The Intersection of DeFi and NFTs - Nick Emmons

uh my name is nick emmons i'm the co-founder and ceo of upshot and we're building uh tools for pricing nfts more efficiently in order to help build out this intersection of d5 and nfts over the past year and a half two years define nfts have exploded and they've become these sort of two core pillars of the the crypto application space d5 gives us this permissionless composable set of tools in which we can build highly expressive financial instruments in a decentralized setting but the issue is we're limited to a fairly finite set of assets in in building those instruments and nfts on the other hand are a structural innovation they quite literally allow us to represent anything on chain in the same setting that these composable tool sets live the issue here is that they're highly illiquid they're difficult to price and so having them interact with this this financial system that we're building is quite difficult and possible in some cases so these these two pillars have exploded in silos in separation what we're trying to do is is to bridge these things and unlock the massive amounts of value that can exist at the intersection of of d5 and nfts and a lot of why they are siloed is because of this nft liquidity problem nft's fundamentally are low velocity non-fungible assets they don't change hands very often and this means that these open market transactions these transfers of ownership that that act as price discovery events for more fungible or liquid assets just don't happen frequently enough for nfts for that to be used as the sole means of pricing them so when you look to the real world the real world to see similar types of assets you see real estate you see physical art being supplemented heavily by the use of appraisals the querying of experts opinions around the values of these assets and so that's what we're trying to to build here to help bridge the gap between define nfts and appraisals are are useful because they uh they don't come with the prerequisite of an asset changing hands when we're buying nfts when we're buying real estate we're buying physical art we're holding these assets for long periods of time but they need to be able to be repriced regularly and accurately in order for them to kind of fit into the d5 mold that that we'd like them to fit into and build these these new exotic financial instruments so how do nft appraisals work this is going to be a very high level overview of nfc appraisals there's a lot more information on our blog and there's a lot more uh kind of ways you're going to appraise nfts we'll just give a high-level overview first before we get into that there are kind of two ways to appraise nfts on one hand there are there are human appraisals there's some experiments being done here we did a lot of early experiments with this over the past year as well and human appraisals are essentially asking experts to manually weigh in on the value of these assets go through a ui or some product and appraise each individual each nft specifically and this runs into some headwinds in terms of scalability and accuracy on the scalability front manually appraising one nft at a time imposes a fairly low ceiling on the number of nfts that can be appraised in any reasonable number of times so repricing these nfts frequently becomes very difficult often impossible at scale and accuracy becomes an issue because when building questions when building systems for human appraisers to elicit their beliefs around the values of these assets you have to take into account account a fairly limited uh cognitive bandwidth you have to ask highly approximative questions to these experts is this nft more valuable than this nft or given this nft in these price ranges which price range does it fall within so you're asking multiple rounds of questions just to reach a reasonably granular estimate of the value of that nft further eating into the scalability further eating at the the frequency in which you can uh reprice these assets and quickly kind of deteriorating the benefits of appraisals in the first place so that's human appraisals on the other end there are kind of model driven approaches machine learning models other kind of statistical models to pricing appraisals and this is kind of the world we're living in right now and the benefit of this is that machine learning models can price many nfts with high levels of accuracy and reprice these things very very frequently and so we'll go into like what goes into these models and some of the things that are considered when when pricing nft is sort of artificially this way so primarily that the things that go into these models when pricing nfcs are appraising nfts is first the metadata the traits the rarity things like this what the nft is the sort of hard market interactions sales mints these these kind of transfers of ownerships that occur in the open market and i think softer market interactions things like bids ask what the order books for these nfts look like and then the secondary considerations that go into these models that we're not really going to touch on in this presentation are things like social sentiment analysis how are people talking about these nfts what is what's the general sentiment across social networks for these things uh twitter discord engagement how big are these communities how fast are they growing how engaged are they things like this and sprinkling in a little bit of of sort of crowd intelligence this this manual expert touch done in small doses can actually improve considerably the the results of of these sort of model driven pricing approaches so what we'll do now is we'll go through a quick example of of a flow of what goes into pricing these things again this is this is very high level this is not everything that goes into it but it should give a a decent idea in how we're able to price nfts without needing them to transact on the open market and do it so efficiently that we're able to reprice them at a kind of high speeds so let's take this dick butt 4835 first what we're going to do is we're going to take the traits from the contract we're going to take the the metadata that has been defined uh directly for this dick but this thick butt has has three traits three non uh three kind of populated traits its background is a sunset its eyes are googly and its skin is light now if we just went off these three traits we'd be only capturing part of the picture of what actually goes into influencing this price primarily what we what we look for in in dick buts and and other types of nfts is how clean it is versus how kind of messy or sloppy it is and so what you have to do is you have to start identifying these meta traits that you can define on top of the metadata that is uh existent in the contract itself and so the meta metatrade that is kind of most useful for dick butts and most other nfts is the number of traits because this will this will allow us to nail down how clean something is or things like this other metatrades may be specific combinations of traits hoodies and sunglasses might be a metatrade in itself because that combination will influence the price positively more than just a hoodie or just sunglasses things like that and so now that we've defined our sort of full set of traits we can start to do in-depth analysis around trait-specific interactions on the open market we can examine trait specific sales so not just tied to that dick but but uh tied to all googly eyed dick butts they're all googly eyed three trait thick butts are all combinations of those traits we'll do the same with order books bids asks assess order book composition at a kind of a trait level we can do things like look at the how the rarest trait in in different nfts or different dick bots is performing and and use that to influence our prices use rarity score last sale price across a collection to gain kind of collection-wide sentiment there's a number of there's a bunch of other features that can go into it this is just a small set of what goes into it but it's to give an idea of the sort of information that's going into to these models and then what we can do is we can grade these models fairly objectively what we do is we can assess the sales history of the nfts against our sort of our timeline of of prices that we've uh produced for these assets these appraised prices and then we can produce a sort of median relative error for for the the comparison of actual sales to predicted prices and what this median relative error is given this price we're able to predict the next sale price with an x percent level of accuracy so if uh that dick was was predicted to be worth two weeks with a 12.75 percent mre the real price is actually two-week plus or minus 12.7 pi 12.75 percent and now we now we have are these absolute prices that are being produced at high frequencies with known error bounds that we can then begin to build different financial infrastructure on top of and that's what we get at the end of this we get real-time price feeds for nfts and i think the real magic in this is seen when you look at an asset that has that has never changed hands or hasn't changed hands in a long period of time like this doodle hasn't changed hands in about four months it was last sold at 1.19 eth but because of how we're we're modeling the uh the nfts in this collection how we're pricing nfts using these approaches we're actually able to have an appraisal that is updated hourly so we have a price fee that that is updated on an hourly cadence and can be used across a number of financial applications um and these are some quick stats around how the how we're appraising nfts right now we're supporting over 240 collections we're uh appraising over 270 000 uh nfts across those collections and we're repricing those nfts uh every hour so we're producing these close to real-time price fees for nfts with an average mre of around eight to fifteen percent which is industry leading right now so uh and then i'm just going to go into quickly kind of what this enables these are just a few examples there's a number of kind of exotic defy x nft primitives that real time price feeds and nft appraisals uh enable but we'll just go into three really quickly so the first one is efficient nft lending protocols the sort of de facto way to build out nft lending protocols right now there's been a number of experiments in the spaces via this pawn shop model this this involves back and forth between two parties it runs into similar price discovery and efficiencies and a number of other kind of sub-optimalities so what you can do when you have these real-time price feeds powered by appraisals you can build a system that looks similar to maker dow i can now deposit an nft that i have into essentially a vault or a cdp and point to a price feed for that nft itself it'll update every hour and that will track whether this position needs to be liquidated whether it's over collateralized or not things like this and then i can draw debt against that position similar to how i would draw that against a cdp or a vault and now i have i have taken out leverage to buy more dick buts maybe i've taken out debt to to purchase a home or something in the real world i've just created value from my previously stagnant asset nfts in general right now are very stagnant reservoirs of capital and they're very large reservoirs of stagnant capital we can't actually do many productive things with that and so we're losing a ton of money because that money is just being left on the table and not being able to earn yield on it not being able to to kind of take out debt from it and building these efficient lending protocols and other kind of defects nft primitives will enable us to make that capital make that pool of capital productive and then when i'm ready to pay off my debt i can just re capitalize the the position pull up my nft assuming it hasn't been under collateralized i've now made my my dick but uh productive from in a capital sense another example of what you can build with these sorts of things are semi-autonomous and nft investment dows so right now nft investment dows are run fairly through verily uh discretionary governance practices it's it's based on human interaction decentralized governments kind of social choice theory but what we can do instead is we can say here are the parameters in which our investment data wants to operate in we want to have this allocation of of this type of asset or this collection and then we can just let the models handle the rest the models can assess orders as they're going onto the open market onto these open order books and purchase or sell nfts based on whether it's a good deal to do so as determined by these models i think this is going to open up a whole design space of how we can think about dows and we can think about investment dows in this actually autonomous sense and we'll really expand the the kind of level of exposure that many people outside of the nft space can can soon get to the nft market and then the last one which is most exciting in my eyes is nft synthetics synthetic assets are just assets that give you exposure to some other asset or some some other price feed and all that's needed to create a synthetic asset is some arbitrary pool of capital and some reliable price feed so now i can put together essentially a pinterest board of nfts i could put together a pinterest port of dick buts for example point to the price feeds for those specific nfts back those price feeds with capital and now i can pull out a a token or an asset that is tracking the market movements or the price movements of those things as determined by these appraisals without needing to hold any of them myself without them needing to be locked in any specific place i can now take theoretically arbitrarily deep positions in these things and use these synthetic representations of these assets to build a number of d5 primitives themselves they become fungible representations of previously non-fungible assets and so these are just some of the examples that of the kind of d5x nft primitives that you can build when you have uh nft price feeds that are not uh reliant on these transfers of ownership for nfts nfc's nfc appraisals will sort of open up the the intersection of d5 and nfts and all this is is exposed through a pretty easy to integrate rest api and and uh soon some adapters to bring them into the the d5 space so if you'd like to integrate any of this if you'd like to build any uh interesting experiments at this intersection check out our website our developer docs are pretty easy to consume and just reach out if you have any questions so thanks