hildobby (Dragonfly) - Surfacing and using CEX onchain data

[Music] Hello everyone. Um, my name is Hill Dobby. I work at venture capital firm called Dragonfly where I run data. Um, I do a lot of public goods kind of data sets and dashboards. So, you might have seen me pop up on Dune and in other places and there's more of that today.

So, you know, just stay tuned. Um the premise of this talk is that central exchange data is actually very under surfaced on chain. Um and I want to change that. Um it's something that I've been working on for a while, but it's kind of culminating into something that's more ready to be used today. Um and I'm going to share more about it.

First off, why why do I even want to cover central exchange data? What's the point? um you know it is prominent and you know it is an important part of the space but there's other reasons the first one being that like it's actually core to other data sets that I used myself and have built in the past um one of them is wash trading for NFTs so that's way back when NFTs were a thing uh and were blooming at the time and you know looks rare and x2 popped up and actually um accidentally or not incentivized wash trading um and we needed to do something about it and actually you know surface those trades from the rest um and one of the mechanics I used is that I want to look at like if two addresses that are interacting with one another trading one another are funded by the same source and I want to flag those as potentially wash trading but I do want to exclude all central exchange addresses because it is likely that you know two people are funded by Binance simultaneously doesn't make them the same entity. Um, then there's staking where I'm trying to track every entity that is staking Ethereum. Um, most notably lately Coinbase which I've been trying to track.

There's also Binance and other central exchanges that do show up in this data. And there's also ETF flows. You might know ETF flows from off-chain data that there's a lot of websites reporting on this data. I do it very differently. So if you look at this next slide over here, you can see that this is actually data from onchain addresses that I'm tracking looking at reported numbers and trying to find those addresses.

So constantly sleuththing for new flows where those are coming from. I know that it's going to be Coinbase Prime that's going to be the custodian for Black Rockck for example. And therefore, you know, if you if you build up enough uristics, you can actually triangle who like which which set of addresses do make up um the AUM that they are holding for both ETH but also for Bitcoin. Um and another reason is that I actually published uh central exchange addresses um data set a while ago. It was just a query I had on Dune and I had like I realized I had a lot of likes on it.

Usually this kind of likes on a query is like people farming likes or something on Dune. Um so that was surprising but I guess it showed there was interest for it. Uh which I was surprised and I was also interested so I thought I might as well you know get into it. And lastly um this data set has also been used in other places. So on ether scan when you go there and other scans I don't know how you call them um you'll find some addresses are tagged and have this little label basically saying I compiled the data set.

Um funnily enough I found out about it because some people uh were getting rugged and tracing the funds. They saw that it was going to Binance and they got confused thinking I rugged them. So I got DMs about it and anyways that's how I found out that Etherscan uses my labels. Um so how do I actually approach this? Um the first step is to understand how central exchanges work on chain.

Uh so I made a little diagram um to try to break it down. If we focus on one central exchange address, let's call it the stereotypical central exchange address. You'll see some inflows from deposit addresses. Those would be the orange ones as well as other orange ones which can be smart contracts in some cases. Kraken for example uses smart contract deposit addresses but others may not.

And then you may also see I mean you will also see withdrawals to all kinds of addresses uh to the user's choosing I guess. Uh and then you also have interactions between addresses and one another. Um so let's keep this diagram in mind and you know go from there and see how we can actually gather this data set of addresses. Um if we zoom in on deposits themselves um there's a little more that goes on. There's two different types of fungeible tokens.

There's ETH and ERC20s. ERC20s when you transfer them from an address you're going to need some gas funding. Hence the difference. Whereas if you actually deposit ETH, you can directly deposit it. You can already have the exchange, you know, transfer it themselves because there's enough for gas unlike ERC20s.

So if it's an ERC20, you need an address to actually deposit some gas for the transaction or it's a smart contract. Um, yeah. So I want to identify central exchange addresses. Let's look at the types and break them down into eight different types. Um, you're going to have hot wallets.

That's going to be the ones users interact with for deposits, withdrawals, and all kinds of shenanigans. Um, cold wallets are going to be the ones that are used, I guess, for internal purposes. Um, and usually to keep funds on um, you know, secure wallets that aren't uh, being actively used um, and for safekeeping purposes. Then there's gas suppliers, like I said before, for when you deposit ERC20s, you're going to need to fund those deposit addresses. And then there's excess gas recipient.

Um, sometimes, you know, you overestimate I mean, you often overestimate the fee and that way you got a little bit left of ETH in deposit addresses. It it's dust amounts, but you know, many dust amounts pile up into a lot of amounts. So, you got to eventually send that through back to uh your address. Uh, you also have addresses creating contracts. So, that can be deposit addresses, but other purposes as well.

You have those executing the contracts uh usually for consolidating deposits. You have wallets used for rotation purposes. So this is you know funds don't stay long there but just one hub into a series of hubs. Um and then you also have central exchanges having their own deposit addresses either for their own exchange or on others. So Binance will have deposit addresses on Coinbase vice versa and those are also worth tagging.

Um, it's not a perfect world. Of course, there's, you know, a lot of things to watch out for. Uh, sometimes you'll have random connection that seem to indicate that a central exchange address is tied to a node of central exchange addresses, but that's not always the case. So, of course, it is prone to some mistakes. Um, there's also cases where central exchanges are just a single node and are self-sufficient on deposits and can therefore supply withdrawals from this and have no connection to the rest of addresses.

Those are also harder to, you know, phase out from the rest. And finally, there's also deposits that are never consolidated for all kind of purposes. Maybe it's too small of a deposit or I don't know many other reasons. I don't work at a central exchange, but it's just assumptions here. Um, so how do I actually approach it and I do I go on Dune?

I have a bunch of queries where I put together huristics based on the things I showed you before. So we know the patterns to look out for and I compile those lists and if I have two or more um sets of huristics that match and lead to uh a central exchange address being you know Coinbase or whatever else, then it is highly likely and I'm going to tag it. Um I also have these hyperlinks. What you'll see is that those are Etherscan, Nensen, Herd, GitHub, and Alchem. Those are all explorers that I use.

So I can just click and actually go through. Um I'm very often going to use Arkham, amazing platform to track transfers. Uh GitHub is awesome for smart contracts because that's where devs have their documentation and tag their contracts. um herd is awesome awesome to go deeper into how the contract actually work and and so on and ether scan you know ether scan you know nansen whatever um like I said it's a data set that I've been compiling for a while so the list of central exchange addresses I have I started in July 2022 when I made NFT wash trading but I've actually been constantly adding more and more um and even more as of late there's a few contributors but I have contributed the overwhelming majority And the black bars and the little slim things is the other contributors. But I welcome everyone to come and contribute.

Submit a PR on Dune um and tag me so I can review and actually you know make sense of why you tag this address. Um because you can't I I want like it doesn't make sense to just tag anything and let anything go. Obviously we need some checking. Once we have central exchange addresses we can actually extrapolate more data sets from it. Um, so you got the manually tagged addresses and then this is for applying for all EVM chains and for every blockchain I have a data set where it checks when those addresses first receive any transfer and tags them accordingly and eventually consolidates into a data set that has all these addresses at a blockchain level.

So you're going to see, you know, some addresses pop up on multiple chains because they are used across different chains. Um, and then once you've got central exchange addresses, there's two more things you can pull from this. Uh, central exchange flows. So, this is all interactions, all transfers in and out of those wallets between one another and all of the kinds. And there's also deposit addresses um, which are very prominent on uh, our blockchains, but um, like I said, undersurfaced.

Oh, and like one little duty tip. If you ever use Dune, watch out. You can use the crop chain table, but if you want more efficient querying, please use the blockchain specific ones. Uh, if that's all you need, just a little FYI, I guess. Um, this is how I tag deposit addresses.

You can actually pause and take a look here. On a high level, I'm looking at the the first time central exchanges receive any token from any address. So the first funding from any address I extrapolate that one day forward look at all the same tokens being sent because people will do test transactions and all kinds of things. You need to account for those and then once I have the maximum like the last time within that time frame that it was deposited from there I look back into when people actually deposited. It's a bit complex but if you want to take a look um I'll upload the slides and you can also rewatch the video but yeah um and you can also inspect the code of course but I'm uh fairly confident in my huristics although I'm very open to anyone coming to change that and you know submitting uh changes to it.

I actually also have started to do some sanity checks. I'm not going to go over all them too much, but essentially checking that the amount consolidated closely resembles the amount deposited because otherwise if you see huge spikes over there, it would be really weird. And then in terms of transfers, central exchanges are going to be more efficient in the way they transfer things than um users are because like I said, test transactions. There's also some contracts sending fees regularly to deposit addresses and then the central exchange only consolidates one once in a while. So anyways, to my to what I I was looking for, this makes sense with what I was expecting and it's one of the other sanity checks that I won't go over here because there's many things you could look at and I'm always, like I said, open to feedback on the data set and constantly iterating on it until um everyone's happy with it.

the actual data uh is probably what you're here for. Um so there's 4,300 uh 4,300 addresses and 92 million deposit addresses and I've tagged uh 350 central exchanges. Of course, it's an unknown unknown how many there are that are missing. Um but like I said, I welcome all you to come forward and let me know. In terms of data, the data we get is actually pretty cool.

So here is the deposit volume across all exchanges across um I would say five of the major chains being Ethereum, BNB, B, Arbitum, Optimism uh and a few others. And here for example you can see you know the FTX fallout in gray uh where Coinbase gobbles up a bunch of their users I guess and takes on market share. You can see the Polex story. You can see HTX and all these are at play and you can also very prominently see Binance uh swallowing the whole market. U so yeah interesting data like I said there's this is just a high level but you can dive into it.

All the data sets are available for you to look at. Um in terms of where the deposits actually happen uh the cool thing is Ethereum is still uh doing great in that category. uh it's get it's lost a little bit of market share to other other chains um a little bit to BNB and then some layer 2s avalanche um but that is more on the transfer count side but if you actually look at the size and the actual volume it's mostly settled uh on Ethereum itself in terms of how much um how much does do central exchanges use in terms of blockchain capacity uh I thought that was another interesting thing to look at. Um, and you can see that in terms of transfers, it used to be 30% for ETH, but it's actually been going down. The pro the thing is it's not really relevant to look at transfers because as time goes, transactions include more and more transfers and things get more complex, especially as DeFi pops up and other things.

So if you look at transactions themselves, you actually see that we are more stable and we're at around 25% of all transactions on Ethereum involve central exchange addresses. So I think this goes back to the point of why are we even looking at this? Uh because it's prominent and we should look at things that are happening on the blockchain. um it's prominent across other chains as well, but um due to the tooling and everything available, Ethereum is going to be where I'm more confident um in terms of tagging and addresses I can sleuth for. Um and it's also like I said, uh where most things are happening terms of fees, uh it's also very prominent.

Now, it is important to note that this is only the fees that central exchanges themselves put on. um in terms of deposits. So here you can see there's 10% 10% of all fees paid on Ethereum are paid by central exchanges. But if you actually look at this one, this is just deposit from users. So you can add five more percent that are users depositing into those deposit addresses.

And there's also some non non- insignificant amount for other blockchains uh that you know obviously there's like I said this oh no no no maybe it's higher but um maybe we go up from here as we tag more. Another interesting use case for this data set is uh as a as a data scientist one of the big problems is there's a lot of spam on blockchains and it's actually hard to discern spam from the rest. Um therefore by looking at deposits that are being consolidated by by central exchanges I do feel like it's some kind of confirmation on the um the tokens uh importance relative to the rest and therefore this is a simple list that just looks at inflow volume all time and ranks the tokens. So you can have some threshold and actually only look at certain tokens and remove all the you know spam tokens and other things that you don't want to look at. Um so yeah this is just a proof of concept idea but I think this can be taken further and hopefully get some you know huristics based uh really good ranking for tokens uh that we crucially are missing and manually tagging would be an insane amount of work given you know the insane amount of tokens that constantly pop up on chain.

How are addresses funding on blockchain? Well, on Ethereum, there's over 85% that are funded by addresses that I've tagged as centralized exchanges. Um, my assumption is that the rest is of course funded by one another uh as people, you know, move their funds and all of the kind of things. Um, but there's also bridge transactions that um is another piece of the puzzle that's missing and um yeah, so it's an unknown unknown, I guess. But what this gets us is that now that we know um where like from where from where addresses that are funded, we can actually look and each of you building your protocol can look at your certain KPI put that into a junquery and then look at where those users are coming from um to better understand your audience.

Um, you can also see, you know, if it's funded by Binance US, you can actually see, well, it's probably a US. Uh, and there's there's more you can get out of this. But my my goal here is to showcase that you can make your own query and look at better profiling your users and hopefully better targeting them and understanding their behavior. Um, and I think that's a part of a whole we're used to, you know, analyzing DEX transactions, but I think this should be just as important um to everyone building protocols over here. Um, another interesting thing, it's a little adjacent to central exchanges themselves, I guess, or at least the centralized aspect of centralized ex exchanges is that I don't know if you've noticed, but lately Binance Smart Chain has taken all the um DEX volume.

Um, I'm not sure if that's something you have, you know, maybe even seen somewhere. Um, but I thought it was important and I was wondering why. And the whole reason is Binance recently launched um well called Dex on sex. I don't know how what's what would be the technical term but essentially um users on the Binance app can now go and trade decentralized exchange tokens and Binance will do the bundling for them and route the transactions. So there's been a lot of volume coming on BNB as a result of this.

Um, you can also expect that other centralized exchange will step into that space eventually. Um, I think it only makes sense, especially if you consider that tailin tokens are much harder to list for. So, you might as well use the onchain liquidity and pools that are readily available. Um, of course, it has a side effect. The more there's, you know, tail and tokens being bundled, it also leads to a lot of arbitrage transactions.

Um, and that's another dashboard I've put together a while ago on me and arbitrage. But essentially what I'm getting at is having more of these dex on sex transactions has also binary like secondary effects that also cause more transactions such as arbitrage. Hence why um there's such a significant jump in their volume uh which should not be taken lightly and yeah interesting data. There's more things for centralized exchange coming. Uh yesterday uh Robin Hood announced uh essentially their uh stock tokens uh on arbitum.

Um I think that's another part of the whole centralized exchange kind of category that we should look at. Um I'm very curious and I'm going to start you know looking at who's trading open AI on arbitrum I guess. Um yeah interesting things to look out for. All these data sets you can query for free on June. Uh this below is a link to a dashboard that I have where I've tagged a bunch of data sets that I think are relevant and are cool and you should uh consider trustworthy if you trust me.

Um so yeah, this is the list for centralized exchanges, but of course there's more there uh for you to look at. Um there's also a dashboard I've put together on centralized exchange flows. Uh it's mostly charts I haven't shown here actually. Um, so yeah, go take a look. Um, hillbexes.

Um, yeah, a bunch of data and I'll keep adding to it. I'm open to feedback. Please do share some feedback. Please do DM me if you have anything that you think is interesting. Um, and also like I said, the Dex onex category or whatever you want to call it, I think is here to grow.

Um so I started also putting together a dashboard on this for Binance but later on who knows most probably other exchanges. Um yeah it's here is the the share um relative to the total DEX uh total DEX volume. Uh so it is a prominent amount relative to the total and I can see it growing from here most probably. But of course there's more to find. It's an imperfect data set.

Uh, no data set is perfect, but this one is not either. Um, so if you have any feedback, I've left a QR code for you to scan and it's a forum where you can give me any info. So that could be there's a centralized exchange and you feel under represented, please come forward. I'm happy to tag you. There's you're a blockchain and you feel like you should be more prominently shown relative to the other blockchains.

Same again. um or even just a hunch of like well this doesn't feel right this doesn't feel like Binance's numbers they should be much higher anything helps so please come forward please send me uh all your feedback um so far it includes all these blockchains um but of course there's space for more uh I think you know the EVM space is only growing uh and my goal is to focus on EVMS but of course there's other ecosystems to look at. There's Bitcoin, there's Solana most prominently and then there's bunch of others. If you are from one of these ecosystems and are um really knowledgeable and how those work, how central exchanges work on those, please come and help. I'm really happy to have you as a contributor and and guide you into contributing to Dune Spellbook um to make this a really good place.

Um I'm not going to be able to track everything forever. Eventually, my goal is to hand to have a really really solid seed data set that can hand off to some AI that scrolls the blockchain, finds uristics, and essentially is self-sufficient into tagging. That's good. Like that's looking forward, but at the pace of which AI is growing, who knows? It could be, you know, we could be weeks away.

I have no idea. Um, huge shout out to June. This is obviously enabled by their platform by spellbook itself allowing me to create data sets uh which anyone can and yeah so huge shout out to them it's really an awesome platform I guess I'm biased but yeah I don't know uh RAM is also an awesome platform and I've used a lot for not only cross-checking tags because they have a lot readily tagged but also because transfer wise it's really really interesting in the it's it's perfect for actually tracking central exchange because most of what you want to see is token transfers and you don't want to see all the rest all the fluff that you're going to see on Etherscan um which are good for other purposes but in this case ARM was the better choice. Uh lastly also uh I actually got my start at a centralized exchange data company uh a while ago. Uh so that was in 2019.

Uh you can see I was put into a cardboard form in my very professional attire uh at their office. So if you see this there, uh that's me. Uh please come say hi and um thanks for to all of Kai code to actually um putting me on this path and and you know um it's how I got my uh how I got started in crypto and how I got even interested in the first place. Um next up, of course, I'm never I'm always looking for new things to analyze. Uh there's still a lot more to do on centralized exchanges, but there's another big piece of the pie that's missing.

Um and I'm looking for any help. Uh I'm looking at bridges interrupt. It's a hot topic at the moment. Um my goal is to compile all deposits and all withdrawals into a single uh um data set. So you can not only look at volume and everything, but also latency, you can look at which are intense protocol, which are not and all these kind of nuances.

Um so yeah there's even CEOs fighting on Twitter uh for who which has the better interop protocol. So maybe we can help them uh by providing data to you know uh back up their arguments. Um thank you very much. Um I hope you enjoyed my talk. Um if you have any questions please come forward.

Uh I'd love to hear Um I actually have one question regarding the Binance uh decentralized centralized exchanges thing uh rooting is it happening specifically on Pancake Swap uh because on our end so I'm from Kao uh we have seen that the volume on Pancake Swap significantly increased recently and so this is maybe related Uh I haven't looked into which of the dexes it is uh precisely. My guess would be pancake swap because I think for tailin tokens on BNB that's where most things happen. And the goal here for Binance is not to route for you know the larger tokens. Those are already listing and going to have much deeper liquidity on the Binance app itself. Um so my hunch would be that yes it would be Pancake Swap but I haven't looked into it uh too deep.

Okay. Thank you. Any other questions? Okay. No.

Well, thank you for listening. Um, I hope you enjoyed and please, um, come forward if you have any more questions. Feel free to reach out on Twitter, Telegram, or anywhere else. Feel free to use my data sets and provide feedback on them. Thank you very much.

[Music]