Understanding NFT Price Moves through Tweets Keywords Analysis

TL;DR

This paper investigates how Twitter community activity and specific keywords influence NFT price movements, demonstrating that social media data can effectively predict NFT price directions and magnitudes.

Contribution

It introduces a novel analysis of Twitter keywords' impact on NFT prices, combining causality tests and predictive modeling to uncover market-behavior insights.

Findings

Number of tweets Granger-causes NFT prices for most top projects.

Social media words significantly improve NFT price move predictions.

Market-related and event-related words positively influence price change predictions.

Abstract

Non-Fungible Token (NFT) is evolving with the rise of the cryptocurrency market and the development of blockchain techniques, which leads to an emerging NFT market that has become prosperous rapidly then followed by a cooldown. Nevertheless, the overall rise procedure of the NFT market has not been well understood. To this end, we consider that social media communities evolving alongside the market growth, are worth exploring and reasoning about, as the mineable information might unveil the market behaviors. We explore the procedure from the perspective of NFT Twitter communities and its impact on the NFT price moves with two experiments. We perform a Granger causality test on the number of tweets and the NFT price time series and find that the number of tweets has a positive impact on (Granger-causes) the price or reversely for larger part of the 19 top authentic projects but seldom copycat projects. Besides, to investigate the price moves predictability, we experiment on predicting Markov normalized NFT price (representing the direction and magnitude of price moves) given tweets-extracted word features and interpret the feature importance to find insights. Our results show that social media words as the predictors result in all 19 top projects having a testing accuracy evidently above the baseline. Based on the feature importance analysis, we find that both general market-related words and NFT event-related words have a markedly positive contribution in predicting price moves. We summarize the characteristics including categorization and sentiment for the words with the top and least feature importance.