Using machine learning to trade NFTs

NFT OnChained is a data analytics platform which leverages machine learning to estimate the current fair value of NFTs.

Having fair value estimates has these 3 main use cases:

  • Spotting mispriced items

  • Portfolio tracking

  • Knowing what price to list an NFT for

How does it work?

We use the trait data, sales data and other market data as input for our proprietary machine learning pipeline to estimate the current fair value of NFTs. Rarity alone is not a good predictor of the value of an NFT since certain traits can be very desirable for aesthetic or utility reasons. The model will learn these patterns.

Does it really work tho?

A fair question. We display the estimate error for each collection to indicate the accuracy of the model used.

blog image

For Cyberkongz VX the estimate error currently is 5% e.g.

Calculation of the estimate error: Average difference between price estimates and actual sale prices for the last 100 sales.

Meaning we store the predictions of the model and later compare them to the actual sales prices. For example if the model estimates a price of 1 ETH for a certain NFT but it later sells for 1.05 ETH, the error would be 5%.

Trading Challenge

We wanted to try out our model ourselves. That’s why we started a fresh OpenSea account and bought and sold NFTs that were deemed mispriced or cheap by the model. Here are the results of our challenge which started November 30th 2021.

You can check our OpenSea account here:

https://opensea.io/OnChainedCom?tab=activity

blog image

Example of a mfer trade

Results

We started our challenge with 2 ETH on November 30th 2021. At the time of this writing (February 21st 2022) our balance was 4.9 ETH (4.3 ETH liquid and 0.6 ETH currently invested). We used 0.3 ETH from our balance to pay some bills, that’s why the profit balance below shows 5.2 ETH. You can check all numbers on OpenSea and etherscan — that’s the beauty of blockchain after all.

blog image

Trading Balance over time

blog image

Profit in USD per trade

blog image

Profit per trade in ETH

blog image

Profit in Percent per trade in ETH terms

blog image

Profit in Percent per trade in USD terms

Conclusion

With our trading challenge we demonstrated the ability to successfully trade NFTs with the aid of machine learning.

We also paid attention to other factors when we pulled the trigger for a trade:

  • decent volume

  • price in uptrend

These factors can increase the probability of success of a trade.

Here is the link to our platform in case you want to try it out:

https://nft.onchained.com/

Or contact us on twitter: