The NFT market has been growing rapidly, with new creative digital arts and skyrocketing prices. Many people are trying to find the next Crypto Punks, Bored Ape Yacht Club or Azuki before the price becomes too high. Although it is difficult to predict the price of any NFT due to the variety of factors involved, it is possible to predict the minimum price in the near future using time-series statistical models.

1. Overview

There are generally two types of price prediction that buyers are interested in:

1. Collection Floor Price Prediction: To predict the lowest price of the entire collection on any given day – the floor price prediction.
2. Token Price Prediction: To predict the price of a specific token within one NFT collection. i.e. within the BAYC collection, based on the different traits and features and historic sale price of the specific token, predict the price of the token on any given day.

The following model is only for predicting the floor price of the collection. The token price prediction will be explained in the next article.

To download data from Covalent, you need to load the required packages and define the functions needed. You also need to request an authentication key from the site. There are other NFT data providers such as Flipside Crypto, Dune, etc. that you can use to download historic NFT prices and sale volumes.

2. Data Preparation

You can download historic NFT data from Covalent API by using the NFT contract address. For example, here is the address for Bored Ape Yacht Club (BAYC).

3. Floor Price Prediction Models

3.1 Model Options

The dependent variable to predict is the floor price of a given NFT collection. Since the past sale price is usually a good indicator of the future price, this can be interpreted as the prediction of a time-dependent event:

Y(c|tn) = f(w,x,Y(c|t0,...,tn-1))

The price of collection c at time tn is represented by Y and is influenced by time dependent independent variables x. The prices from t0 to tn-1 represent the past prices that can be used to predict the price now.

We know the prices of a lot NFTs have sky-rocketed since last year and there is surely a trend in the price. Given most of the time-series models require stationarity (no trend) in the data, generally Y needs to be transformed into a difference in order to remove trending in the price:

dY(c|t) = f(w,x,dY(c|t0,...,tn-1))

The percentage change in the price of a collection (or token) from one month to the next is represented by dY.

Some of the most commonly used time-series models are:

• Autoregressive model
• ARMA
• ARIMA

In the following sections, we will use an ARIMA model to predict the floor price. This model incorporates the differencing term dY(c|t), the lag of the difference dY(c|t-1), and the lag of the error terms e(c|t).

3.2 ARIMA

ARIMA models are generally the most general class of models for forecasting a time series which can be made to be “stationary” by differencing (see more details here). The model form ARIMA(p,d,q) comes with three components:

• p is the number of autoregressive terms,
• d is the number of nonseasonal differences needed for stationarity, and
• q is the number of lagged forecast errors in the prediction equation.

3.2.1 Stationary

You can use the autocorrelation function ACF and partial autocorrelation function PACF to help you decide what values of p, d and q to use in the ARIMA model.

The graphs below demonstrate that the initial daily floor price time-series is trending (non-stationary). However, once the first difference is used, the series becomes stationary as is evidenced by the ACF and PACF plots. The second difference is over-differencing as is suggested by the over-shooting lag from ACF and PACF.

The pricing revolution that NFTs need may not be a complete reversal of what has been done before, but a steady stream of imaginative proposals to improve the efficiency of price discovery. In terms of NFTs and DeFi collaboration, fragmentation is still a direction worth exploring.

The benefits of fragmentation are democratization, liquidity, and price discovery.

Some people argue that most non-fungible tokens (NFTs) are owned by individuals and only provide benefits to those individuals. This goes against the idea of decentralization that is proposed by blockchain technology. Therefore, there is a need for fragmented NFTs that are designed specifically for decentralized ownership and use.

Some artists who have embraced NFT have not been able to sell their products, while many NFT-based projects have closed down On the one hand, some people argue that the lack of liquidity in the NFT market is due to the fact that NFTs are not easily interchangeable. This is because the value of an NFT is determined by its uniqueness, which makes it difficult to find a buyer willing to pay the same price for it. On the other hand, others argue that the high prices of NFTs are preventing them from being sold, as people are not willing to pay such high prices for something that does not have a tangible form. However, many artists who have embraced NFTs have been successful in selling their products, and many NFT-based projects are still up and running.

We believe that NFTs need a pricing revolution in order to improve their usefulness and encourage further development. The current system of price discovery is seen as a limitation by many in the industry, and is preventing NFTs from reaching their full potential.

How to Comprehend “Price”

Let’s re-understand the concept of price in NFT markets from two perspectives. These two concepts answer two of the most frequently asked questions about NFT:

1. Why can a JPEG be sold for thousands of dollars or even millions of dollars?
2. What does this JPEG sold for thousands of dollars or millions of dollars mean?

• Prices are determined by the value judgments of consumers

Let’s review the seven pricing principles of “The Psychological of Price”:

1. Prices should be based on consumers’ perceived value of the product, not what you think it costs.
2. Prices should be realistic, so consumers understand what they are getting for their money.
3. Prices should be comparable within your control.
4. If you want to adjust the price, you have to change the structure of the service or product.
5. Price differentials are essential contributors to profitability.
6. Pricing communication influences how consumers perceive the value of goods.
7. If you want to increase profits, you must be prepared to sacrifice some sales.

There are two main types of value for goods in economic literature: functional value and emotional value. Functional value is “What can I do with this thing?” and emotional value is “How much I love this stuff.”

The two principles state that the main factor that determines price is the value that consumers place on an item. In order to understand the pricing process, we need to go back to the basics of trade-offs. This means that we need to understand what users are willing to trade-off in order to get the item.

Most NFT users are more concerned with thefunctional value of their assets rather than the sentimental value. This is because in a speculative market, there are very few people who are willing to pay for art. The trade-offs for most NFT users are the investment benefits and social capital.

Why can a JPEG be sold for such high prices?

The value of an NFT is determined by the users, based on how much they are willing to trade-off.

• Information can only be realized through price

The first part of the sentence states that price is how knowledge (economic information) is spread and the information system of economic production. This means that in order for knowledge to be spread effectively, it must be through price. Any attempt to control prices would be disastrous because it would interfere with the flow of knowledge.

The social and economic problem is one of making sure that people can use what they know to improve their lives.

Ludwig Von Mises and Friedrich August Hayek, the representatives of the Austrian school of economics, regard price as the primary means of achieving dispersed knowledge in the market. Hayek emphasized price as a signal of free-market coordination, stating that information can only be realized through price. For example, if one notices that bread is going up in price, it can be inferred that more people are involved in the production of bread.

The price of NFTs, which is irrational, is a result of the market’s information system that produces economic activity.

This JPEG being sold for thousands or millions of dollars is a sign of the market’s consensus on the value of NFTs as well as a signal of coordinated market activity.

The NFT Dilemma: Liquidity and Financialization

Why are NFTs having trouble finding liquidity? It is mostly because it is hard to determine how much an NFT is worth.

First, NFTs are indivisible and there are few transactions. This lack of data makes it difficult to assess their value, resulting in low capital efficiency. Secondly, the current NFT market is highly speculative. This means that even NFTs with comparable value can have very different prices, for example land in Decentraland that is being sold speculatively.

The trend is clear that NFT is becoming a new asset class with a large market.

In his article, “All Digital Content Is Going on-chain”, the founder of CoinFund argues that we should think of NFT as “liquid IP”. He argues that as a result of this, trillions of digital contents will move to the secondary market as intellectual property is transferred to the blockchain as NFT. This will unlock huge illiquid value and become the most significant asset category in the blockchain area.

However, NFT also shares some similarities with low-velocity assets in the traditional world, such as real estate and antiques. Just like those assets, NFT can be valued in two ways: trading and valuation. The main problems with NFT are that there are fewer transactions and less data, which makes it more difficult to value. Additionally, NFTs often come with potential costs that must be paid for by seeking out third-party estimates, which can be opaque.

We can’t use NFTs as collateral for a Compound loan because their value is too hard to determine.

If NFT does not have a way to effectively value its assets, it will be difficult for those assets to be included in the financialization process, which could be a hindrance to the development of NFT.

Pricing Attempt

These liquidity protocols provide a way to determine the value of NFTs for applications that require it.

At present, there are three solutions:

The price set by the buyer is the first solution.

One way to provide liquidity for an NFT is to use seller pricing. For example, NiftEx/NFT20/NFTX will generate an ERC-20 token to be used for pricing when the NFT is first created.

The third solution is called third-party pricing, which is where a company generates its prices based on the predictive behavior of random users.

While NFTs may be low-velocity assets, it is still difficult to price them correctly due to the dynamic nature of price discovery in the futures market. Open competition between supply and demand drives the price of the commodity to the world price level.

In other words, price discovery is finding out what the prices are for items in the world. It is also one of the key functions that NFT markets serve. By understanding how prices are discovered in these markets, one can better understand how NFT prices are set. This, in turn, can help to improve one’s thinking on the subject.

At present, there are NFT price discovery methods:

1. Sale at a fixed price

There is an initial price set when each NFT is put up for sale. For example, the NFT known as Bored Ape Yacht Club has a low initial price, set at 0.08 ETH. However, it’s worth noting that if not many people are interested in buying the NFT, and there is a weak community background with little information on pricing (such as reference to historical data or endorsement from key opinion leaders), then the NFT will have poor liquidity.

2. Blind box

Also, a kind of sales mechanism was buyers snapping them up at fixed prices.

3. Auction

The process of bidding on items is part of how commodity prices are determined.

An example of this would be PartyBid, which is a product that allows people to bid on NFTs collectively. With PartyBid, anyone can create or join a group, pool funds together, place bids on auctions, and have fun. Once a specific NFT auction is won, the NFT will be deposited in a vault that is jointly owned by DAO holders. DAO holders jointly own the assets.

For example, the cross-chain DeFi platform Alpha Finance Lab launched the NFT floor bid auction platform Alpha Buy Wall. This platform conducts auctions in the form of call auctions, which can be sold in real-time without slippage. The only fees charged are Gas fees. Buyers in this platform are not bidding for a certain NFT, but the lowest purchase price (floor price) of a certain NFT series. When the seller believes that the current bid has reached their expected price, they can sell at the highest bid. In addition, the platform allows sellers to set the minimum bid accepted, preventing the cancellation of the maximum bid when the transaction is completed.

4. Multi-form auction

Beeple’s work was auctioned off on the well-known British auction platform Christie’s for a record price of US$69.34 million. This once again illustrates the fact that art is more subjective and therefore more suited for auction formats. Specifically, there are the following types of auctions:

• Dutch Auction

After accepting the bid, the auction ends immediately. The company Algorand used the Dutch Auction method to raise 60 million dollars in 4 hours, making Dutch Auctions widely known. In the NFT market, Dutch Auction is also a standard method. The specific auction mechanism is as follows: In the bidding process, the bidding price of the auction object is increasing according to the bidding ladder from low to high, and the highest bidder becomes the winner of the bidding, and the highest bidder becomes the winner of the bidding. The seller (the creator of this auction) can accept any bid at any time, and there is no time limit. After accepting the bid, the auction ends immediately.

Bounce. finance will be launched as a decentralized auction tool to support the Dutch auction function of NFT auctions.

• Radical auction

This system incentivizes bidders to help the price of a lot reach the reasonable pricing of the market quickly. The price of the lot increases steadily according to the joint curve model set by the system, and the bidders who help discover the price can get income compensation corresponding to the risk.

This means that, with TopBidder, combined with the auction protocol of the radical market theory, you can produce and issue crypto assets that are not permanently occupied. As long as your bid is not less than 10% of the bid of the person above you, you can obtain ownership of the asset.

• Simultaneous multi-round auction

The Nobel Prize in Economics was awarded to Paul Milgrom and Robert Wilson for their improvements to auction theory and inventions of new auction formats. The format they proposed is called a simultaneous multi-round auction. In this type of auction, all items up for sale are available at the same time. The auction starts with a low price, and bidders can make repeated bids. This format reduces uncertainty for potential buyers. Milgrom was hired as an Algorand consultant in September of this year.

Bidders in traditional Dutch and English auctions may attempt to conceal their true intentions to avoid the winner’s curse (the winning bidder overvalues the auction item and its income is lower than normal or even negative). SMRA forces bidders to be active from the beginning of the auction. The comprehensive information and timely disclosure strategy not only avoids the winner’s curse but also provides sellers with an ideal price to the greatest extent. In the same period, bidders with diversified needs (including bidders from multiple NFTs) bid for multiple NFTs simultaneously. This result is also more meaningful for the price discovery of non-standard assets.

PawnHouse’s NFT mortgage lending platform has seen a transaction volume of over 100 million dollars in a month. By using more open and timely information, it encourages and strengthens bidding behavior, thus creating a “price corridor” for NFT assets through a real and dense quotation range.

5. The Oracle mechanism

The ChainLink network provides essential data and logic connectivity for decentralized applications (DeFi), enabling them to securely interact with off-chain resources like data providers, smart contracts, and APIs. Without products like ChainLink, it would be difficult to improve infrastructure for decentralized applications, such as non-fungible token (NFT) index products.

Upshot One, launched by Upshot, uses a design mechanism called “Peer Prediction” to motivate people to answer questions honestly. The mechanism uses “mutual information” (a generalization of relevance) to evaluate quality, with more honest answers receiving more mutual information. In these environments, participants are compensated for answering questions, with the amount of mutual information they provide determining the level of the reward. This creates a clear motivation to answer questions honestly. Upshot’s Peer Prediction is the first on-chain mechanism widely applicable to NFT price discovery.

Upshot’s machine learning model estimated the value of the Sotheby’s NFT collection to be 25.3 million USD, which is better than Sotheby’s own estimation of 12-18 million USD. Upshot’s evaluation system is now being introduced to the fragmented NFT trading platform Unicly through a partnership between the two companies.