[Paper Review] Bubble Prediction of Non-Fungible Tokens (NFTs): An Empirical Investigation

Kensuke Ito, Kyohei Shibano+1 more|arXiv (Cornell University)|Mar 22, 2022

Blockchain Technology Applications and Security9 citations

TL;DR

This study applies the logarithmic periodic power law (LPPL) model to time-series price data of four major NFT projects to predict market bubbles. As of December 20, 2021, it finds that NFTs overall and the Decentraland project are in small to medium bubbles indicating expected price declines, while Ethereum Name Service and ArtBlocks are in small negative bubbles suggesting potential price increases.

ABSTRACT

Our study empirically predicts the bubble of non-fungible tokens (NFTs): transferable and unique digital assets on public blockchains. This topic is important because, despite their strong market growth in 2021, NFTs on a project basis have not been investigated in terms of bubble prediction. Specifically, we applied the logarithmic periodic power law (LPPL) model to time-series price data associated with four major NFT projects. The results indicate that, as of December 20, 2021, (i) NFTs, in general, are in a small bubble (a price decline is predicted), (ii) the Decentraland project is in a medium bubble (a price decline is predicted), and (iii) the Ethereum Name Service and ArtBlocks projects are in a small negative bubble (a price increase is predicted). A future work will involve a prediction refinement considering the heterogeneity of NFTs, comparison with other methods, and the use of more enriched data.

Motivation & Objective

• To empirically investigate bubble formation in non-fungible tokens (NFTs) using time-series price data.
• To assess whether the LPPL model can predict bubble dynamics in NFT markets, particularly at the project level.
• To identify whether specific NFT projects are in positive (upward price pressure) or negative (downward price pressure) bubble states as of late 2021.
• To lay the foundation for future research on heterogeneous NFT characteristics, alternative prediction methods, and enriched data integration.

Proposed method

• The logarithmic periodic power law (LPPL) model is applied to weekly moving-average price data of NFTs from four major projects: Decentraland, CryptoPunks, Ethereum Name Service, and ArtBlocks.
• The LPPL model estimates two bubble indicators: $bubbleindicator(pos)$ for positive bubbles (expected price decline) and $bubbleindicator(neg)$ for negative bubbles (expected price increase).
• The model uses time-series fitting to detect log-periodic patterns indicative of critical transitions in asset prices.
• Data were collected from nonfungible.com, covering price movements from late 2017 to December 20, 2021, with weekly aggregation to reduce volatility.
• The model's predictive accuracy was evaluated by comparing predicted bubble indicators with actual price trends.
• Results were visualized using time-series plots of both bubble indicators to assess predictive performance.

Experimental results

Research questions

• RQ1 Are NFT markets, particularly at the project level, exhibiting signs of speculative bubbles as of late 2021?
• RQ2 Can the LPPL model effectively detect and predict bubble dynamics in NFT price time-series data?
• RQ3 Do different NFT projects show divergent bubble states—such as positive, negative, or no bubble—indicating varying near-term price trajectories?
• RQ4 How does the LPPL model’s performance compare across heterogeneous NFT projects with distinct underlying assets and market behaviors?

Key findings

• As of December 20, 2021, the overall NFT market was in a small bubble, indicating a predicted price decline, with $bubbleindicator(pos) \approx 0.2$.
• The Decentraland project was in a medium bubble, signaling a higher likelihood of price decline, with $bubbleindicator(pos) \approx 0.4$.
• The Ethereum Name Service and ArtBlocks projects were in small negative bubbles, suggesting a predicted price increase, with $bubbleindicator(neg) \approx 0.1$ each.
• The LPPL model successfully captured the direction of price changes in most cases, though it failed to predict the prolonged price surge in CryptoPunks from October 2020 to March 2021.
• The model demonstrated predictive capability for both positive and negative bubble trends, particularly in the latter half of 2021.
• The study is the first empirical investigation to apply the LPPL model specifically to NFT projects, offering a novel framework for bubble detection in digital collectibles.