Correlations versus noise in the NFT market

The non-fungible token (NFT) market emerges as a recent trading innovation leveraging blockchain technology, mirroring the dynamics of the cryptocurrency market. The current study is based on the capitalization changes and transaction volumes across a large number of token collections on the Ethereum platform. In order to deepen the understanding of the market dynamics, the inter-collection dependencies are examined by using the multivariate formalism of detrended correlation coefficient and correlation matrix. It appears that correlation strength is lower here than that observed in previously studied markets. Consequently, the eigenvalue spectra of the correlation matrix more closely follow the Marchenko-Pastur distribution, still, some departures indicating the existence of correlations remain. The comparison of results obtained from the correlation matrix built from the Pearson coefficients and, independently, from the detrended cross-correlation coefficients suggests that the global correlations in the NFT market arise from higher frequency fluctuations. Corresponding minimal spanning trees for capitalization variability exhibit a scale-free character while, for the number of transactions, they are somewhat more decentralized.

💡 Research Summary

The paper provides a systematic quantitative investigation of the nascent non‑fungible token (NFT) market, focusing on the degree to which price and activity across thousands of Ethereum‑based collections are correlated versus dominated by random noise. The authors first assemble a high‑frequency dataset covering daily changes in market capitalization and the number of transactions for a large set of NFT collections over a two‑year period. After log‑differencing to achieve stationarity, they compute two distinct correlation measures: the conventional Pearson correlation, which captures linear, synchronous co‑movements, and the detrended cross‑correlation coefficient (DCCA), which is robust to non‑stationarity and can separate low‑frequency (trend) from high‑frequency (fluctuation) components. Using each measure they construct an N × N correlation matrix (N = number of collections).

Eigenvalue analysis of these matrices is then performed. By comparing the empirical eigenvalue spectra with the Marchenko‑Pastur (MP) distribution that describes the spectrum of a purely random matrix, the authors assess how much genuine structure exists. The bulk of eigenvalues falls within the MP bounds, indicating that most inter‑collection relationships are indistinguishable from noise. Nevertheless, a handful of the largest eigenvalues lie well above the MP upper edge, suggesting the presence of genuine collective modes—likely driven by shared external shocks such as major drops or spikes in overall crypto sentiment, high‑profile drops of popular collections, or platform‑wide events.

A key contribution is the side‑by‑side comparison of Pearson‑based and DCCA‑based matrices. The DCCA spectrum shows markedly stronger correlations at short time scales (high‑frequency fluctuations), whereas long‑term (low‑frequency) correlations remain at the noise level. This pattern implies that NFT price dynamics are dominated by rapid, speculative trading and short‑lived demand spikes, with little evidence of a stable, fundamentals‑driven co‑movement that characterises mature asset classes.

To visualize the network topology implied by the correlation structure, the authors build minimum spanning trees (MSTs) from both the market‑cap and transaction‑count matrices. The MST derived from market‑cap fluctuations exhibits a classic scale‑free architecture: a few “hub” collections (e.g., CryptoPunks, Bored Ape Yacht Club) possess a disproportionately high degree, while the majority of nodes are peripheral. This hub‑spoke pattern reflects a nascent “core‑periphery” organization where large, well‑known NFTs anchor market movements. In contrast, the MST based on transaction counts is more evenly connected, lacking dominant hubs, which indicates that trading activity is more diffusely spread across many smaller collections.

Overall, the study concludes that the NFT market is considerably less correlated than traditional equity, bond, or even major cryptocurrency markets, with most observed relationships attributable to statistical noise. However, the existence of a few significant eigenvalues and the high‑frequency DCCA correlations reveal that non‑trivial, short‑term collective behavior does exist, driven by speculative dynamics and shared external events. The differing MST structures for capitalization versus transaction volume further highlight distinct underlying mechanisms—value concentration versus activity dispersion. The authors suggest that as regulatory frameworks mature, market participants gain better valuation tools, and platform standards evolve, the correlation structure may become stronger and the market more efficient, potentially transitioning from a noise‑dominated regime toward one with clearer, longer‑term inter‑asset linkages.