Methodology and Research

Our Methodology


We build on our academic expertise [2,3,4,6,7,8,9,10] in analyzing cryptocurrency markets to aggregate and validate NFT transaction data. Based on this data, we construct the comprehensive set of indices and analytics for the overall NFT market and for its subcomponents. The high-quality data and the aggregate indices are foundational for both understanding NFTs as assets and for developing a future ecosystem of core market applications and services built around NFTs.


Our methodology is developed in detail in [1].


Data


We analyze more than 30 millions transactions and essentially cover most of the universe of the NFT trades. We aggregate the data from all of the major NFT exchanges and also directly tap the different blockchains to cross-validate random subsamples of our data. The raw data is then extensively inspected and refined to assemble a high-quality dataset that is consistent across a variety of data sources and is robust to outliers and contamination. The comprehensiveness and the quality of the dataset is essential in constructing both the NFT market indices and, even more saliently, in constructing high-resolution analysis of the market components such as collections and categories.


In creating this data we build on our prior research in assembling large datasets for cross-sections of tokens in cryptocurrency markets [2,3,7] and in aggregating data from multiple cryptocurrency platforms and exchanges [4,9].


Index methodology


Construction of a comprehensive NFT index is significantly more involved than constructing market indices for either equities or cryptocurrencies. The primary challenges are that individual NFTs are often unique, differ vastly from each other by their characteristics and prices even within a given collection or a class, and are traded infrequently. These features make it challenging to aggregate their prices.


We proceed by decomposing the valuation of each NFT into a common component and an idiosyncratic component. We extract the common component using statistical methods and construct a representative comprehensive index for the NFT market. Furthermore, we apply a similar procedure to the popular NFT collections and construct high-resolution sub-indices for major NFT collections and categories.


Our indices stand in a marked contrast with the indices that are based on cross-sectional moments of the data such as the floor price, median price, and average price. Those result in spurious movements depending on the changing characteristics of the individual trades, making the construction of an overall NFT market index based on this method impossible. Not surprisingly, the idea of constructing price indices based on such cross-sectional sample moments is largely discarded in the academic literature. Our methodology is immune from any composition change in sales.


Valuation (AVM)

We construct an AVM (automated valuation model) model that optimally combines information from the BLT indices and machine learning techniques to generate updated high-precision and reliable valuations of individual NFTs and NFT collections. The AVM is a Multi-Layer Perceptron Regressor model, which is a feed-forward neural network model with multiple hidden layers uses for regression tasks [1]. The AVM is designed to incoporate the information from visual characteristics of NFTs extracted from high-quality images as JPG files using the convolutional neural networks ResNet-18 and AlexNet as classifiers. In [1] we show that incorporating the information from visual characteristics improves the out-of-sample R-squared by 38% (ResNet) and 64% (AlexNet).


Analytics


We develop detailed analysis of the NFT market using the constructed high-quality data and indices. This analysis uncovers both the properties of the indices and its components such as various collections as well as the drivers and the predictors of the market performance. Specifically, we focus on three main areas. First, we develop a variety of more specialized indices that capture various aspects of the NFT market and its components such as the indices of the most expensive NFTs. Second, we construct a number of related indices that provide additional important characterization for the behavior of the NFT market such as the indices of the NFT related cryptocurrencies and indices of NFT attention. Third, we describe a variety of the core statistics for the NFT market (such as the Sharpe ratio) and the statistically valid predictors and driving factors of the behavior of the indices over time and across NFTs.

Our Research

 

1. "The Economics of Non-Fungible Tokens", by Nicola Borri, Yukun Liu and Aleh Tsyvinski, Available at SSRN, 2022 [videos: IMF 10th Statistical Forum |Chainlink Research | CBER Forum | Blockchain@UBC]

2. "Risks and Returns of Cryptocurrencies", by Yukun Liu and Aleh Tsyvinski, Review of Financial Studies, 2021

3. "Common Risk Factors in Cryptocurrency", by Yukun Liu, Aleh Tsyvinski and Xi Wu, Journal of Finance, Forthcoming

4. "The Cross-Section of Cryptocurrency Returns", by Nicola Borri and Kirill Shakhnov, Review of Asset Pricing Studies, Forthcoming

5. "Regulation Spillovers across Cryptocurrency Markets", by Nicola Borri and Kirill Shakhnov, Finance Research Letters, 2020

6. "Conditional Tail-Risk in Cryptocurrency Markets", by Nicola Borri, Journal of Empirical Finance, 2019

7. "Accounting for Cryptocurrency Value", by Yukun Liu, Aleh Tsyvinski and Xi Wu, 2021, invited submission to the Review of Financial Studies

8. "Crypto Premium: Higher-Order Moments and Tail Risk", by Nicola Borri and Paolo Santucci de Magistris, 2021

9. "Cryptomarket Discounts", by Nicola Borri and Kirill Shakhnov, 2020

10. "Do Cryptocurrencies Have Fundamental Value?", by Yukun Liu, Jinfei Sheng and Wanyi Wang, 2019