- The speaker is a co-founder of Keyrock, a large algorithmic trading firm in the DeFi industry.
- They believe that as power users, they can provide valuable insights and learnings from their experience.
- They initially focused on providing liquidity to decentralized finance (DeFi) platforms, particularly automated market makers (AMMs) like Uniswap.
- They found that AMMs are not necessarily better than traditional market making, but they provide better access to market making intelligence.
- They explored different ways to provide liquidity in DeFi, including order books, automated pricing systems (like Uniswap and GMX), and RFQ systems.
- They specifically focused on automated pricing systems, particularly Uniswap v3, and developed strategies to minimize impermanent loss and maximize fees.
- They also experimented with derivative products built on top of Uniswap, such as perpetuals, and explored new innovations like split market making functions and interest rate swaps using oracles.
- They emphasized the importance of data, latency, and risk management in DeFi trading.

- On-chain data is free and can be obtained at a low cost or through paid providers
- Real-time decision making is best done using an RPC node, while data analysis and historic tracking can be done through an API provider
- Hosting your own node provides access to real-time data and main pool data, allowing for more informed decision-making
- Execution in decentralized finance (DeFi) requires good quality and can be managed through a server communicating with an RPC provider or by injecting code into your own node for faster execution
- Being fast in program execution is more important than trip latency in DeFi
- Tracking on-chain data can help prevent mistakes and vulnerabilities in protocols
- The options market in DeFi is experiencing massive adoption and growth
- Both order book-based systems and AMM systems have value and complexity, with no clear preference for one over the other
- The speaker is open to experimenting with and supporting innovation in the industry.

Learn more about TOKEN2049: https://token2049.com
Follow us on X: https://x.com/token2049

Keynote: Journey of a DeFi Power User

Speaker:
Juan David Mendieta, Co-Founder and CSO @ Keyrock

Stage: KuCoin Stage

#token2049 #keyrock #defi #marketmaker #algorithmictrading

I'm not gonna talk too much about Keyrock today because I felt that the point of this presentation should be to add value to the audience.

But in case you don't know too much about Keyrock, so we are a large algorithmic trading firm, we're a market maker, we have a multitude of services, and if you wanna learn more about them, please get in touch.

But for today, what I want to talk about is something else.

I want to be able to give a little bit of value about what has been our journey as a DeFi power user.

So around 2 years ago, we started on a mission to understand where this technology is taking us and what we see is that a lot of people in the industry are builders.

They are building protocols and majority of presentations and information out there is from the point of view of builders.

We are power users so we are the other side We're using this tech.

I want to share with you today some of the learnings that we had from that experience.

So I'm gonna start with something very basic.

So of course when we look at decentralized finance, the first thing that we tried to figure it out was how do we provide liquidity to this thing?

As a lot of you know, there is the concept of AMMs, correct?

And just before I go into the presentation, I think it's important to take a step back and really understand what the fuck is an AMM.

So I don't know if you ever had a situation where you were in university or maybe in high school And you were taught for the first time calculus and how an integral would work you were probably taught that the area under the curve of a derivative of of an integral will be equivalent to an infinite sum of pieces.

And that's basically the exact same difference between an order book and a function-based market maker.

It's just that 1 is a discrete version of the other.

1 is a function, the other 1 is discrete.

But you can actually generate the same with both.

So we look at that and then we conclude, actually the reality of AMMs is that they are not an improvement on market making.

They are an improvement on access to market making intelligence, which is slightly different, because they are not actually better algorithms, but they do provide a huge value in terms of access.

With that, we started to experiment and, you know, like enter into the industry, understand what type of liquidity systems are there.

What we realize is that today we basically can provide liquidity in 3 different ways in DeFi.

You can go to an order book where they do exist, it's just that the settlement is on chain, correct?

You can go to what we call automated pricing systems.

So we don't call it AMMs, we call them automated pricing.

Inside of automated pricing we'll have 2 types.

We'll have function-based systems like Uniswap and Oracle-based systems like GMX.

And last but not least we'll have RFQ systems with a market maker, we just match a price directly.

For today, because I don't have enough time, we're only going to focus on the 1 in the middle, because I think it's anyway the most interesting 1.

Okay, so we're going to go into automated pricing systems and describe a bit better how they operate and the experience that we have.

So starting with something very basic, everybody knows Uniswap.

I do like, and I think it's interesting to have a step back to look at it from a visual perspective.

When you look at Uniswap, I mean, everybody knows there is impermanent loss.

But I think when you look at it in a chart, it is more telling.

So the diagram that you see here, it's the visualization of the risk, the return curve of a Uniswap V2 pool.

I mean, what's the message here is actually very simple, which is this is a terrible structure.

If you provide liquidity, you'll probably end up losing money as soon as the price moves more than 30%.

So it's basically a strategy which has a very, very fat hidden tail.

I mean, if any of you provided liquidity to Uniswap v2, you probably lost money.

I mean, it's just the reality of it, correct?

But fortunately for us, then, there was the implementation of concentrated ranges.

We started to develop technology and systems to be able to manage these ranges.

As we did that, we realized that we can provide liquidity on tight ranges.

We can use multi-call atomic contracts to move that liquidity around.

So we started to play with it and you know like just deploy liquidity and follow the price.

And I'm not gonna talk exactly on how it works but imagine there are many different algorithms that you can develop in Uniswap because again it's just a function similar to an order book but what is interesting for you to understand is just that if you would have a market where you want to be 100% all the time, correct?

Like if you have a restriction that you really want to provide liquidity all the time, it's possible to actually subtract and learn from how do you make that payoff work for you, correct?

And like what we've done is a little bit of a hypothetical case where we say, okay, if we provide liquidity on a range and we have an assumption of implied volatility in that range, how often we would get out of the range and assume that we rebalance, or we take the impermanent loss of that range, how much loss do we accrue, and how many times we need to rebalance?

And what that generates is a little bit of a function.

You have it in front of you, and that function would actually curve and find an optimal point.

That optimal point would be a point where the losses from the impermanent loss get to a minimum phase, correct?

So like, they would start to get close to 0, but your range is as tight as possible.

And this is the mindset that is used when you provide liquidity in something like Uniswap v3, or at least this is how we do it.

After we did that, then we realized this is very interesting.

And then composability started to become a thing.

So protocols started to build very interesting derivative products on top of Uniswap.

And I don't know if some of you have looked at perp.com.

So perp.com is an implementation of perpetuals on top of Uniswap v3, which meant that all of a sudden, we had access to leverage, cross-margin, and high capital efficiency to test Uniswap v3 strategy.

And it was actually very interesting because it proved that you can use concentrated ranges not only for a spot, but you can also build complex derivative products which is very interesting, correct?

After we did that and we kept on experimenting and looking at it and there are new up-and-coming innovations into the space.

We're not a hundred percent sure of course where is this gonna go but a very interesting use case of these has been implementations like Carbon, where the implementation of the market making function gets split in the buy side and the sell side.

So what that means as a market maker is that we'll be able to basically create artificial spreads within concentrated range models.

It starts to look very similar to an order book at that point and this is a little bit where it's going.

After we did this research we started to look at Oracle based market making systems and those ones, these ones are very very interesting actually.

So for those of you that have looked at something like GMX, correct, it works in a very simplistic way.

You have a basket of assets, that basket of assets accrues 3 different types of revenue.

So you would have revenue 1 being the funding rate of the traders, revenue 2 being trading fees and potential revenue tree being the losses or the gains of the traders, correct?

So the traders, if they lose money, the LP makes money, but the other way around.

This risk is not a permanent loss, but there is a risk.

And We started to deploy capital into the systems to understand how they work and figure it out if we could actually use on-chain data to detect trader behavior and offset that risk whenever we wanted to.

So we did that and that took us to understand how these whole dynamics operate all the way until we started to look for companies that had better implementations of this Oracle-based technology.

We end up investing in a company called ARC, which does something very similar.

But This is evolving to implement mechanics where the protocol itself can use incentives to offset risk for the LPs.

When you look at Oracle-based systems, they unlock a wide variety of opportunities.

And the reason that I'm going to bring this example here, which is the implementation of interest rate swaps.

Interest rate swaps are a very complex market.

Probably many of you would not be aware of how they operate.

With a very simple setup, using oracles and basically an approach of trading against the average of traders like in GMX, we can now create swap markets.

The only difference between a perp market that uses an oracle for market making and an interest rate swap is the actual cash flow obligations.

So in the chart here, just for you to understand a little bit better, you have a range.

You can provide liquidity on that range, similar to in Uniswap.

It's just that you're providing range on a rate instead of on an asset.

So on Uniswap v3 you would say, okay, my range is Ethereum from 1000 to 1500.

Here you say the staking rate of of Ethereum is gonna be between 6 and 8 and you can provide liquidity to that particular range.

After you do that, cash obligations start accruing between what is the fixed rate and the variable rate.

These protocols exist and we've actually stress tested them.

It's super interesting the kind of things that you can already start building on DeFi.

And this market is still very young but this is exactly the kind of things that we are researching.

So if also you're building anything like that, please get in touch.

So we experimented a lot, tested a lot of different ways to provide liquidity and what we wanted to do today was just to bring expertise on 3 elements that anybody that is playing with DeFi have to consider, correct?

So we're gonna start with the first 1, pretty simple.

So you need to have a good setup for your data.

But there are 2 basically main ways that you can do this.

You can look at API providers or you can look at an RPC node.

First and foremost, it's important to consider rule number 1, which is on-chain data is free.

So if you know what you're doing, you can actually get a lot of data at a very low cost.

Now, if you go to a provider through an API, my suggestion would be that you mainly use that for data science, data analysis, historic tracking, correct?

But real-time decision making, only with an RPC node, correct?

Like, That's the only thing that actually makes sense because of the latency.

The reality of it is that when you get access to an RPC node, there is also 2 things that you can do.

You can use a provider, or you can host 1 yourself.

So we went ahead and we tested that as well, correct?

What is very interesting is that you start realizing when you get access to an RPC note, you get access to real-time data.

When you host your own note, you get access to real-time data plus main pool data, which means that you can have a little bit of a view on the future.

You're not 100% sure, but you can use probability to understand what would come and take even better decisions from your data.

Compared to centralized finance, execution here is very different.

And we need to make sure execution is good quality.

Plus or minus, there are 2 ways of managing execution whenever you run any type of algorithm into a decentralized platform.

We're going to have a server where the server communicates with an RPC provider, or you host your own node, and you inject the code of your algorithm inside of the node itself.

This has tremendous advantages, especially because you're going to be a lot faster.

I mean, the reality of it is that when you look at the rate limits, the external provider, you have rate limits, and the latency is going to be something in between 10 milliseconds to 600 milliseconds.

When you look at hosting your own nodes, we start actually going to the microsecond realm and you become very fast.

In DeFi, being very, very fast is not as important in terms of trip latency as it is in program execution.

Taking a decision or the calculation of the decision is more important than actually the trip to make that decision.

Because you need to be able to take a decision before 1 block finishes.

If you're in a chain where 1 block takes 10 seconds, That's actually fine.

You have 10 seconds to take your decision.

But if you're in a chain that is very fast, you need to be able to make a lot of simulations and calculations a lot faster.

So if you take 10 milliseconds and you multiply it by a computation that is 10,000, you actually don't manage to compute fast enough, while with your own node you will.

Last but not least, unfortunately, majority of protocols are still having issues.

And the protocols that I'm bringing today as an example, they are great protocols, but they have taught us a lot of things.

Outside of the smart contract vulnerability, there are a lot of interesting things to look at here, and especially the fact that all of these protocols have on-chain data you can track to prevent mistakes.

You can look at tracking collateral that is on the protocols as a way to get alerts when something weird is happening with the collateral.

You can look at how do you track oracles and make sure that if oracles have vulnerability, you yourself take advantage of that inefficiency as a way to protect your own investment.

On the last 1, it's actually a very interesting story.

There is this protocol called Vaults, an interest rate swap platform.

And we discovered a vulnerability in this protocol because we tried to withdraw capital that we knew that we couldn't withdraw.

And actually, the transaction went through.

At that point, we knew the protocol was going to default.

We contacted the founder of the protocol and explained to him, we think your protocol is going to default.

Unfortunately, the answer that we got back was, don't worry, it's a UI mistake.

2 hours later, the protocol closed, unfortunately.

Well, this is basically what I wanted to tell you today in terms of learnings for DeFi.

I ran out of time, so I had a couple of information in terms of trends.

But if There is something else that I just want to say is, we see massive adoption in terms of the growth of options.

It's 1 of the latest trends that we see, both in order book-based systems as well as in function-based approach.

So we believe that, overall, both types of markets are going to continue growing.

There is no, in our point of view, clear definition where order books would overcome any type of AMM systems, but I think they both provide different type of value and they both become more complex as time goes by.

We'll be testing all of these protocols and if you're building anything in this realm, please get in touch Happy to experiment with all the innovation in the industry.


Speaker 0: Thank you for having me. So I hope I can add some value today. So I'm 1 of the co-founders of Keyrock. I'm not gonna talk too much about Keyrock today because I felt that the point of this presentation should be to add value to the audience. But in case you don't know too much about Keyrock, so we are a large algorithmic trading firm, we're a market maker, we have a multitude of services, and if you wanna learn more about them, please get in touch.

We have a booth. But for today, what I want to talk about is something else. I want to be able to give a little bit of value about what has been our journey as a DeFi power user. So around 2 years ago, we started on a mission to understand where this technology is taking us and what we see is that a lot of people in the industry are builders. They are building protocols and majority of presentations and information out there is from the point of view of builders.

We are power users so we are the other side We're using this tech. I want to share with you today some of the learnings that we had from that experience. Okay. So I'm gonna start with something very basic. We are a market maker.

So of course when we look at decentralized finance, the first thing that we tried to figure it out was how do we provide liquidity to this thing? How does it work? As a lot of you know, there is the concept of AMMs, correct? And just before I go into the presentation, I think it's important to take a step back and really understand what the fuck is an AMM. So I don't know if you ever had a situation where you were in university or maybe in high school And you were taught for the first time calculus and how an integral would work you were probably taught that the area under the curve of a derivative of of an integral will be equivalent to an infinite sum of pieces.

And that's basically the exact same difference between an order book and a function-based market maker. They are actually, in a sense, the same. It's just that 1 is a discrete version of the other. 1 is a function, the other 1 is discrete. But you can actually generate the same with both.

So we look at that and then we conclude, actually the reality of AMMs is that they are not an improvement on market making. They are an improvement on access to market making intelligence, which is slightly different, because they are not actually better algorithms, but they do provide a huge value in terms of access. With that, we started to experiment and, you know, like enter into the industry, understand what type of liquidity systems are there. What we realize is that today we basically can provide liquidity in 3 different ways in DeFi. You can go to an order book where they do exist, it's just that the settlement is on chain, correct?

You can go to what we call automated pricing systems. So we don't call it AMMs, we call them automated pricing. Inside of automated pricing we'll have 2 types. We'll have function-based systems like Uniswap and Oracle-based systems like GMX. And last but not least we'll have RFQ systems with a market maker, we just match a price directly.

For today, because I don't have enough time, we're only going to focus on the 1 in the middle, because I think it's anyway the most interesting 1. Okay, so we're going to go into automated pricing systems and describe a bit better how they operate and the experience that we have. So starting with something very basic, everybody knows Uniswap. I do like, and I think it's interesting to have a step back to look at it from a visual perspective. When you look at Uniswap, I mean, everybody knows there is impermanent loss.

But I think when you look at it in a chart, it is more telling. So the diagram that you see here, it's the visualization of the risk, the return curve of a Uniswap V2 pool. What do you see there? I mean, what's the message here is actually very simple, which is this is a terrible structure. You have very, very fat tails.

If you provide liquidity, you'll probably end up losing money as soon as the price moves more than 30%. So it's basically a strategy which has a very, very fat hidden tail. And that actually happened. I mean, if any of you provided liquidity to Uniswap v2, you probably lost money. I mean, it's just the reality of it, correct?

But fortunately for us, then, there was the implementation of concentrated ranges. And here we got more serious. We started to develop technology and systems to be able to manage these ranges. As we did that, we realized that we can provide liquidity on tight ranges. We can use multi-call atomic contracts to move that liquidity around.

So we started to play with it and you know like just deploy liquidity and follow the price. And I'm not gonna talk exactly on how it works but imagine there are many different algorithms that you can develop in Uniswap because again it's just a function similar to an order book but what is interesting for you to understand is just that if you would have a market where you want to be 100% all the time, correct? Like if you have a restriction that you really want to provide liquidity all the time, it's possible to actually subtract and learn from how do you make that payoff work for you, correct? And like what we've done is a little bit of a hypothetical case where we say, okay, if we provide liquidity on a range and we have an assumption of implied volatility in that range, how often we would get out of the range and assume that we rebalance, or we take the impermanent loss of that range, how much loss do we accrue, and how many times we need to rebalance? And what that generates is a little bit of a function.

You have it in front of you, and that function would actually curve and find an optimal point. That optimal point would be a point where the losses from the impermanent loss get to a minimum phase, correct? So like, they would start to get close to 0, but your range is as tight as possible. Therefore, you collect the most fees. And this is the mindset that is used when you provide liquidity in something like Uniswap v3, or at least this is how we do it.

After we did that, then we realized this is very interesting. And then composability started to become a thing. So protocols started to build very interesting derivative products on top of Uniswap. And I don't know if some of you have looked at perp.com. So perp.com is an implementation of perpetuals on top of Uniswap v3, which meant that all of a sudden, we had access to leverage, cross-margin, and high capital efficiency to test Uniswap v3 strategy.

So we went ahead and we did that. And it was actually very interesting because it proved that you can use concentrated ranges not only for a spot, but you can also build complex derivative products which is very interesting, correct? After we did that and we kept on experimenting and looking at it and there are new up-and-coming innovations into the space. We're not a hundred percent sure of course where is this gonna go but a very interesting use case of these has been implementations like Carbon, where the implementation of the market making function gets split in the buy side and the sell side. So what that means as a market maker is that we'll be able to basically create artificial spreads within concentrated range models.

It starts to look very similar to an order book at that point and this is a little bit where it's going. After we did this research we started to look at Oracle based market making systems and those ones, these ones are very very interesting actually. So for those of you that have looked at something like GMX, correct, it works in a very simplistic way. You have a basket of assets, that basket of assets accrues 3 different types of revenue. So you would have revenue 1 being the funding rate of the traders, revenue 2 being trading fees and potential revenue tree being the losses or the gains of the traders, correct?

So the traders, if they lose money, the LP makes money, but the other way around. So the LP starts to carry a risk. This risk is not a permanent loss, but there is a risk. And We started to deploy capital into the systems to understand how they work and figure it out if we could actually use on-chain data to detect trader behavior and offset that risk whenever we wanted to. So we did that and that took us to understand how these whole dynamics operate all the way until we started to look for companies that had better implementations of this Oracle-based technology.

We end up investing in a company called ARC, which does something very similar. But This is evolving to implement mechanics where the protocol itself can use incentives to offset risk for the LPs. And this is really where it's going. But it doesn't really stop there. When you look at Oracle-based systems, they unlock a wide variety of opportunities.

And the reason that I'm going to bring this example here, which is the implementation of interest rate swaps. Interest rate swaps are a very complex market. Probably many of you would not be aware of how they operate. But it's very interesting. With a very simple setup, using oracles and basically an approach of trading against the average of traders like in GMX, we can now create swap markets.

The only difference between a perp market that uses an oracle for market making and an interest rate swap is the actual cash flow obligations. So in the chart here, just for you to understand a little bit better, you have a range. You can provide liquidity on that range, similar to in Uniswap. It's just that you're providing range on a rate instead of on an asset. So on Uniswap v3 you would say, okay, my range is Ethereum from 1000 to 1500.

Here you say the staking rate of of Ethereum is gonna be between 6 and 8 and you can provide liquidity to that particular range. After you do that, cash obligations start accruing between what is the fixed rate and the variable rate. These protocols exist and we've actually stress tested them. It's super interesting the kind of things that you can already start building on DeFi. And this market is still very young but this is exactly the kind of things that we are researching.

So if also you're building anything like that, please get in touch. So we experimented a lot, tested a lot of different ways to provide liquidity and what we wanted to do today was just to bring expertise on 3 elements that anybody that is playing with DeFi have to consider, correct? Which is data, latency, and risk. So we're gonna start with the first 1, pretty simple. Everything that is data.

So you need to have a good setup for your data. But there are 2 basically main ways that you can do this. You can look at API providers or you can look at an RPC node. First and foremost, it's important to consider rule number 1, which is on-chain data is free. So if you know what you're doing, you can actually get a lot of data at a very low cost.

But of course, you can also pay for it. You can go out to a provider. Now, if you go to a provider through an API, my suggestion would be that you mainly use that for data science, data analysis, historic tracking, correct? But real-time decision making, only with an RPC node, correct? Like, That's the only thing that actually makes sense because of the latency.

The reality of it is that when you get access to an RPC node, there is also 2 things that you can do. You can use a provider, or you can host 1 yourself. So we went ahead and we tested that as well, correct? Like, how about hosting a run node? How would it work?

What is very interesting is that you start realizing when you get access to an RPC note, you get access to real-time data. When you host your own note, you get access to real-time data plus main pool data, which means that you can have a little bit of a view on the future. You're not 100% sure, but you can use probability to understand what would come and take even better decisions from your data. Yeah. Number 2 is gonna be execution.

Compared to centralized finance, execution here is very different. And we need to make sure execution is good quality. How do you actually manage execution? Plus or minus, there are 2 ways of managing execution whenever you run any type of algorithm into a decentralized platform. We're going to have a server where the server communicates with an RPC provider, or you host your own node, and you inject the code of your algorithm inside of the node itself.

This has tremendous advantages, especially because you're going to be a lot faster. How much faster? I mean, the reality of it is that when you look at the rate limits, the external provider, you have rate limits, and the latency is going to be something in between 10 milliseconds to 600 milliseconds. When you look at hosting your own nodes, we start actually going to the microsecond realm and you become very fast. Now, why is this important?

In DeFi, being very, very fast is not as important in terms of trip latency as it is in program execution. What does that mean? Taking a decision or the calculation of the decision is more important than actually the trip to make that decision. Why? Because you need to be able to take a decision before 1 block finishes.

If you're in a chain where 1 block takes 10 seconds, That's actually fine. You have 10 seconds to take your decision. But if you're in a chain that is very fast, you need to be able to make a lot of simulations and calculations a lot faster. So if you take 10 milliseconds and you multiply it by a computation that is 10,000, you actually don't manage to compute fast enough, while with your own node you will. Last but not least, unfortunately, majority of protocols are still having issues.

And the protocols that I'm bringing today as an example, they are great protocols, but they have taught us a lot of things. Outside of the smart contract vulnerability, there are a lot of interesting things to look at here, and especially the fact that all of these protocols have on-chain data you can track to prevent mistakes. You can look at tracking collateral that is on the protocols as a way to get alerts when something weird is happening with the collateral. You can look at how do you track oracles and make sure that if oracles have vulnerability, you yourself take advantage of that inefficiency as a way to protect your own investment. How do you stress test protocols?

On the last 1, it's actually a very interesting story. There is this protocol called Vaults, an interest rate swap platform. And we discovered a vulnerability in this protocol because we tried to withdraw capital that we knew that we couldn't withdraw. And actually, the transaction went through. At that point, we knew the protocol was going to default.

We contacted the founder of the protocol and explained to him, we think your protocol is going to default. Unfortunately, the answer that we got back was, don't worry, it's a UI mistake. 2 hours later, the protocol closed, unfortunately. Well, this is basically what I wanted to tell you today in terms of learnings for DeFi. I ran out of time, so I had a couple of information in terms of trends.

But if There is something else that I just want to say is, we see massive adoption in terms of the growth of options. And the options market is growing a lot. It's 1 of the latest trends that we see, both in order book-based systems as well as in function-based approach. So we believe that, overall, both types of markets are going to continue growing. There is no, in our point of view, clear definition where order books would overcome any type of AMM systems, but I think they both provide different type of value and they both become more complex as time goes by.

We'll be testing all of these protocols and if you're building anything in this realm, please get in touch Happy to experiment with all the innovation in the industry. Thank you very much

See all TOKEN2049 transcripts on Youtube

Journey of a DeFi Power User - TOKEN2049 Singapore 2023