Aave V4: Building Better DeFi

Transcript

[Music] Okay, mic is on. It's time. We can start. Hi guys. Uh my name is Alio. Uh thanks for being here um in this venue. I am happy to be here in ITC and happy to talk about what we are working on these days as a labs as part of the aid. Um I am VP of engineering uh at A Labs which means that I'm supervising everything that is uh D def

DI related with the uh within A and we are part of the DAO as service providers and we got in charge uh last year of building the new protocol iteration. So I already gave a presentation about uh a V4 on a stable summit and uh it was relatively short but still uh I managed to talk about a lot of stuff. I think what's also interesting to talk about is like how uh did we came up with a new iteration of the protocol? challenges we have been facing and in general the thought process behind uh trying to improve a protocol that has been uh now live for

many years. So I guess we can move forward uh to the next slide. So what is a who doesn't know in here what a is raise your hand. No it looks like it's the right crowd. Everyone knows what a is. So uh just to give a quick refresh uh a is the largest uh uh D5 protocol at the moment with over 40 billions in uh in uh assets uh under management. Uh the protocol has been live uh for uh over five years. It's been going through three different iteration. we had a v1, v2 v3 and uh each iteration including the last

one v3 has been like built up over the original model. So uh why is that? Uh why we have always been iterating over the original model for a variety of reason. First is a model that works. It's a model that scales. It's the model that allowed a to become like the largest uh D5 protocol and also uh we have been observing how the protocol uh worked over the years and trying to identify uh what could be improved and what innovation we could bring to the table right so we started with a v1 which was released uh shortly after compound v1 or v2 I don't remember now was released

uh v1 had some major implement some major improvements uh over the existing D5 protocols. One of the major uh features that we was released was flash loans. Probably not everyone knows but a was the first protocol introducing flash loans. Uh are they good? Are they bad? I think now everyone agrees that flash loans are good. like because like at the beginning we were hearing all the time about uh price manipulation hacks used uh through flash loans. Uh but uh turns out this actually made the ecosystem much stronger. So now you don't hear about that anymore. Why? because everyone learned how to actually handle

uh economic attacks that like uh flash loans a lot capital for everyone but also made uh a lot of uh DeFi more efficient, made the liquidations more efficient, made uh capital transfer more efficient, made uh moving positions across DeFi more efficient. So now we have a flash loans that work under the hood and people don't even know that they're actually using flash loons and at the same time uh they improve uh they are a net improvement for the whole space. Then we had V2 uh in V2 we had uh optimized uh design optimized architecture with less gas cost. Uh we had introduced credit delegation which

allowed users to um use their position to delegate credit to others. So people could borrow uh using my collateral uh up to the threshold that I would decide to so I could uh delegate credit to other people and add better improved granular management. All these three were coming uh from observing how V1 was working. And then we had a v3. A v3 is an an improvement over a v2. Uh which I would say the major feature was absolutely emo. Uh essentially what we have been noticing uh especially with the rise of uh staking derivatives is that there was a

lot of demand to leverage um correlated assets. Uh so therefore we came up with a solution to u allow users using LSTs as collateral to uh leverage with higher borrowing power if they were willing to only borrow a certain asset classes. So price correlated assets that was absolutely amazing feature of V3 was recently improved by BGD labs u extending its capabilities further so making it even more flexible. Uh but there were also other improvements like isolation mode which allowed to list longer tail assets uh with reduced risk exposure for the protocol and like a marginally improved

liquidation uh engine over uh the previous iteration. So why do we need V4? to be honest like when I uh when we finished the uh development of V3 and we released it I was 100% convinced that we would not need a V4 and there were a variety of reason as well I thought we had everything we needed to actually uh um become the largest proto which we actually did uh and like there was built-in approability so I thought like okay anything that we need uh we can definitely add over time.

But uh in the last three plus years in which V3 has been live, we have been observing how continuously observing how a tree works and how um uh what could be improved, right? So one of the first things that we improve we we noticed that could be improved is that the complexities around releasing uh new markets. So uh what a v3 is is basically when you deploy the a protocol what you deploy is a market. What we call market is like a set of smart contracts that allow to supply certain assets and to borrow against them and which enforces collateralization uh constraints uh provides uh liquidation mechanisms and

so on and so forth and provides the yield uh distribution infrastructure for people only looking for yield. Right? So uh every time uh a DAO tries to launch a new market we face the same challenge that the market starts from zero and you need some somehow to provision liquidity to actually bootstrap the borrowing demand right so that's a major challenge that we face and at the same time you look maybe at the a core market on Ethereum and you have I don't know a couple of billions of stable coins sitting there as idle liquidity uh not generating any yield and that that was already the first uh bell ringing in our head. Um

uh the monolytic architecture also of av3 because essentially being a monolyic market this make it so that there is um no separation between the liquidity provisioning part and the uh borrowing part. So because of that like you only have one way to borrow assets from a you use a collateral which is needs to be an ERC20. So if you want to use uh different assets classes you need to find a ways to tokenize them. And uh you can only collateralize these assets to be able to borrow right. Uh and another challenge that we noticed uh o over these three years is uh difficulties for users that are looking for uh different

risk and yield profiles compared to what have to offer. So a design is basically most in most cases like one a market per network which provides provides the best risk adjusted u uh yield for the vast majority of users. But there is also class of users who are you looking for uh different tiers of risk and yield. Right? So it's uh with the current design is hard to tailor uh uh solutions for this category of users. Right? So uh another problem is that that we have is that the borrow rates are uniform across all the collaterals. What this

means is that whenever an asset gets collateralization power power on a V3, it will pay the same rate as any other asset. So independently of what kind of risk this collateral brings to the system. So if I'm uh uh borrowing used uh using uh tokenized uh LP share with volatile assets inside which brings uh some kind of additional smart contract risk and pricing risk. I will pay the same in the same uh rate as a borrower that is borrowing using ETH. Uh which is a problem because this is not how real life works. So in real life uh rates are used tailored uh on the degree of risk that you bring to the lender. So if uh

the lender perceives a risk for a borrower higher, he will demand probably higher rates, right? So uh what did we do to improve this? First of all, we decided to move completely away from the architecture of uh a v3 and we redesigned the prop the protocol completely from the ground up. So the first step was to separate the liquidity provisioning part from the borrowing part. So uh we built a liquidity provisioning engine which is completely abstract away from how uh liquidity is borrowed. Uh the only thing that the liquidity provisioning layer does we call it liquidity hub is to keep

accounting of who has provided uh what and what depth has been taken and how much depth must be returned including the interest right. Uh these are some interesting uh uh consequences. The first one is that now instead of having one single borrowing strategy, we can have multiple borrowing strategy plugging uh all in the same liquidity layer. So we call uh borrowing strategies uh spokes. So we have this hub spoke design. Uh so we can have a spoke that behaves exactly as a v3 but we can also have a spoke that I don't know let you borrow from uh unis swap.

This is actually something that's going to happen. And we created a proposal on the unis swap dowo to implement a spoke that will plug into a v4 and while your collateral is in unis swap generating trading fees you can actually borrow from a and this is something that with v3 you cannot do or maybe you can but like it's quite hard to do because the architecture is not designed to do that while in a v4 will come for free but that's not only that you can imagine custodians that will build their own custom spoke and they will uh uh plug into their own custodianship uh infrastructure into AB. So the coll the collateral stays with the custodian. Uh

the infrastructure enforces collateralization risk and the users providing liquidity to this custodian are actually able to borrow. So the whole goal of this is to actually optimize demand and go and tap into areas that are untapped today for uh unlocking liquidity today cannot be easily unlocked. So that's the first goal. So here we have a more visual aspect in comparison between V3 and V4. You can see uh V3 is like vertically integrated with like uh uh no segregation between the liquidity provisioning part and the borrowing part while uh a v4 as more modular with the

possibility of segregating like borrowing strategies from the liquidity provisioning. Uh this has al also another interesting um consequence that allows us to solve another problem that we want to to solve which is uh bootstrap uh new markets for new asset classes more easily. So what you can do is for example create multiple hubs and these hubs also uh can interact with each other. So uh we have here a representation of how emode uh for those who not familiar with emode in a v3 is the um is the feature that allows you to borrow with higher collateralization. This is how emode

will be structured in in v4. There actually not be uh any emote specific anymore, but there will just be uh multiple hubs with just uh with the same code base but just a set of disc different uh uh risk parameters where uh those that are functioning in an equivalent way as EMO today will just have higher collateralization uh uh thresholds and uh listing correlated assets. But the interesting part is that you can also plug uh hubs into each other. So uh let's say that we bootstrap a new Atheina emode hub that starts from zero. But we can also plug this hub into

the main hub where the majority of the liquidity is and there is maybe a lot of idle liquidity and therefore user using this atina mode hub they can actually start borrowing even if there is no liquidity in there because they draw liquidity from the main hub where the liquidity is available. And this has even another benefit which is now we have the possibility uh uh for users that are looking for higher yields to supply actually liquidity directly against Atheina collateral if that's what they're looking for to actually earn more yield compared to the core market.

So at the same time we solve uh the liquidity uh segregation issue that like having multiple parallel market brings uh because when you instantiate a new one is always empty and we also solve the problem of giving ease access easy access for users to different risk and yellow profile without uh sacri sacrificing on UX and capital segregation. Uh we also have how much time do I have? Still five minutes. Good. Um, we also have another feature that is going to be core for the protocol which we call risk premiums. Uh, risk premiums aim to solve the issue that I mentioned before where all collaterals pay the same borrow APR.

So on the left you see V3. If I try to borrow go, I will borrow go at 10% whether I'm using Atheina or I'm using Lido SD E or I'm using ETH. Uh these three assets have a different risk profile because ETH being the most decentralized and liquid one. Um Lido uh has additional smart contract risk and redemption risk compared to if Athena is a completely different asset class. They should not pay the same borrow rate. Right? So in uh uh V4 we will have the capability of setting a base rate which is the rate that the users with their lowest risk tier collateral will pay and then other

assets will actually have pay a premium on top and that premium will be determined by how much additional risk these assets bring to the system. So for example you will have east paying 10% uh lido paying 12 and Athena paying 12. Uh so that's aims at solving the u majority of the uh issues that we identified around um uh uh liquid capital efficiency and uh risk pricing. Right? Then we also have other features that we implemented. Uh one of the major ones is like aims to solve one of like really annoying problem that we have today. we always had with all the

iterations of a um that is the fact that you cannot easily change uh collateralization factor for collaterals once it's assigned because changing the collateralization factor affects all the users. So if you drop the collateralization factor of an asset too much you might cause liquidations for example. This is very painful because every time uh risk managers have to do an adjustment uh they cannot react as in in the same um with the right measures but they need to always be mindful of the fact that collateralization factors cannot be changed that drastically.

Therefore it creates a lot of governance overload because there are a lot of proposals to to create uh due to this maintenance uh and it it um brings suboptimal uh risk management. So what we did in before is that uh we have this new feature called um dynamic risk configuration where you can essentially change the uh the collateralization factor of an asset only for new users uh keeping all the users unaffected. This is uh very important because this way like we can reduce the governance overload, offload the risk management to risk managers completely without changing trust assumptions because there is no way of

forcing previous user uh um to get liquidated. You can still change previous configurations by going through governance too. So the previous uh feature is still available if that is needed. For example, whenever you need to off to offload an asset. Uh I only have two minutes left. Let's see what I have here. So very very quickly reopening the design space. This is something I'm very excited about. Uh the a new architecture brings a lot of possibilities to build on top compared to a uh the current a iteration where the design space on top is relatively um stagnant at the moment.

uh with the new architecture we will actually manage to bring a lot of innovation on top of a you can bring new borrowing strategies uh uh you can build position managers with another feature that we implemented uh to facilitate u um uh positions man position man management automated position management you have will have a reinvestment engine that allows uh the protocol to reinvest uh idle liquidity for better yields and uh potentially we can have creators on top that actually move liquidity across markets to optimize yield.

So conclusion a before like we really it's not just V3 on steroids like V3 was V2 on steroids is completely new D5 primitive. It aims uh to a more structural and comprehensive way of uh having uh onchain yield and onchain borrowing. Um and it's designed to meet uh the needs of both uh DeFi users uh our famous Digents and also institutionals because like the features that we have uh allows maximum flexibility for any kind of configuration. Also I just want to add is not in the slides but the code is super simple, super small, much much simpler than a V3 much easier to deploy

permissionlessly. Uh so we will have a lot of fun in the next months I think. That's it. I don't know if I have time for questions. No, I don't. Unfortunately, I don't. But yeah, that's it. Thank you guys. [Music]