pool has 200 asset A and 200 asset B and a user looking to buy asset A sets the lower bound for asset A to 198 and the upper bound for asset B to 202 it seems like the SPO making the block can look at the actual balance of the pool and decide to create a tx that buys some asset A, then process the user's trade, then create a tx that sells some asset A, perfectly extracting the surplus.

Hm, it would certainly be interesting if fees could mitigate the impermanent loss from volatility, ultimately though being an LP is a bet for a sideways market, so I assume there is a limit to how much you can change that. Edit: I wonder how much of the fact that most LP's would have done better holding the assets could be mitigated by a better understanding of the type of investment being an LP is. When thinking about ada/stable pools, many users probably believe that in the long term the price of ada will significantly increase relative to usd. Perhaps a better long term investment for them would be a strategy that starts with only stablecoins and accumulates ada as the price drops, then benefits from volatility as the price moves in that range. This strategy means even if the price of ada were to rise significantly the user wouldn't experience an impermanent loss they feel forced to realize so they don't lose more value to ada price appreciation.

[fallen-icarus]

I don't see why it isn't possible to steal the surplus that is intended for the LP's as fees.

Hmm... that is a good point but this should be easy to socially police. Unlike on Ethereum with its global state, all you need is the current Cardano block. What are the odds a block has all its trades cancel out for zero surplus? The probability of five or ten independent trades all hitting their worst-case bound exactly to the decimal point is effectively zero. That is economically unnatural. We don't even need to identify the individual traders. If an SPO consistently makes blocks whose trades cancel out, that is very likely to be MEV. And it is trivial to detect. So wallets could have a banner like "Your SPO is acting maliciously! Please delegate to a new pool."

So in a way, this is like an MEV honeypot. End-users are unaffected and stealing surplus from LPs is trivial to catch. When Linear Leios goes live, it would still be easy to police since we can track who proposes an endorser block. If the same SPO consistently proposes endorser blocks whose trades cancel out, that is very likely to be MEV.

I know some people think social policing doesn't work, but I disagree. That is literally how optimistic rollups work. This is optimistic MEV protection. Wallets just need to nudge users to move their delegation.

What do you think? If Cardano's SPOs are actually decentralized, will this actually be an issue? We don't need to stop all MEV; we need to stop at least enough for LPs to still meaningfully profit compared to existing AMMs. 🤔

[ken-underscore]

I do think it's possible that social policing works, but I'd be afraid of depending on it to work. I still remember when Cardano staking went live and the discussions about whether it was socially acceptable for some SPOs to run multiple pools or if they should consolidate to a single pool to spread power among more independent operators. At least for me, it isn't easy to predict how these social interactions will play out.

Some potential problems are private pools which aren't beholden to delegators and also pools which promise to share the value they extract with their delegators. There is also potentially anxiety for LP's who may always wonder if SPOs are skimming a little bit off the top, trying to take just enough so they aren't noticed.

My personal take is the concept is still great even if it were to use an explicit fee instead of an intrinsic one based on surplus. If the AMM is using concentrated liquidity then the drift and slippage may contract to a point where over the long run an explicit fee might not be such a bad deal 🤔

[fallen-icarus]

I would like to run this social policing experiment because it would at the very least give empirical data on it. The conditions have to be right, but if Ouroboros can be based on a social trust assumption (51% honest majority) why can't this? Even if 40% of the surplus is skimmed by MEV, that would still be a better result with its dynamic fees. I would only be willing to use explicit fees as a last resort because static (or slow to update) prices are usually a bad idea in economics...