As governments worldwide grapple with how to regulate digital assets, a fundamental question remains surprisingly unanswered: what did investors in non-fungible tokens (NFTs) actually earn? The NFT market exploded in 2021, with record-setting sales at Christie’s and Sotheby’s, celebrity endorsements, and breathless media coverage. By late 2022, prices had collapsed. But between the hype and the hangover, the basic economics of NFT investment have been obscured by a statistical illusion that plagues all infrequently traded asset markets.

The returns you see are not the returns you get

Researchers studying illiquid asset markets – housing, art, venture capital, private equity – have long recognised that observed transaction prices can be misleading. We only see prices when a sale occurs, and sales are not random events. Borri et al. (2022) document the broad economics of the NFT market and highlight its extreme return characteristics. Liu et al. (2022) show that cryptocurrency-specific factors like momentum and investor attention strongly predict performance (see also Tsyvinski and Liu 2018). But an even more basic problem confronts anyone trying to measure NFT investment performance: the returns we observe are conditioned on a sale having occurred, and sellers systematically avoid realising losses.

This disposition effect – the well-known tendency of investors to hold losers and sell winners, first documented by Shefrin and Statman (1985) and Odean (1998) – has particularly severe consequences in markets where assets are unique and trading is infrequent. Korteweg et al. (2016) showed that correcting for this selection bias dramatically changes the apparent returns to investing in fine art. In our paper (Goetzmann et al. 2026), we apply and extend this approach to NFTs, using blockchain data that records every transaction with unusual completeness.

A real-time laboratory for bubble economics

We study two major NFT platforms. SuperRare, a curated marketplace, provides data on 96,688 unique digital artworks created between April 2018 and December 2024. OpenSea, the largest uncurated marketplace, gives us over 9 million sales of art NFTs. The blockchain records allow us to track every creation, bid, listing, sale, and transfer – a level of data completeness that researchers studying traditional art or real estate markets can only dream of.

The raw numbers are staggering. On SuperRare, the median realised return on resold NFTs was 170% in dollar terms. Our repeat-sales price index – the standard methodology used since Bailey et al. (1963) to measure returns in markets for unique assets – peaked in October 2021 at roughly 490 times its initial level. That would make NFTs one of the greatest investment opportunities in history.

But there is a catch. Only 6.2% of NFTs on SuperRare ever resold. The 170% median return is computed only for the small fraction of assets that traded twice – precisely the ones whose owners had good reason to sell. This is the selection bias problem. When we look at asking prices, the pattern is stark: during the boom, the average seller listed their NFT at six times what they paid. Sellers were anchoring to large gains and refusing to sell at a loss.

Correcting the illusion

Using a Bayesian Markov chain Monte Carlo methodology adapted from the art market literature (Korteweg et al. 2016, Huang and Goetzmann 2024), we estimate a selection-corrected price index that accounts for the fact that losing investments are systematically less likely to appear in the transaction record.

The corrected picture is dramatically different. The selection-corrected index peaked at roughly one-tenth the level of the unadjusted index and peaked earlier, in March 2021 rather than October. The disposition effect did not just inflate apparent returns; it delayed the apparent timing of the crash by seven months. Researchers studying housing markets have documented similar phenomena: Goetzmann and Peng (2006) and DeFusco et al. (2022) showed that irrational selling behaviour due to reference dependence or base-rate neglect can mask the true timing of booms and busts in repeat-sales indexes, a dynamic that likely contributed to the delayed recognition of the US housing bubble (Sockin et al. 2020).

Even after the correction, the NFT price rise was extraordinary. The corrected index still implies a peak increase of roughly 60 times – which, if accurate, would make the NFT bubble one of the largest in recorded financial history, likely surpassing even the South Sea Bubble of 1720 (Frehen et al. 2020).

Who made money?

We identify 199 active intermediaries on SuperRare – agents who purchased more than 30 NFTs. Their mean return was 94%, but the median was negative 85% when unsold inventory is valued at zero. Nearly two-thirds lost money. The mean was pulled up by a handful of spectacular successes – the maximum return was 73 times the initial investment.

To understand what a disciplined investment strategy could have achieved, we simulate a ‘liquid artist’ strategy: each month, buy NFTs from the most actively traded creators, priced below $10,000. This strategy generated an impressive 14% monthly return. But even this best-case analysis reveals the fragility of NFT profits. Dropping just the top 0.6% of trades – 75 out of 13,019 purchases – reduces the strategy’s return to zero. The entire profitability of NFT investing rested on a tiny number of extreme winners.

Diversification, the standard remedy for idiosyncratic risk, offers limited comfort. Our simulations show that an investor needs at least 400 NFTs in their portfolio to have a 90% probability of earning a positive return. The vast majority of active investors held far fewer. NFT investing, in practice, was less like asset allocation and more like playing the lottery.

Wash trading and false liquidity

On OpenSea, we document another concern for regulators: wash trading, where sellers transact with themselves to create the appearance of market activity. About 5% of transactions show signs of wash trading, and including wash-traded NFTs in a liquidity-oriented back-testing portfolio doubles the portfolio size but reduces returns – from 5.04% to 4.70% per month – because the inflated prices attract genuine buyers who subsequently lose money. This finding is consistent with broader concerns about market manipulation on blockchain platforms raised by Cong et al. (2023), Li et al. (2025), Wang et al. (2025), and Oh (2025), and echoes discussions of how speculative dynamics can be amplified by credit and information distortions (Schularick et al. 2015).

What NFTs tell us about speculative markets

The NFT market provides an unprecedented real-time laboratory for bubble economics. The blockchain records everything – every creation, every listing, every bid, every sale, every refusal to sell. Our findings carry lessons beyond digital art.

First, selection bias is not an academic curiosity. It can make a collapsing market look like it is still booming. Regulators and investors who rely on transaction-based indexes in any illiquid market – housing, private equity, fine art, collectibles – should be aware that the disposition effect systematically inflates apparent returns and delays the recognition of downturns.

Second, extreme tail dependence – where all profits come from less than 1% of trades – is a warning sign for any market dominated by unique, illiquid assets. When the distribution of returns is this skewed, standard portfolio theory provides little guidance and most participants will lose.

Third, the blockchain’s transparency offers a model for market oversight. The ability to identify wash trading, track the disposition effect in real time, and decompose profits across market participants is something traditional markets lack. As policymakers consider regulatory frameworks for digital assets – including the recent US stablecoin legislation (Cecchetti and Schoenholtz 2025) – the analytical tools demonstrated by the NFT experience deserve attention.

Successful NFT investing during the bubble required an almost perfect confluence of timing, liquidity, and luck. As our data confirm, very few achieved it.

References

Bailey, M J, R F Muth, and H O Nourse (1963), “A regression method for real estate price index construction”, Journal of the American Statistical Association 58(304): 933–42.

Borri, N, Y Liu, and A Tsyvinski (2022), “The economics of non-fungible tokens”, working paper.

Cecchetti, S, and K L Schoenholtz (2025), “Crypto, tokenisation, and the future of payments”, VoxEU.org, 22 August.

Cong, L W, X Li, K Tang, and Y Yang (2023), “Crypto wash trading”, Management Science 69(11): 6427–54.

DeFusco, A A, C G Nathanson, and E Zwick (2022), “Speculative dynamics of prices and volume”, Journal of Financial Economics 146(1): 205–29.

Frehen, R, E Jerphanion, and F Braggion (2020), “Credit and trading behaviour: New evidence from the South Sea Bubble”, VoxEU.org, 19 April.

Goetzmann, W N, and L Peng (2006), “Estimating house price indexes in the presence of seller reservation prices”, Review of Economics and Statistics 88(1): 100–112.

Goetzmann, W N, D Huang, and M Nozari (2026), “Non-fungible tokens as investment”, NBER Working Paper 34837.

Huang, D, and W N Goetzmann (2024), “Selection-neglect in the NFT bubble”, working paper.

Korteweg, A, R Kräussl, and P Verwijmeren (2016), “Does it pay to invest in art? A selection-corrected returns perspective”, Review of Financial Studies 29(4): 1007–38.

Li, T, D Shin, and B Wang (2025), “Cryptocurrency pump-and-dump schemes”, Journal of Financial and Quantitative Analysis 60(8): 3622–59.

Liu, Y, A Tsyvinski, and X Wu (2022), “Common risk factors in cryptocurrency”, Journal of Finance 77(2): 1133–77.

Odean, T (1998), “Are investors reluctant to realize their losses?”, Journal of Finance 53(5): 1775–98.

Oh, S (2025), “Market manipulation on blockchain”, working paper.

Schularick, M, A M Taylor, and O Jordà (2015), “Leveraged bubbles”, VoxEU.org, 1 September.

Shefrin, H, and M Statman (1985), “The disposition to sell winners too early and ride losers too long: Theory and evidence”, Journal of Finance 40(3): 777–90.

Sockin, M, W Xiong, and Z Gao (2020), “Housing speculation and its economic consequences”, VoxEU.org, 26 January.

Tsyvinski, A, and Y Liu (2018), “Risks and returns of cryptocurrencies”, VoxEU.org, 6 September.

Wang, Y, M Cheng, and S Zhang (2025), “Wash trading and insider sales in NFT markets”, working paper.