The financialization of Non-Fungible Tokens
The financialization of Non-Fungible Tokens
Market research on the NFT lending space and its pricing infrastructure solutions.
In this piece, I showcase the current reality of the NFT space and deep-dive into the keys to the next growth phase. Please remember that I’m just a random anon on the internet. I’m not a financial advisor, and none of what I write should be taken as an endorsement. I don’t hold any of the discussed projects, but I plan on investing in some of them. These are just my thoughts.
NFT was chosen by the Collins Dictionary as word of the year for 2021. This achievement is due to the sharp growth and the impressive adoption numbers that the NFT industry saw last year.
Some companies such as Visa estimated the NFT market valuation to be around $100B
. Other institutional sources such as Jeffries investment banking group are more conservative in their valuations, but still announced that they expect the NFT market to reach an $80 billion valuation by 2025. What is clear is that NFTs managed to get everyone’s attention.Last years’ statistics were astonishing. The number of sales increased a 1,836% when compared to 2020
. The traded volume (in USD terms) growth was even more impressive, achieving a 21,350% increase with more than $17B traded ($14.6B of which took place on the Ethereum Mainnet).
Collections such as the CryptoPunks or Bored Ape Yacht Club saw extraordinary repricings, becoming two of the most iconic and expensive collections in the whole space. The floor prices and the avg sale price increased substantially for both collections, bringing each of their market valuations above $2.5B.
This phenomenon is only explainable because NFTs have managed to capture retail interest successfully. Unlike Decentralized Finance, where the next adoption phase is highly dependent on institutional investors, we have already seen the first signs of mass adoption in the NFT space. The combination of art, culture, and speculation is more attractive to the average person than financial innovation.
The “problem” of sudden growth
As mentioned earlier, BAYC has become the most valued project in the whole industry. In less than a year these profile pictures managed to climb to the top of the NFT leaderboard.
The floor price of the listed apes, which minted at 0.08Ξ less than a year ago, sits at 109Ξ at the time of writing. That’s a 136,150% increase in value (in ETH terms).
What happens when new investors join the space and end up holding one of these unicorn NFTs?
If we query on-chain data to analyze the wallets of the BAYC owners, we can see that this project is quite distributed and its holders are long-term oriented.
The 10,000 apes are held among 6,441 different wallets. Over 60% of them haven’t been transferred for more than 3 months, and almost 40% haven’t been transferred for more than 6 months.
A common assumption on Twitter is that most BAYC holders are not wealthy but at the same time they are really committed to the project and don’t want to realize profits by selling their NFTs. This hypothesis can be easily validated by fetching the wallets of the holders and their balances. To simplify the investigation, this validation was performed by only looking at the ETH balances of these wallets, since ETH is the most commonly held asset in the industry. Although not certain, it seems fair to assume that the ETH holdings of the wallet on Ethereum Mainnet should be representative of the overall wealth of that wallet.
According to this data, 75% of the holders have less than 1/10th of the collection’s floor value in their wallets. Although BAYC owners may have other assets and/or other wallets, it seems fair to assume that, due to the illiquid nature of their NFTs, these investors will suffer from opportunity costs. On top of that, some previous holders may have probably sold because their NFTs acquired substantial and the only way to get liquidity was by selling them in the secondary market.
Driving liquidity to illiquid JPEGs
Due to the current valuation of the NFT space and the growth rate that we saw last year, it can be considered a dormant giant. The space is rapidly accumulating value, but its utility is highly constrained by its illiquid nature. This restrictive dynamic is preventing the explosion that will come once NFTs manage to unleash their full potential.
Take a step back for a moment. Look at the past history of DeFi. There was a time when it wasn't possible to borrow against ETH. Eventually, AAVE and Compound changed that. As the first lending protocols, they became money legos, enabling a new layer of financial use-cases that could be built on top of them. As a result, their valuations and TVL skyrocketed, becoming a catalyst for growth in the whole DeFi space.
Since most NFTs don’t have any utility and are mainly cultural and/or speculative assets, using them as collateral to get access to temporary liquidity seems like a great fit. Several actors have noticed this opportunity and are working on the first NFT lending protocols. If some of them manage to create a successful product, we will experience a similar outcome as we saw with AAVE and Compound.
Why would anyone take a loan against their NFTs?
As with ERC-20 tokens, there are two main reasons.
To get leverage and perform any sort of operations such as investing in underpriced assets, buying more NFTs, maybe shorting some overvalued NFTs, or even buying a house.
To convert their idle NFTs into a source of passive income by using the borrowed amount to earn yield (for this to happen the obtained yield would have to be higher than the borrowing fee).
Both options are interesting, but the second one seems to be especially appealing to combine with the illiquid and passive nature of NFTs. Other than the flippers, most of the individuals who hold NFTs aim to do it for the mid to long-term.
Because of that, regardless of the risk profile of the NFT holders, it should be extremely appealing to them to make their NFTs productive (note that the yield strategy and the LTV can be adjusted to match their risk appetite). So it can be concluded that, before accounting for technical constraints, the total addressable market, in the best-case scenario, for NFT lending is quite big.
The current landscape
As mentioned before, those who noticed the need for NFT lending have already started a race to develop the protocols that will unleash the next phase in the NFT space.
So far, there seem to be two different approaches toward lending against someone's NFTs:
Protocols who believe there will be demand to borrow someone else’s NFTs and want to create a P2P marketplace to enable these activities. The business model of these protocols relies on borrowers paying fees to liquidity providers to get temporary access to their tokens and keep a cut of the fees.
Protocols who are willing to provide liquidity themselves to the NFT holders. On the one hand, the NFT owner can repay their loan, plus a fee, to get their NFT back. On the other hand, if the owner fails to repay the debt, the protocol would liquidate the collateral and book the profit.
The current landscape of the NFT lending space.
Peer-to-Peer lending has two main applications depending on the intentions of the liquidity lender (NFT borrower):
The liquidity lenders borrow the NFT prices they would be comfortable buying the NFT at. Therefore, this strategy is like selling puts on NFTs. The liquidity providers either get paid their principal back and book the interest or the borrower defaults and they get the NFT.
The liquidity providers are willing to temporarily get lend their liquidity to get access to the NFT for a while. This behavior is expected to happen if they want to take advantage of any use case that the NFT may have (such as access to events, boosts in yield farms, or maybe just to flex it).
Despite its interesting use-cases and NFTfi’s growing popularity, since P2P lending is extremely reliant on each individual’s personal valuation of NFTs, it won’t be further analyzed in this paper. Instead, we will focus on B2P Lending protocols.
Bridging the gap
We have realized that there is a use case for NFT lending and that the market demand is extremely likely to be there. Nevertheless, to effectively bridge the gap between DeFi and NFTs and achieve product-market fit, it is crucial to build a solid product first.
How do lending protocols work?
Before going down the NFT lending rabbit hole, let's consider the mechanisms and properties that make a successful lending protocol. To do so, let's look at the properties that the leading DeFi blue-chips (AAVE and Compound) and some of their competitors/forks (Rari Capital) have in common.
Interest rates: Rates on the lender and borrower sides are usually determined by a curve. The more an asset is utilized, the higher the rate lenders earn and borrowers pay. Generally, the lower the spread between both rates, the more attractive it will be for users.
Terms: Determined by a risk team that assesses market conditions to minimize risk to protocol collateral. Terms dictate how much a user can borrow, at what point they are liquidated (max loan-to-value), and how much they pay when they are liquidated.
Terms also set which assets can be provided as collateral and which assets can be lent. Usually, whitelisted assets have deep liquidity and broad adoption, so the risk of suffering exploits is reduced (the deeper the liquidity the more difficult it is to manipulate the price oracle for a sustained amount of time). Some forks such as Rari Capital use permissionless pools, so any type of asset can be used but their risk is limited to the pools they belong to.
Liquidations: Unlike traditional loans, on-chain loans are liquidated autonomously when a user's LTV ratio falls below certain thresholds. A large enough drop in the value of someone's collateral will allow other users to bid for the collateral, sell it, and pocket a profit. The health of the lending protocol depends on the efficiency of the liquidations. At the same time, one of the key factors in the efficiency of these liquidations revolves around the pricing oracles. By having a trusted price feed, the protocol can prevent bad debt or, at least, reduce this risk since there are other factors involved. On top of that, a reliable price feed also minimizes the risk of exploits.
Now that we understand how lending protocols work, let’s apply those principles. This time, instead of using ERC-20 tokens as collateral, we will be using NFTs.
Reliable NFT valuations are key for mass adoption
The main reason why financial products haven't been built on top of NFT yet is the difficulty to appraise them.
Because of their scarcity and illiquid nature, the industry has missed a mechanism, such as an AMM or an order book, that can constantly value them fairly. So far, one of the only protocols that tried to do so (by using a combination of NFT vaults and ERC20 tokens that represent a claim on a random asset from the vault) is NFTX. Despite its interesting mechanics, the protocol hasn’t been able to gain wider market adoption because they struggle to get deep-enough liquidity.
Note: Protocols such as FloorDAO, which can be described as a decentralized NFT market-making protocol, aim to make the most out of NFTX (while increasing its liquidity at the same time) by using OlympusDAO’s bonding mechanism to accumulate productive NFT liquidity, which is then deployed into NFTX vaults to generate yield.
Honorable mention: Although there aren’t any AMM or order books for NFTs yet, the Sudoswap (a decentralized OTC protocol) team is building the SudoAMM. A concentrated liquidity AMM designed specifically for NFTs. The SudoAMM aims to take over Opensea by lowering the fees and offering continuous liquidity pools while maintaining buy and sell walls (like bids/asks).
Nevertheless, as showcased in the previous figure, in this piece we will focus on the lending landscape, where 2 different mechanisms that aim to continuously value NFTs have emerged.
Oracle-based Pricing
The first and most intuitive type of pricing mechanism relies on the usage of oracles to retrieve on-chain data that help predict the fair valuation of NFTs.
This category can also be split into different categories based on the information used and the complexity of the estimations.
Price Oracle: The most basic estimations just use a price oracle to value an item.
A common practice among protocols is to use the floor price of the collection (the lowest price at which a piece of that collection is being sold at a given time) since theoretically, it ensures that they won’t issue loans greater than what the market will pay for an item of that collection. Other common valuation methods consist of accounting for the lowest sale in a given period, or the average of the X lowest sold items in a given period.
Most protocols decide to start with one of these simple approaches and build more complex valuation models afterward.
Oracle + Item traits: The previous method values all the items of a collation at the same price, failing to acknowledge the uniqueness of each NFT. Because of that, some pricing mechanisms also take the attributes of each item (and their rarity) into account when estimating the valuations.
Since every collection is different and has unique attributes, accounting for this information in the valuations adds a lot of complexity to the pricing models. Because of that, some projects such as JPEG'd or DropsDAO will create custom valuation models for each collection and, therefore, will only whitelist blue-chip projects.
Machine Learning: Having to develop custom valuation models for each NFT collection and account for its attributes is time-consuming and not scalable. Because of that, protocols such as UpShot have developed machine learning models that take into account all the available information (item traits, last sale price of the item, last sale price of an item from the collection, etc.) to predict the fair valuation of an NFT at any given time. They claim that their ML model achieved a surprising Median Relative Error (MRE) of only 11.6% when backtesting it to appraise the Cool Cats collection
.
Optimistic Pool Pricing
The second type of lending protocol is built on top of Abacus Spot, an ambitious but still untested approach at NFT valuation. Abacus uses a combination of principles from optimistic roll-ups and proof of stake consensus mechanisms to create what they define as optimistic proof of stake. The core idea of the Spot pools revolves around the premise that the NFT real value is the amount of ETH that is locked (staked) by the liquidity providers (validators) in its corresponding pool.
Since this system is new and not as intuitive as the traditional pricing models that were previously described, we will dive into it further. Let’s try to answer a couple of questions.
How does the Abacus valuation system work?
Users can buy a position in a given pool by locking their ETH for a desired period of time (between 1 and 12 weeks). The aggregation of all the active pool positions determines the market valuation of the NFT.
Due to the perishable nature of the positions, unless new liquidity joins the pool, the valuation of an NFT will drop as positions unlock and become inactive.
Why shouldn’t the valuation be overpriced?
Because the liquidity providers lock their ETH for a given period of time and the pool can be closed at any given moment.
Since all the stakers are liquidated in a FIFO manner, whenever a pool is closed and an NFT is auctioned (due to liquidation or because the owner thinks that the pool valuation is too overvalued) their “truthfulness” is tested by exchanging the pool liquidity for the auction revenue.
Note that since the staked ETH is distributed on a FIFO basis, if the NFT is overvalued, those who overvalued the NFT are penalized and their capital is slashed.
Why would people lock their ETH in a pool to help determine the fair valuation of an NFT?
Because of incentives. When a pool is closed by the NFT owner, all the stakers are liquidated by following a FIFO logic. If an NFT is sold at a price above the determined by the pool, the premium is split among the liquidity providers. Since late stakers incur more risk, they are further incentivized to stake by getting a bigger cut of the fees.
On top of that, when the pool is still active all stakers receive yield (similar to nodes in a proof of stake system) in the form of a decreasing cost basis for ABC tokens in exchange for validating the price and guaranteeing the integrity of that pool’s NFT valuation.
Comparing pricing methods for lending protocols
When analyzing the strengths and the flaws of each valuation method, it is important to not only account for their accuracy (versus the market price at which the assets are finally sold), but also for how they impact the mechanics/behavior of the lending protocols.
Ability to accurately predict the price of future sales:
The valuation system that Abacus offers is characterized by the combined perception of the pool stakers. Therefore, we can say that its valuation is dependent on the knowledge and accuracy of these actors.
It seems reasonable to assume that the strategies that these actors will use to value the NFTs are the same as the ones used by basic pricing oracles (mainly floor price and item rarity based on its traits). Because of that, it is to be expected that the Abacus valuation system will be more accurate than most basic oracle pricing methods.
Nevertheless, it also seems unlikely that these human appraisers can outperform machine learning algorithms (especially if they keep improving) in a sustained manner and across all the NFT collections by only using their knowledge. Therefore, protocols such as Upshot seem to have a chance of becoming market leaders when estimating the fair valuation of NFTs.
Note: If Upshot’s appraisals are proven to be really accurate, it is also to be expected that people will use their price estimations in combination with the Abacus Spot pools to get as much yield as possible on their ETH. Therefore, it is likely that Abacus pools end up reflecting Upshot’s valuations over time.
Ability to guarantee profitable liquidations:
Despite believing that Upshot, or any other trusted appraisal model, can achieve great accuracy, lending protocols must guarantee that if bad debt needs to be liquidated, they will be able to do so at the expected price. Unfortunately, prediction is different than certainty. Because of that, lending protocols that are based on prediction models (what we called Oracle-based pricing methods) will always need to establish a somehow conservative LTV regardless of their accuracy.
Why do lending protocols based on prediction models need to have a conservative LTV?
Even in liquid asset lending (i.e. ETH or DAI), loans are overcollateralized. Lenders do not want to risk their assets, therefore they want to avoid any potential insolvency issues.
Unlike AAVE or Compound, due to the illiquid nature of NFTs flash loans are not an option here. Because of that, NFT lending protocols need to be even more careful with liquidations. Unless they are willing to hold the NFTs themselves (which is unlikely since it is extremely risky), they will have to wait for someone to liquidate the loan by buying the NFT at a discount.
In summary, regardless of their appraisal accuracy, they will have to use conservative LTVs because they must avoid bad debt and guarantee that they will find a buyer when liquidating an NFT.
Is Abacus any different?
Yes, it is completely different. Thanks to its design, lending protocols that are built on top of Abacus do not need to worry about liquidations. The mechanics ensure that there will be a known amount of liquidity at a known time schedule. This property gives certainty to the protocol that it will be able to effectively liquidate the NFTs without having to rely on third parties.
If a loan needs to be liquidated, they will take ownership of the NFT and close the spot pool. Then, they will get the staked ETH and the NFT will be auctioned in the open market. If it was overvalued the stakers will get slashed and if it was undervalued they will book some profits.
Thanks to this intrinsic liquidity property, lending protocols built on top of Abacus will have the ability to issue loans at a max LTV of 100%. Note that some protocols may opt for a max LTV of 100%, and others (to ease loan managing efforts on the borrower side) may adjust it based on the minimum ETH amount locked in the pool before maturity.
As it can be seen in the figure, since stakers commit for a period of time, borrowers also benefit from the predictability of the price evolution. They know with certainty the minimum price that their NFT will have 12 weeks in advance, and in the best case, some new liquidity providers may join the pool and even stake more ETH. This property is a high value add from the borrower’s side, because it helps them plan ahead the maximum amount of ETH that they can borrow at any given time.
But so far, Abacus is only an idea. It has to prove its value to the market and it will have to bootstrap liquidity to succeed. Note that the whole value proposition relies on being able to attract enough liquidity to its pools so that any NFT can be properly priced. Therefore, if the protocol doesn’t manage to bootstrap enough liquidity it will end up as only a nice idea.
How will Abacus bootstrap liquidity?
Because of its tokenomics design, Abacus will only manage to bootstrap liquidity if users are interested in acquiring ABC tokens. As explained before, the rewards for helping secure the Abacus network are issued in the form of ABC tokens. Therefore, people will only lock their ETH (and secure the network) if ABC tokens are valuable enough for them.
The value proposition of ABC tokens can be split into several use cases.
Pay network fees: ABC is required to do certain core actions in the Abacus ecosystem such as buying into pools, creating pending orders (priority access) to lock ETH once positions mature, or closing pools.
Governance voting power: ABC can be vested in exchange for voting escrowed ABC (veABC). Voting participation in Abacus governance decisions will be limited to veABC holders.
Earn platform revenue: One of the biggest benefits for veABC holders is that 90% of all ETH revenue generated by the network will be paid back to them in the form of ETH.
Gauge control: Every epoch, the protocol will offer extra emissions to the NFT collections selected by the distribution gauge control. Since these incentives will increase the EV that liquidity providers can obtain in the boosted collections, it is expected that those pools will receive higher liquidity. If these pools manage to get more liquidity, the owners of the NFTs will benefit from an increase in the amount of liquidity that they can borrow on lending protocols, increasing their expected yield.
The protocol believes that gauge control can create dynamics similar to the curve wars. Therefore, in addition to manually allocating to specific collections, collection agnostic veABC holders can auto allocate their tokens. Every epoch, interested parties can bribe auto allocators for control over allocation usage.
Brief Summary
We have seen that the protocols that already started the NFT lending race are using two different approaches to fixing the LTV challenge.
On the one hand, we have those who decided to go through the prediction route by using oracle-based methods. Their success will be directly related to their ability to accurately predict future NFT valuations, since liquidity and effective liquidations will depend on those valuations. Upshot seems to be the best-positioned protocol right now, because they have a scalable mechanism that is compatible with any collection and seems to have quite accurate appraising skills. Nevertheless, protocols such as JPEG’d or DropsDAO could get to compete if they manage to create high efficient pricing methods tailored to the whitelisted collections.
On the other hand, we have those protocols which decided to go through the Optimistic Pricing route and use Abacus as their underlying infra. These protocols are betting on an untested technology that, if proven to be successful, will make them market leaders. Optimistic pool pricing will give Gradient and Unlockd the ability to offer liquidation-free loans to users (if the positions are properly managed according to the liquidity unlock schedule) and also give them the certainty that they can effectively liquidate bad debt by accessing the underlying pool liquidity. Another huge advantage of this mechanism (as it happens with ML methods) is that it is scalable and that can be used for any collection.
Closing thoughts
Regardless of the underlying pricing infra that NFT lending protocols decide to use, it is clear that they will unleash the next expansion phase of the crypto space. A lot of idle capital will be put to work and the whole space will benefit from it. NFT owners will get access to liquidity, DeFi protocols will see their TVLs increased, and they may even onboard new users who have never interacted with DeFi yet.
Although oracle prediction-based pricing models such as the one Upshot is building may end up being extremely accurate, the intrinsic liquidity and certainty attached to Abacus make it a superior option for lending protocols.
Thanks to the predicting ML models that Upshot is building, they could pivot/expand their business model to sell it as an appraising service. The technology that Upshot is developing is very valuable and will help users better understand the fair value of NFTs and allow them to have an edge when buying/selling based on those insights. On top of that, they are already building Upshot Analytics which seems like the perfect way to make the most out of their technology. In a future where both Abacus and Upshot are successful, it is easy to envision people paying for a tool that accurately predicts the valuation of NFTs to earn yield by locking their ETH in the Abacus pools (win-win for the best technological solutions).
Right now, only Gradient Finance and Unlockd Finance are building on top of Abacus, and they could benefit from a first-mover advantage (especially Gradient, who have been working closely with the Abacus team for a while). Nevertheless, any of the listed protocols (such as JPEG’d, Upshot, OpenSky Finance, or DropsDAO) that decided to initially launch with a predictive oracle-based model, could decide to switch to Abacus too. We could even see protocols adopting both mechanisms at the same time, so that if there is more liquidity on the Abacus’ pools, they would allow loans up to that value. If Abacus’ liquidity was lower than the LTV based on the prediction pricing, they could still benefit from that liquidity as a final backstop that limits risk (in case they are unable to effectively liquidate the NFT at higher prices). That is the power that infrastructure products have, they allow others to build on top of them.
Despite Abacus technology being promising and having incredible upside, it is still to be proven. To be successful, Abacus will have to overcome the bootstrapping liquidity phase. Only time will tell us if the value proposition for holding ABC tokens is good enough and if they manage to do so.
The setup for the next big narrative is clear. The only remaining questions are: When will this narrative get traction? Which pricing infra will be broadly used? Who will manage to establish themselves as the NFT lending market leaders?
Visa Article - A program to help creators navigate NFTs
Coindesk Article - Jefferies’ predictions on NFTs
NonFungible Report - 2021 Yearly Report
Nansen Report - State of The Crypto Industry 2021
Upshot Article - Explaining NFT Pricing with Machine Learning
Check NFTBank for another example for Machine Learning base NFT appraisal service!