
Upshot’s mission is, and always has been, to enable the creation of efficient financial markets for *anything*.

To realize this, we’re proud to announce the Upshot Machine Intelligence Network, a network designed to crowdsource financial alpha produced by machine learning models and powered by our new “Proof of Alpha” reward mechanism.

The Upshot Machine Intelligence Network and Proof of Alpha usher in a new era for DeFi. Now, more sophisticated DeFi protocols can be built leveraging decentralized AI.

In this post, we dive deeper into:

- The problems we sought to solve with our network
- What the Upshot Machine Intelligence Network is
- How it works
- The role zkML plays in it
- The applications being built on top of it, such as Ditto, the first yield protocol with AI-powered strategies
- And what all of this means for the next era of DeFi

## How We Got Here
Our machine intelligence network (or MIN) is the culmination of years spent evolving and fine-tuning our approach as a company.

Two and a half years ago, we launched a product called UpshotOne. With so many disparate assets in crypto, there was no way a single entity could accurately price and create markets for it all. So UpshotOne was an attempt to leverage the wisdom of crowds to incorporate that information. To that end, we created peer prediction models to price assets based on subjective opinions from “experts” in the space (e.g. whales, collectors, traders).

![Hot or Not for NFTs](//images.ctfassets.net/lhkcghnr8ql4/4dRCxzA11dfl6ZaEkypn1c/eb269721c71879bf88db9ee15c564504/Hot_or_Not_for_NFTs.jpg)

UpshotOne showed promise, but human crowdsourcing alone wasn’t scalable or accurate enough to be a sustainable solution for the long-tail pricing problem. Moreover, we learned that we needed a solution to reward people based on the efficacy of their contributions.

So we explored other options.

We realized that using AI would be far more powerful than crowdsourcing alone. Our models would allow us to unlock more accurate and scalable price predictions for assets, which could power new innovations in lending, analytics, and trading strategies. Initially, we focused solely on developing AI models for NFT price predictions. This allowed us to build:

- Lending strategies
- Market making strategies
- Appraisals
- Indices
- Analytical tools
- And more

![upshot appraisals](//images.ctfassets.net/lhkcghnr8ql4/6C3DSLNljKYhYj1b0J1xYb/918f05aabf6ea8d52f75f969315bf071/upshot_apprasails.jpg)

As our models proved useful across crypto, not just NFTs, we expanded our infrastructure to encapsulate more markets (such as RWAs, gaming, and DeFi more broadly).
But as a single entity, we ran the risk of centralization. One model creator isn’t sustainable enough to support the entire universe of potential markets or AI-influenced primitives. And there was no way, even with AI, we could possibly coagulate all the financial insights in the world without passing through the crowd at some point.
This led us to build the Upshot Machine Intelligence Network––a distillation of all our past learnings building novel crowdsourcing infrastructure and industry-leading machine learning models.

By combining these two disciplines, we’re creating a scalable and ever-learning network of collective intelligence. It takes the best of both previous approaches––

- The efficiency of AI
- And the wisdom of the collective

––to realize our mission of enabling the creation of efficient markets for anything.

The Upshot Machine Intelligence Network is the culmination of our successes, mistakes, lessons, and experiments over the past three years.
## The Benefits of Decentralized AI
To address the challenges of rewarding crowdsourced intelligence, as well as the challenges of bringing together model creators in a decentralized way, we’ve created a novel solution that aligns incentives for all participants in a decentralized AI network.

Decentralized AI can yield multiple benefits:

1. __Collective intelligence__: A networked AI ecosystem can harness the collective intelligence of various participants, including traders, data providers, and model creators. This collective wisdom can lead to better market predictions and informed decision-making.

2. __Reducing information asymmetry__: Entities holding specialized data, such as a Rolex marketplace, may lack the necessary analytical skills to derive insights from their data. Conversely, those with data science expertise may lack access to niche data sources or the capital required to invest in such specialized markets. When participants are incentivized within a network to share insights and predictions, it becomes harder for a select few to monopolize critical information, leading to more accurate predictions.

__3. Liquidity and market accessibility__: AI-driven algorithms can enhance liquidity in markets by providing continuous, automated trading and market-making activities. This can make markets more accessible to a wider range of participants, increasing trading volume and efficiency.

__4. Market innovation__: In a networked AI environment, market innovations and improvements can be developed collaboratively, leading to more efficient trading mechanisms, fairer market structures, and new kinds of financial products.

To realize these benefits of decentralized AI, we developed a mechanism that not only incentivizes contributions to the network, but also ensures that rewards are distributed based on the accuracy and usefulness of each contribution.

That mechanism is called Proof of Alpha.
## What Is Proof of Alpha?
Proof of Alpha is the core reward mechanism powering the Upshot Machine Intelligence Network.

Upshot MIN participants contribute alpha in the form of proprietary data, predictive features, machine learning models, and other elements aimed at enhancing the network’s ability to optimize specific objective functions. The participants who contribute the most useful alpha will earn the most rewards.

![proof-of-alpha](//images.ctfassets.net/lhkcghnr8ql4/5DFNBjvpBKexbzt8Niqfpk/0510e4d7a9e52bfc4a8fd25807434f98/proof_of_alpha_imagery.jpg)

The result is a system somewhat akin to a perpetual prediction market, with regular rewards distributed based on contribution quality and accuracy. Proof of Alpha ensures the most effective models for each subnetwork are compensated accordingly.

Historically, profiting from unique insights required extensive business infrastructure:

- Attracting investors
- Hiring a team
- Customer outreach
- And more

This precluded most people from financially benefiting from their alpha.

Now, because of Proof of Alpha, the Upshot MIN enables anyone to contribute specialized data/insights and receive rewards based on that alpha’s measured accuracy. Almost no other overhead required.
## How Does the Upshot Machine Intelligence Network Work?
A MIN is a new kind of decentralized network that brings together different ML models and incentivizes them to optimize some objective together, such as:

- Minimizing mean absolute direction loss in predicting the future price of an asset
- Maximizing Sharpe ratio with a proposed portfolio of assets
- Forecasting volatility of an index of assets of a specified time period

The power of this networked approach is amplified through the deflationary capabilities of machine learning and AI. These technologies use vast computational power and data-driven algorithms to uncover insights, make predictions, and drive optimizations that are far beyond the scope of human cognition or what can currently be expressed on-chain.

![deflationary-ai](//images.ctfassets.net/lhkcghnr8ql4/2edAjvAFCOXfRUOUd6ttSs/40f19dbba2521b30eabb6dbf1c4b4b4b/deflationary_AI.png)

This abstraction provides the bedrock for an abundance of use cases; any application that could benefit from ML can be improved by sourcing its insights from a network of collaborative machine intelligence.

Here’s a more detailed breakdown of how it works:
### 1. Create or Find the Best Subnetwork for Your Alpha
Each subnetwork in the Upshot MIN is defined by a specific objective function for participants to optimize around. For example, subnetworks might focus on the following objectives:

- Predicting the future price of an asset, such as the price of ETH/USD in 1 hour or the median sales price of CryptoPunks over the course of one week; e.g. minimizing the mean absolute directional loss of those predictions.
- Managing a portfolio of crypto assets; e.g. maximizing the Sharpe ratio of that portfolio.

Once you create or find the subnetwork best suited to your alpha, you’ll stake in it and start contributing alpha to earn rewards.
### 2. Start Mining Alpha
After joining your subnetwork of choice, you’ll provide your alpha in the form of:

- Relevant data
- Predictive model features
- ML models

The quality of your alpha will be represented by its “stake,” or the amount of money placed behind it, signaling its worth. You may place a stake in your own expertise, and others may delegate their stake to you, adding to your total stake. Your total stake represents how influential you are in your chosen subnetwork, and will fluctuate based on the realized usefulness of your alpha.

You, along with other alpha miners in your subnetwork, will regularly learn from each other, boosting your respective models based on the results of your peers. You’ll also score your own peers as well based on how useful their contributions were. Usefulness is measured by how well the subnetwork’s objective function was optimized.

![proof of alpha peer scores](//images.ctfassets.net/lhkcghnr8ql4/1xiBqDmEligmSuznJFHaIS/74b28ab87bd3c8ad68d0d23187311fc8/proof_of_alpha_peer_scores.jpg)
### 3. The Proof of Alpha Mechanism Aggregates Scores and Updates Alpha Miners’ Stakes
This process by which the network aggregates scores and updates participants’ stakes is called “Proof of Alpha.” It’s used to reach consensus on how useful each participant's alpha is to the subnetwork’s objective. Specifically, a miner’s Proof of Alpha score determines how many rewards they receive as incentive, or how many are deducted as penalties.

Whenever participant scores are updated, the network distributes (or slashes) rewards related to contributors' scores and stakes; participants with higher stakes earn or lose more rewards relative to participants with smaller stakes, and participants earn or lose depending on the direction of their score. This means that more insightful participants “rise to the top,” meaning that the network becomes smarter over time.

Earnings are paid for by network consumers–those who request predictions from the network. If they’re interested in querying for, say, the price of some NFT, they’ll pay a fee to the subnetwork devoted to pricing NFTs. Given finite computational resources, as demand for such predictions grows, so do the fees. In other words, more adoption begets more rewards for those supplying the most useful alpha.
### 4. A Novel zkPredictor Verifies Model Outputs
If we stopped with the network as described so far, we’d have a mechanism that generates increasingly accurate predictions of future events. However, for these predictions to be maximally useful, they need to be delivered on-chain, which raises a problem:

__How can an on-chain application verify that predictions received actually came from models of a particular set of miners?__

Imagine a DeFi platform that uses AI for high-frequency trading to profit from arbitrage. You see it 100x your friend’s portfolio, so you want to invest too. But how can you verify the platform is using the same profitable models to manage your capital?

Zero-knowledge (zk) proofs would prove pivotal in this situation and many similar situations in DeFi. With zk proofs, miners can verify that predictions came from their proprietary model without revealing their model and therefore revealing their IP and means of recurring revenue.

Fortunately, recent advances in zkML technology allows us to do just that. Hence, we’ve partnered with [Modulus Labs](https://www.modulus.xyz/) to create the first zkPredictor, which leverages zkSNARKS to verify the output of certain machine learning models. Our first version of the zkPredictor uses a novel proof system called Remainder, built by Modulus Labs, to cryptographically verify the outputs of tree-based machine learning models (a class of models that [is most adept](https://arxiv.org/pdf/2207.08815.pdf) at tackling ML problems with tabular datasets, which describe many financial applications of ML).

We plan to seed our network with a zkPredictor utilizing a model of our own, a verified version of our flagship long-tail asset valuation model, which has been exhibiting industry-leading performance in predicting NFT prices for over 2 years.

This zkPredictor showcases a substantial leap forward in the practical use of zkML. With this new zkPredictor, the number of verifiable outputs may approach 10,000 per hour (previously, the record was nearly 12 outputs per day), a throughput suitable for early crypto x AI applications. As the network expands, participants can apply a zkPredictor of the same capacity to their own models.
### 5. DeFi Protocols Benefit from Curated ML Models
The Proof of Alpha reward mechanism allows the network to curate the best alpha and ML models into a “meta-model” that represents the collective intelligence of a given subnetwork. The results of this meta-model can then be provably linked to reputable models that have already demonstrated their efficacy.

At this point, model results are ready to be consumed by applications, especially DeFi protocols, which can now remain decentralized while benefiting from the efficiencies of ML. This is especially important for protocols that manage larger amounts of capital, such as:

- Perpetual platforms
- Decentralized insurance platforms
- Investment DAOs using AI strategies to help manage their treasuries

What’s more, our novel foundation for serving decentralized ML can power a variety of novel use cases, such as:

- Asset managers can create manipulation-resistant index funds for niche markets.
- Lending platforms can offer more unique collateral types and advanced term logic with lending terms created by machine intelligence.
- Tokenized vaults can use advanced AI-powered strategies for potentially higher yield.
- And much more.
### Ditto: The First App on the Upshot MIN
The first application built on the Upshot Machine Intelligence Network will be [Ditto](https://dittohq.xyz/), a decentralized yield protocol that gives users passive access to verifiable, AI-powered strategies.

![ditto header](//images.ctfassets.net/lhkcghnr8ql4/pj81piByKGFI1ZyJu8KjK/299bfb65328461b1c5f75dd0af8cf17c/ditto_header.jpg)

DeFi has shown us how decentralization, composability, and verifiable computation can enable us to significantly speed up the rate of innovation in our financial system. However, existing DeFi primitives are currently limited by what can be expressed on-chain, thus restricting strategy sophistication and therefore limiting yield.

In Ditto, users deposit capital into on-chain vaults, each managed by machine-powered strategies from the Upshot Machine Intelligence Network. The goal of each strategy is to use the capital in the corresponding vault to generate yield for its depositors. These strategies are proven to have been made by these models using our pioneering zkPredictor.

![ditto how it works](//images.ctfassets.net/lhkcghnr8ql4/788nLQCv0tyCoD5Sd3es0j/a59dee0e6d690affd7bce41fba1e52a9/ditto_how_it_works.jpg)

Ditto combines passive on-chain capital provision with advanced AI-powered strategies and zkML technology, bringing more powerful, higher-yield strategies to DeFi and NFTs.
## Monetize Your Alpha
At first glance, the idea of a decentralized machine intelligence network may seem abstract. Yet its roots stem from our mission of enabling the creation of efficient financial markets for anything.

The Upshot Machine Intelligence Network incentivizes the collaboration and aggregation of machine intelligence to generate financial alpha for on-chain applications. It provides a means for people to earn from their unique insights, thanks to the Proof of Alpha reward mechanism.

At the same time, it enables the development of advanced protocols like Ditto, which leverage the wisdom of decentralized AI networks. As more developers plug into the network, the financial markets the network supports will grow more sophisticated.

But with that growing sophistication comes a need for security, transparency, and fairness. That is why innovations like the zkPredictor are so critical: they ensure decentralized AI is provably trustworthy.

With the combined powers of decentralized AI, Proof of Alpha, and the zkPredictor, we believe the Upshot Machine Intelligence Network will create new opportunities to build sophisticated, inclusive, and trustworthy financial systems.

![dev graph](//images.ctfassets.net/lhkcghnr8ql4/7Cwe8aI6Q5zD0tgRk1Yx29/7237c17c4c01231e91203da653d28a4f/dev_graph.jpg)
### How to Get Involved
If you’re interested in joining the Upshot Machine Intelligence Network, you can choose one of the following roles based on your skills and preferences:

- __Data providers__: If you have access to proprietary or valuable data that can be used to train machine learning models, you can become a data provider and earn rewards for sharing your data with our network. [Learn more here](https://docs.upshot.xyz/docs/data-provisioning).

- __Alpha miners__: If you have expertise in machine learning and can create or improve models that generate useful alpha, you can become an alpha miner and earn rewards for contributing your models to our network. [Learn more here](https://docs.upshot.xyz/docs/alpha-mining-1).

- __Developers__: If you need financial insights for your on-chain applications or DeFi protocols, you can become a developer and get access to the aggregated insights from our network. [Learn more here](https://docs.upshot.xyz/docs/protocol-overview).

- __Join the community__: If you’d like to contribute to the research and continued development of the Upshot Machine Intelligence Network, you can join our community Discord. [Join our Discord here](https://discord.gg/upshot).

Our new Machine Intelligence Network leverages decentralized AI and Proof of Alpha to enable advanced DeFi protocols and opportunities for anyone to contribute and monetize their financial insights.

Announcing the Upshot Machine Intelligence Network

Let’s face it: NFT interest and trading volumes have declined significantly in 2023. However, lending against NFTs remains an important option for investors seeking liquidity.

To make the most out of the opportunities available to us in a bear market, it’s important to determine appropriate lending terms. This requires carefully calculating recovery rates–i.e., the percentage of collateral value recovered in default. This is essential for assessing your risks, even if fewer deals are being written.

In this blog post, we’ll dive into the calculation, usage, and adjustment of recovery rates for NFT lending.

Understanding this process provides key insights, regardless of market conditions.

## The Complex Calculations Behind NFT Recovery Rates

Determining appropriate NFT lending terms requires estimating recovery rates. But how exactly are these rates calculated for different NFT collections?

We use some robust calculations. To start:

![recovery rate formula](//images.ctfassets.net/lhkcghnr8ql4/6QgOJuwa1bveSxX05fbhhT/8a1981885ff88c14aa7b91f3da98a94e/recovery_rate_formula.png)

In this calculation, recovery rate R serves as a forecast of the change in an NFT's appraised value over a fixed time horizon. This rate is expressed as a relative change, where R < 1 signifies a decline in asset value, while R > 1 indicates an increase.

Recovery rates are inherently uncertain, and 
therefore, we treat R as a random variable following a skew-normal distribution.

To estimate the parameters of this distribution, we utilize the daily relative changes in floor prices based on Upshot appraisals, considering them as realizations of R.

It's important to note that when using floor prices, the resulting distribution applies to an entire collection, assuming that recovery rates remain consistent across assets within the same collection.

For point forecasting purposes, the mean of each skew-normal distribution serves as an estimate of the expected recovery rate for a given collection and time horizon.

Additionally, the default probability of a loan for a specific time horizon is determined by the probability that an asset's value depreciates below its appraised value at the loan's inception.

This probability can be derived from the cumulative distribution function (CDF) of R, specifically:

![recovery rate formula 2](//images.ctfassets.net/lhkcghnr8ql4/3I5eRhdRUz3u4W2q6sA6qD/fbe7a62a783cf32bed27601120346fda/recovery_rate_formula_2.png)

While these calculations provide valuable insights, it's crucial to acknowledge that they assume rational investor behavior.

It can be challenging to model and account for irrational behavioral patterns, which may impact the accuracy of these estimates.

To address this uncertainty, lenders must validate and adjust recovery rate forecasts to account for real-world irrationalities.

One approach is to compare projected returns to actual market performance over time. The gaps between projections and reality can then be used to refine your recovery rate estimates.

## Adjusting Recovery Rates for Forecasting Uncertainty

To assess forecast performance, we compare forward-looking returns based on floor prices with the forecast distribution means over rolling windows of 60 days. By calculating the mean absolute percentage error for each time horizon, we can adjust the recovery rate downward to account for forecasting uncertainty.

This adjustment, known as the "uncertainty margin," is calculated as the uncertainty-adjusted recovery rate equals the recovery rate multiplied by (1 - uncertainty margin) for each collection and time horizon.

With rigorously validated recovery rate estimates, lenders can incorporate these projections into competitive lending terms.

## Utilizing Recovery Rates in the Lending Process

The recovery rate model's output serves as a critical component in determining two key aspects of the lending terms: the loan amount and the interest rate.

### Loan Amount Determination

The uncertainty-adjusted recovery rate plays a pivotal role in determining the loan amount. For each collection, the loan amount is calculated as the maximum of the uncertainty-adjusted recovery rate multiplied by the appraised value of the asset.

The maximum uncertainty-adjusted recovery rate corresponds to the time horizon selected as the loan duration.

However, to prevent concentration risk and ensure a diversified lending strategy, the loan amount is capped at twice the collection-level loan median.

### Interest Rate Calculation

The interest rate for each asset is calculated at the collection level using the formula:

![Interest rate formula](//images.ctfassets.net/lhkcghnr8ql4/5txevxMsgxShLHHhNbNBUl/2963beaaf68539c981312403bb135ecc/Interest_rate_formula.png)

The risk-free rate is set at 4.3%, reflecting the long-term average of the 10-year US Government bond.

The premium, set at 0%, 2.5% and 5% depending on each NFT’s value bucket, is added to compensate for the additional risk associated with NFT lending.

The default probability utilized in this calculation corresponds to the default probability associated with the best recovery rate. The term (1 - default probability) increases the overall interest rate as the default probability rises, effectively compensating for default risk in the loan portfolio.

Our methodology offers various levers that can be adjusted to tailor the lending terms and enhance competitiveness.

### Interest Rates

The premium contributes to the interest rate formula and can be adjusted to make the rates offered more competitive.

### Loan Duration

Currently, loan durations range from 14 to 30 days. Depending on demand and risk considerations, this range can be extended to accommodate different lending preferences.

### Loan Amount

The loan amount is based on the adjusted recovery rate at a specific time horizon. The adjustment is determined using the mean absolute percentage error. The adjustment can be increased or decreased to align with market conditions, allowing for flexibility in loan amounts.

### Range of Collections

Currently, we generate lending terms for a range of liquid blue chip NFT collections. This selection of collections provides a competitive advantage as many other NFT lenders cover a smaller number of projects.

To further enhance our offerings, we have the option to selectively add more collections to the terms provided, catering to a wider range of investor preferences.

## Building Sustainable NFT Lending with Recovery Rates

Recovery rates are indispensable for structuring competitive, risk-aware NFT lending. Thoroughly understanding rate calculations lets lenders estimate collateral value retention across time horizons. Validating projections and adjusting for uncertainty creates actionable forecasts to inform loan terms.

Ultimately, recovery rates provide the foundation for sustainable NFT lending. Rather than reacting irrationally, lenders can rely on data-driven forecasts to confidently support investors long-term.

Careful rate analysis transforms lending from speculation to discipline. As NFT markets progress, recovery-focused strategies separate resilient programs from short-lived trends.


Learn how carefully calculating, validating, and incorporating NFT recovery rates into lending terms enables sustainable programs that withstand bear markets.

How to Calculate Recovery Rates for NFT Lending (Even in a Bear Market)

It’s easy to get lost in the day-to-day fluctuations of a digital community’s:

- Social media engagement
- Website traffic
- And, in the case of NFT collections, floor prices

But as we always say: __floor price alone don’t tell the full story of a digital community’s value.__

Moreover, sometimes a dip in a key metric might look concerning on the surface, but may not actually signal impending doom like many assume.

To truly understand the health of a digital community, you must look beyond the standard metrics. In this post, we’ll break down how to do an in-depth community analysis—focusing on everything from network structure to engagement trends.

We’ll use a popular NFT collection ([Azuki](https://www.azuki.com/))  as a case study to showcase these techniques. But the insights apply to assessing any online community.

Analyzing how this community responded after the launch of a new collection uncovered some counterintuitive insights. Not least of which, their asset prices dropped but their community grew larger and stronger by many measures.

But we’re getting ahead of ourselves.

First, let's look at how to analyze a community's strength.

## Conducting Network Analysis
Most real world and digital communities can be thought of as networks: sets of people connected together.

Everything from social networks to metabolic networks fit this topology. Most networks have common properties. For example, they often follow a power law distribution, where a small number of nodes are highly connected hubs, but most nodes only have a few connections.

Analysis of network structure can reveal insights about the nature of the underlying community. For example, many networks follow a power law distribution, with a few highly connected hubs and most other nodes having just a few connections.

For NFT networks, this likely represents a few whale wallets transacting with many others, while most holders only interact with a small subgroup.

## Quantifying Subgroups Through Community Detection
One key technique used in network analysis is community detection. This involves identifying densely connected clusters of nodes, with sparser connections between clusters.

Intuitively, these clusters represent subgroups within the broader network - like friend groups within a social network. In an NFT collection, detected communities likely represent groups of closely linked collectors and holders.

The cohesion of communities can be quantified using modularity. Networks with high modularity have dense connections inside communities but sparse connections between them.

So when examining an NFT collection, detecting communities and calculating their modularity provides insights into the composition and connectivity of holder subgroups––going beyond just individual wallet links and floor price.

## Analyzing the Azuki NFT Network
To analyze the Azuki NFT community, and determine the strength of their network beyond the headlines and floor prices, we looked at transaction data over a 6 week period in mid-2023. The snapshots of the collection took place just before and after their Elementals mint on June 27th.

We extracted two key datasets:

1. __Ownership data__: how many Azuki NFTs each wallet held
2. __Transaction data__: a network showing Azuki sales and transfers between wallets

![azuki image](//images.ctfassets.net/lhkcghnr8ql4/HLjvxUMTU6zteB1M6qYtM/0b09f85247ce83e37eb1bf32dfebf536/azuki_image.jpeg)

# How Ownership Transformed
Now, many Azuki holders complained that they experienced technical issues with the mint. Some were unable to mint any NFTs from the new Elementals companion collection. Other holders complained that the art for the collection looked almost identical to the original collection. An egregious error for a hefty 2 ETH mint price.

These factors led to a massive drop in floor price, and many criticizing the project.
![azuki price chart](//images.ctfassets.net/lhkcghnr8ql4/1KyLtbeJJqACKyuQkFVysE/4d278678eb353d4d79e93640d3b9336c/azuki_price_chart.jpeg)
*Image courtesy of [@mrjasonchoi](https://twitter.com/mrjasonchoi) on X.com*

But network analysis shows us that floor price, headlines, and twitter sentiment don’t tell the whole story. 

The data we analyzed showed a steady increase in unique holders––i.e., more community members––from 3,381 in the weeks preceding the mint to 4,372 just a couple weeks after the mint.

We also saw the number of wallets making transactions surged after the Elementals release, from 1,006 to over 6,500 weekly.
![Azuki ownership analysis](//images.ctfassets.net/lhkcghnr8ql4/73qTy1J9g3dwaN8NDPcVU1/81cb712fd4160eda54e6ea5cfb99ffe0/Azuki_ownership_analysis.png)

Also, analyzing the top 100 Azuki holders showed around 60% remained consistent holders before and after the release. So, despite the headlines, the "whales" were mostly staying put.

![Azuki whale analysis](//images.ctfassets.net/lhkcghnr8ql4/3Cp1tNMdrqou4jVczVYuiY/d972fd46602fd63a96161317a084072e/Azuki_whale_analysis.png)

# Insights on Azuki’s Transaction Network
In addition to ownership data, we also looked at the network of Azuki transactions between wallets each week. Studying patterns in these transactions revealed insights about the community’s structure and engagement over time.

One metric we examined was __assortativity__: the extent to which similar wallets connect to each other. High assortativity indicates wallets tend to connect to other similar wallets.

For NFTs, this could mean whales transacting with other whales, or collectors within the same community subgroup trading with each other.

To calculate assortativity, we first used community detection algorithms to identify subgroups of closely connected Azuki holders. We saw the number of communities increase over time as more new holders joined.

This image is a visual representation of the Azuki community’s connections before the Elementals launch (between June 19th and June 26th):
![Azuki community pre launch](//images.ctfassets.net/lhkcghnr8ql4/7nrgWXj4oc0sIzIfZfPfuD/fee05411e5bda44b7596e3d9bafe0471/Azuki_community_pre_launch.png)

This image is a visual representation of the network after the launch, from June 27th to July 3rd:
![azuki community post launch](//images.ctfassets.net/lhkcghnr8ql4/6TioWd0t1nepvOFxJL57tM/2d03a0b615082ab75fcbd82f5731880b/azuki_community_post_launch.png)

As you can see, the community grew significantly, and became more interconnected. This suggests a stronger, more resilient network.

We also saw shifts in the key buyers and sellers week to week. No single whale dominated - different wallets emerged as hubs each period. This was another positive indicator, which shows a diverse network not dominated by single nodes (so to speak).

__Back to the assortativity calculations:__

The lower the assortivity, the less alike each wallet is. Implying a robust, diverse community that’s part of many communities.

The assortivity dipped when Azuki Elementals launched.

![assoritivity ratings](//images.ctfassets.net/lhkcghnr8ql4/4z7xk5mxaXExH23PM0H7Js/1ba2d2f2f74f91446ba3a96f216d7e45/assoritivity_ratings.png)

In short, these network insights suggest the introduction of Elementals brought in new holders from across groups. Azuki's community was diversifying and strengthening despite price drops.

## Key Takeaways
While we focused on one specific community, these network analytics can be used to understand any digital community. Some key lessons:

- Look beyond surface metrics. Network structure better reflects engagement.
- Monitor changes around major events. This reveals if new members join or existing ones leave.
- Identify core members and influencers within detected communities. Engage these VIPs.
- Watch for new hubs emerging over time. Distributed activity is healthy.
- Consider incentivizing key members during volatility to prevent mass exodus.
- Establish network baseline metrics early on. Track how they evolve across milestones.

Of course, every community is unique. But network analysis provides a data-driven toolkit to gain actionable intelligence on your digital community’s health.


Floor price doesn't reveal the full health of a digital community. We break down techniques like network analysis, engagement trends, and community detection to gain deeper insights.

How to Quantify a Digital Community’s Strength

Despite ever-changing market conditions, the NFT lending space represents immense potential for traders, lenders, and developers.

Indeed, its propensity for transformation is part of what makes NFT lending so fruitful.

But with potential comes risk. You must have a strategy to protect yourself, your profits, and your assets.

- Lending on risky platforms with low collateralization ratios could lead to unnecessary losses for you.
- Failure to diversify your portfolio across platforms, collections, and [asset buckets](https://upshot.xyz/post/asset-buckets) could leave you overexposed.
- You might miss out on opportunities for more attractive yields.

This article provides strategic insights on evaluating lending platforms, setting prudent loan terms, diversifying NFT assets, managing liquidation risks, and more.

With these practical tips and risk management strategies, you can confidently maximize returns in NFT lending while minimizing downside risks.

We’ll start by breaking down the two main types of lending models:

__Peer-to-peer and peer-to-pool.__

## What Are the Differences Between Peer-to-Peer and Peer-to-Pool Lending?

- __Peer-to-Peer Lending__: This model facilitates direct transactions between individuals. A typical scenario involves a borrower using a platform like NFTfi or Arcade, offering an NFT as collateral to secure a loan (often in WETH or DAI), then the NFT collateral is safely stored in a vault, governed by the agreed-to terms.

- __Peer-to-Pool Lending__: This model operates a bit differently. In peer-to-pool lending, liquidity providers stake tokens into pools. Borrowers can then tap into this liquidity by placing their NFTs in vaults. BendDAO is an example of a platform that champions this lending approach.

Now that we have a cursory understanding of the different NFT lending types, let’s break down a few of the most popular lending platforms, what makes them unique, and the most common ways people earn money on them.

# Peer-to-Pool Lending
Peer-to-pool NFT lending utilizes pooled funds contributed by multiple lenders. Borrowers can access these funds by staking their NFTs as collateral into smart contract vaults.

__This provides easy liquidity without needing to sell your NFTs.__

For lenders, peer-to-pool lending offers a hands-off way to earn attractive yields. By supplying capital to lending pools, they can passively earn interest and platform fees based on pool activity.

In general, this model provides higher liquidity than direct peer-to-peer lending.

### Lending for Enhanced Yield on BendDAO
BendDAO in particular offers a lucrative proposition for lenders:

By simply depositing ETH, you can watch your investments grow as the platform lends it out to eager borrowers. But the benefits don't stop there.

Those who stake the platform’s BEND token (and possess veBEND) are entitled to an administrative fee. This fee is calculated based on the holder's [veBEND](https://docs.benddao.xyz/portal/governance/vote-escrowed-bend-vebend) balance in relation to the overall veBEND supply.

Remarkably, this [administrative fee](https://docs.benddao.xyz/portal/governance/fee-collection-and-distribution) constitutes about 30% of the total interest revenue amassed from all NFT loans. And for the cherry on top? All these fees, paid exclusively in ETH, are disbursed to veBEND token holders on a weekly cadence.

# Peer-to-Peer Lending
Peer-to-peer lending facilitates direct loans between individuals using NFTs as collateral. On platforms like NFTfi, X2Y2, and Arcade, borrowers and lenders connect directly to agree on loan terms, and the NFT acts as security if the borrower defaults.

A key advantage of peer-to-peer lending is lower fees, as there are no intermediaries taking a cut. Lenders can also potentially earn higher interest rates than on pooled lending platforms. However, liquidity on peer-to-peer platforms can be lower as loans require manual matching. There’s also higher risk without centralized oversight.

For borrowers, peer-to-peer lending unlocks liquidity from NFTs without having to sell them. They can leverage NFTs to obtain loans tailored to their needs. Likewise, lenders can earn yields on a decentralized market outside the traditional financial system.

Overall, peer-to-peer NFT lending provides a direct and decentralized way for borrowers to monetize their NFTs and for lenders to earn yields. But it comes with the risks of a nascent market lacking intermediaries or oversight.

### How People Maximize Returns with NFTfi
NFTfi, one of the most reputable peer-to-peer NFT lending platforms, supports over 150 NFT collections. The platform employs a multifaceted approach to appraise NFT collateral value, considering factors like:

1. Current floor prices
2. Project averages
3. And valuations from [NFTBank](https://nftbank.ai/)

Borrowers benefit from the platform's robust, tamper-proof permissionless smart contract infrastructure, accessing liquidity through wETH, USDC, and DAI loans. Notably, while NFTfi waives borrowing fees, lenders incur a [5% fee](https://nftfi.com/how-it-works/) on the interest accrued from successful loans.

*__1. Lending for yield__*

This strategy can be an enticing alternative to traditional liquidity provision on platforms like [Uniswap](https://uniswap.org/).

For optimal returns, lenders might diversify by proposing multiple loan offers, each crafted with a focus on risk-minimization, such as spreading offers across different collections, and structuring conservative loan terms. Prioritizing loans on liquid assets and collections—those that can be swiftly sold if necessary—might also be beneficial.

*__2. Lending to acquire__*

Proactive lenders can leverage this strategy to make offers on desired NFTs, often securing them for a cheaper price than through an open market purchase. A boon for NFT aficionados and collectors, this strategy facilitates cost-effective NFT acquisition over time. The principles of diversified loan offers and asset liquidity still remain crucial with this strategy.

# Perpetual Peer-to-Peer Lending with Blend
Blend, [Blur](https://blur.io/)’s NFT lending platform, quickly emerged as a leader in the NFT lending space. What sets Blend apart is its deviation from the norm:

It operates independently of oracles and imposes no constraints on loan durations.

This flexibility allows borrowers to maintain their positions indefinitely, either until they repay or face liquidation. The interest rates, rather than being fixed, are shaped by market dynamics.

By channeling ETH into Blend, users not only reap yields but also accumulate Lending Points on the Blur platform. These points empower lenders to make loan offers from their Blur Pool funds. The "Max Borrow" value, chosen by the lender, dictates the maximum ETH amount that can be borrowed against a single NFT, and in turn, amplifies their point earnings.

When a borrower greenlights a loan offer, their NFT is secured, and the lender's interest earnings, denominated in ETH, commence. The loan remains flexible, allowing closure at any point to retrieve the principal and interest. However, borrowers are bound by a 30-hour window to either repay or refinance their loan with a fresh offer.

### Blur Season 2
This gamified take on lending has led to much of Blur’s early success.

The $BLUR [Season 2 Airdrop](https://blur.io/airdrop) is designed to incentivize specific user behaviors, such as bidding, listing, and lending. By partaking in these activities, Blur liquidity providers can earn rewards.

Bidding has emerged as the favorite among farmers, primarily because it demands exposure solely to ETH—a relatively low-risk asset compared to NFTs. This exposure can be further hedged using short futures, theoretically rendering farming almost risk-free.

### How Lenders Are Harvesting Blur Points
Blur Points serve as a testament to a user's engagement level, determined by the length and intensity of their participation in lending, bidding, and listing activities. In the case of lending, you earn Blur Points by loaning your NFTs to other users. Note that those points are on top of the yield you earn from the loan itself.

It’s a risky, high stakes yield strategy. Which is likely what has made it so popular. Which leads us to to the next section:

## NFT Lending Risks (and How to Manage Them)
In NFT lending, the key to maximizing your returns is effective risk management. There are several important aspects to creating a risk-resilient portfolio of NFT loans.

Let’s break some of them down here:

# 1. Setting Your Lending Parameters

Risk management in NFT lending relies heavily on the configuration of lending parameters. As a lender, you’re able to define the NFT collections you’re comfortable lending against, set the loan-to-value (LTV) ratio, determine the loan duration, and establish the interest rate. By carefully calibrating these parameters, you can ensure that your lending activities align with your risk tolerance.

For example, opting for a lower LTV ratio (e.g., 50% instead of 75%) provides a safety net in case the value of the collateral NFT experiences a sharp decline. Similarly, setting a higher interest rate can offset the risks associated with lending against NFTs.

You can also mitigate your risks by choosing shorter loan durations, such as 1-6 months, rather than longer 12-month loans. Diversifying your loans across multiple borrowers, collections, and even platforms can be prudent as well.

# 2. Navigating Loan Defaults

In the event of a default, the lender takes possession of the NFT the borrower used as collateral. At this point, as a lender you have the choice to keep the NFT, potentially benefiting from future value increase, or to liquidate it to recover their funds.

The decision of whether or not to take custody of an NFT can play a large role in your success as a lender. It may be tempting to hang onto the NFT, but if the price has dropped to the point that the borrower liquidated, you might be holding a depreciating asset.

When deciding, analyze the recovery potential of your new asset by looking at the collection's growth indicators:

- Check for positive community sentiment
- Upcoming catalysts like launches or events
- Recent sales trends

Also assess the liquidity of the collection to estimate how easily the NFT could be sold. More liquid collections generally have higher recovery rates if sold post-default.

If indicators point to potential near-term appreciation, holding the NFT may be worth your while. But if the value has dropped significantly, liquidating the asset quickly may maximize recovery.

# 3. Inadequate Appraisals
Loans are made based on a shared understanding of the appraised value of the collateralized asset. When the [valuation is inaccurate](https://upshot.xyz/post/asset-buckets), lenders risk financial loss, while borrowers risk receiving less than the actual value of their NFT asset.

It's important to benchmark your platform’s valuation models by backtesting accuracy across different market environments. Models relying solely on current floor prices aren’t enough. Sufficient out-of-sample data should be used to evaluate performance.

# 4. Concentration Risks
Diversification is an essential part of risk management for NFT lending. By lending across various NFT collections and platforms, lenders can spread their risk, thereby mitigating any negative impact if the value of a specific collection plummets significantly.

It’s also important to consider systemic risks such as cascading liquidations. If many loans from the same collection share the same liquidation price, drops in value can trigger sequential liquidations, amplifying price drops, liquidations, and more.

Furthermore, lenders can employ a risk overlay based on their current portfolio of active loans. This overlay can fine-tune the loan terms by considering various factors, such as:

- Collection-level concentration
- Ownership concentration
- The number of open NFT loans in the marketplace

Lenders can even employ more sophisticated techniques like quadratic programming to find the optimal asset weights that minimize portfolio volatility.

# 5. Market Volatility
NFT prices can fluctuate immensely in short periods of time. Regularly monitoring the market and adjusting your lending parameters accordingly can be an effective risk management strategy.

In the event of a specific collection’s value dropping, lenders might consider readjusting the LTV ratio or temporarily suspending lending against that collection until the market stabilizes.

# 6. Hedging Risks
NFT lending serves as a popular method for owners to hedge their long positions in NFTs. Essentially, taking out an NFT-collateralized loan functions similarly to purchasing a put option with an upfront payment.

If the market value of the NFT dips below the loan amount, the borrower might find it beneficial to default on the loan and surrender the NFT, thereby effectively utilizing it as a hedging tool.

However, NFT lending isn’t the sole strategy for risk mitigation. Financial instruments like perps and options offer alternative risk hedging avenues.

For example, borrowers concerned about a potential decline in [Bored Ape NFTs](https://app.upshot.xyz/indexes/yuga) could short $APE futures to counterbalance their long exposure. This provides downside protection if Ape prices decline, offsetting the impact of potential defaults on loans collateralized by the NFTs. Utilizing derivative instruments on correlated assets allows lenders to hedge risks beyond portfolio diversification.

## Future Prospects of NFT-Backed Lending
NFT-backed loans have emerged as a transformative tool, enabling asset owners to unlock liquidity in the NFT market and have been the go-to source for downside protection of NFTs. However, the journey isn’t without its challenges. Recent declines in NFT prices have cast shadows over the robustness of many NFT x DeFi protocols, as evidenced by dwindling user engagement.

There's a tendency among market participants to downplay the probability of adverse future events. Such an oversight can culminate in significant losses. It's imperative for investors to adopt a forward-thinking approach to risk management.

Despite the hurdles, the monetization and securitization of NFTs, a domain still in its nascent stages, brims with immense promise.

Learn strategies to maximize returns and minimize risks in NFT lending by understanding key differences between peer-to-peer and peer-to-pool models, setting prudent loan terms, diversifying, and more.