Algorithmic pricing
Last updated
V1
Last updated
Algorithmic pricing models (like Machine Learning based ones) ingest historical sales data and NFT metadata to construct features based on this information to generate accurate, reliable pricings. They validate the predictions by examining their accuracy on data not used in the training process and obtain error bounds by comparing the predictions to realized sale prices. Both the predicted pricings and error bounds provide critical information to Unlockd to know the price of the NFTs being deposited as collateral, thus providing the basis for calculating the Loan-To-Value.
Machine Learning allows us to incorporate data that simpler models do not take into consideration, such as pooling the sales histories of NFTs to arrive at a prediction for a single NFT and utilizing a range of NFT metadata.
As a result, the ML approach excels in predicting the prices of NFTs where others fall short.
Unlockd works with partners that have developed and trained these algorithmic models, whose much of their research effort has focused on constructing different predictor variables, using automated methods to uncover the most important ones, and iterating to arrive at lean but powerful models.
Upshot is a platform that aims to make NFT financial markets more accessible through machine learning and new types of DeFi x NFT primitives. It is the leading NFT appraisal platform, providing accurate, near-real-time appraisals for 57m+ NFTs from over 32,000 collections. By delivering accurate appraisals at unparalleled scale through machine learning, Upshot enables the creation of powerful solutions at the intersection of DeFi and NFTs.
The steps that go into their appraisal models can be summarized as follows:
They index large amounts of on-chain and off-chain data such as mints, sales, bids, asks, and asset metadata. They then enhance this data in some ways by creating useful metatraits, constructing graphs to analyze holder activity, etc.
They extract features that are predictive of future prices based on rarity, trait value, recent sale prices and combine into a single data frame, which they split into 3 temporally contiguous pieces: a test set, a validation set, and a training set.
They train their models using median-standardised prices as the target and use the trained models to refresh appraisals every hour. They then compare historical NFT sales to historical appraisal prices to calculate the Median Absolute Percentage Error % for each collection (e.g. the appraisal is accurate within +/- MAPE%). Their MAPE is industry-leading at 3-10% for our most accurately appraised collections.
Learn more about Upshot in their
Upshot is the main NFT pricing data oracle that the Unlockd protocol uses to appraise NFT prices. Our model also relies on other partners as backup oracles:
NFTBank Machine Learning models analyze NFT metadata traits and sales history. As the model matured, they have developed methodologies to most accurately predict individual assets such as applying different weighting to trait values, categories, groupings of trait values, time of sales and more.
NFT Bank is continuously developing their Artificial Intelligence model to catch edge cases such as wash trading. Still, their model may work better with certain collections, especially with those with rich past sales data and low volatility.
Nabu's advanced machine learning algorithms combine on-chain data, transaction histories, traits, and rarity to provide state-of-the-art price predictions.
Learn more about NFT Bank in their .
Nabu allows Unlockd to algorithmically each individual NFT according to different models, input variables, and Machine Learning techniques.