DeepNFTValue NFT Index

The deprecated dashboard can be found here.

WARNING: As the data provider, Flipside Crypto, has deprecated DeepNFTValue data, this markdown serves as documentation for this work. There are some learnings from this approach that I think are worth documenting for future use.

Background and Motivation

As the name implies, Non-Fungible Tokens (NFTs) are tokens (or assets) with various characteristics that make each asset different from one another. This is the "non-fungible" component, because they are not interchangeable.

This also implies that, due to differences in certain attributes, the market prices these assets differently.

You can think of it this way: Perhaps a house is worth more because there is less glare from the morning sun, or it is near a school. No two homes would cost the same, or are interchangeable, because there are plenty of factors that affect their price. This applies to NFTs as well (assuming their characteristics have differences).

Likewise, some NFTs are valued more because they have cool attributes that people want: For instance, a red panda with various clothes is worth a lot more than a plain-looking cat.

Problem Statement

Due to the reasons above, a few questions arise. To help non-crypto people reading this, I've attempted to "translate" the implications and reasoning behind the questions to help you frame them better 🙂

1. What are some price trends given the attributes of an NFT collection?

   • Translated: Are homes near schools getting more expensive over time?

2. What is the driver of an NFT's price?

   • Translated: What makes a home expensive? Distance? Number of bedrooms?

3. What is the general price trend across price cohorts for a collection? Can we compare this with other NFT collections?

   • Translated: Are expensive homes in New York getting more expensive at a faster rate than cheaper homes in New York?
   • Translated: Are expensive homes in New York getting more expensive at a faster rate than expensive homes in Texas?

Why are these questions important?

1. More expensive NFTs usually indicate a healthy and coveted NFT collection. Purchases and price appreciation in the most desirable NFTs demonstrate strong conviction, and this data aids investment decisions.

2. If an NFT investor wants to enter a collection, should they buy the cheapest NFT or prioritize quality?

Methodology

Data Scope

DeepNFTValue

The price provided is from DeepNFTValue, a company that uses machine learning to predict NFT prices by accounting for their trading history and attributes. This is necessary because not all NFTs are traded, and to get a complete picture of them, we need to "fill in the gaps."

Think of it this way:

• Home A, with 3 bedrooms and a driveway, is sold.
• But Home B, with 3 bedrooms and no driveway, isn't.

What is the worth of Home B? We can use other homes (e.g., Home A) to make a guess for Home B.

Flipside Crypto

This is a data provider that has on-chain crypto data. They also host DeepNFTValue's data and have free compute and dashboard-making capabilities.

NFT Collection Coverage

We are limited to the number of collections available from Flipside. Here is the list of NFT collections:

1. CryptoPunks
2. Bored Ape Yacht Club
3. Pudgy Penguins
4. Azuki
5. Mutant Ape Yacht Club
6. Azuki
7. Nakamigos

A Mental Model on How NFTs Are Categorized

In the NFT community, there are usually four implicit ways NFTs are grouped:

1. Grails: These are usually the rarest in terms of a particular trait or are among the top few most expensive due to a series of highly desired attributes.

2. Premium: These also have rare traits and are fairly expensive relative to the rest of the NFTs, but are a little cheaper than grails.

3. Mids: An arbitrary mid-tier.

4. Floor: These are usually the cheapest in the collection. Interestingly, floor prices are used to monitor the general price performance of an NFT because if the cheapest tokens are priced higher, then the rest of the above tiers would follow as well. Think of it as Lowest Price + some financial premium = NFT Price. This is why the floor price is a widely followed metric to gauge the health of an NFT collection.

So what does the above tell us? The attributes, or traits, of the NFT largely drive price. By mapping out and ranking the trait's price, we can find what traits are valuable, assign a category, then associate the most valuable trait label that a particular NFT has to an NFT.

For instance, here is a graph of the Red Panda trait. We see that NFTs with the red panda trait are sold at a price close to the median price of the Red Panda trait.

The observant reader may notice: "Hey, why are some NFTs sold at a much higher price than the median red panda price?" The answer is because these NFTs probably have a trait that is more valuable. For instance:

We see that although this is a red panda (and we can't see it!), it has a very rare blue bucket headgear (9 in existence). Therefore, it commands a higher premium.

To recap, based on what we have learned:

1. NFT prices are (usually) driven by their most expensive traits.
2. The value of a trait can be proxied by looking at the series of NFTs with that particular trait.

For this analytical project, the median price of all NFTs with a particular trait is calculated and mapped.

Categorizing NFTs by looking at their most valuable traits.

Here is a quick overview of how their categories are determined

Based on the SQL script, here's how the different NFT price tiers are determined through a series of sequential filters and calculations:

Floor Tokens

1. First removes supergrails from the dataset (becase skew)
2. Calculates weights using the formula: sqrt(1 / (trait_adjusted_price / sum(trait_adjusted_price)))
3. Floor tokens are those with weights >= average weights

Mids (Split into Lower and Upper)

Lower Mids:

1. Removes floor tokens
2. Uses similar weight calculation as floor
3. Lower mids are tokens with weights >= average weights

Upper Mids:

1. Removes lower mids tokens
2. Calculates weights using: 1 / (trait_adjusted_price / sum(trait_adjusted_price))
3. Upper mids are tokens with weights >= average weights

Premium (Split into Lower and Upper)

Lower Premium:

1. Removes upper mids tokens
2. Uses same weight formula as upper mids
3. Lower premium tokens are those with weights >= average weights

Upper Premium:

1. Removes lower premium tokens
2. Calculates weights using: (trait_adjusted_price / sum(trait_adjusted_price))²
3. Upper premium tokens are those with weights < average weights

Grails

Determined through multiple methods:

1. Supergrails via 99.9th percentile
2. One-of-one grails (unique tokens)
3. Non-one-of-one grails (based on weighted calculations)
4. Any remaining tokens above upper premium threshold

The final price tier labels are assigned as:

• Floor: <= max_floor price
• Mid: Between max_floor and max_upper_mids
• Premium: Between max_upper_mids and min_grail
• Grail: >= min_grail price

Tracking a Collection's Category Performance

From here on out, the job becomes easier because we've assigned a label to each NFT. We then calculate their price performance over time by creating a baseline of 100 points. For each day, the change from that initial 100 points is added or subtracted to get the performance in absolute terms.

This is similar to how the S&P 500 is calculated.

Here we can see that grails are largely flat in value, but the rest of the other tiers fluctuate wildly (especially floor prices). This suggests that, for traders, it's better to trade floors, while collectors should focus on grails for price stability.

Building the Index

This index tracks the performance of NFTs across different collections by categorizing them into labels (like "floor," "mid," "premium," and "grail") and then calculating a weighted average of their returns.

Monthly Rebalancing:

The index is rebalanced monthly. This means the label assigned to each NFT is held constant for the entire month. The label assigned on the last day of a month is used for the whole following month. This prevents excessive trading and reflects a buy-and-hold strategy typical of many NFT investors.

Market Capitalization Weighting:

The index uses market capitalization weighting. This means the return contribution of each collection is proportional to its market capitalization. A collection with a larger market cap will have a bigger influence on the overall index performance than a collection with a smaller market cap. This reflects the significance of each collection within the broader NFT market.

Index Calculation:

The index is calculated daily using the following steps:

1. Daily Returns: The percentage change in price is calculated for each NFT within its assigned label.

2. Weighted Averaging: These daily returns are then averaged within each collection and label. These average returns are further weighted by the collection's market capitalization proportion, giving more weight to larger collections.

3. Blended Returns: Finally, these weighted average returns are summed across all collections for each label, producing a single, blended daily return for each label.

4. Index Level: The daily returns for each label are used to calculate the label index value from a base value using the compounding effects of returns. This approach is the same across all labels allowing meaningful comparison on returns between different rarity groups or baskets within NFT projects. The overall index represents a combined, weighted picture of the performance of these labeled segments of the NFT market.

Now we are able to see the price performance for each category with across the 6 collections listed above. This acts as a barometer in understanding boom and busts.

This is the market cap of the 6 collections. The relative proportions are used to weight returns