Pricing crypto assets when the sample itself is biased

Cryptocurrency markets are large, volatile, and — by now — extensively studied. Much of the empirical evidence, however, rests on samples that condition on survival: coins that failed, were delisted, or traded only briefly are silently dropped. The resulting portfolios overstate cross-sectional return predictability and obscure which anomalies are real. This is not a minor data-handling concern. In a universe where a substantial fraction of listed assets eventually disappears, survivorship conditioning is first-order for inference.

This project does two things. First, it builds unbiased infrastructure — data sources and R packages that make the full, delisting-inclusive cross-section of cryptocurrencies available to researchers. Second, it uses that infrastructure to re-examine central crypto-finance findings (size, momentum, herding) and to measure how crypto risk propagates into traditional banking.

Bias-corrected cross-sectional evidence

Survivorship and Delisting Bias in Cryptocurrency Markets documents annualised survivorship bias of 0.93% for value-weighted and 62.19% for equal-weighted buy-and-hold portfolios in a 3,904-asset cryptocurrency universe from 2014 to 2021. Once delisting returns are accounted for, a size effect is confirmed but substantially overstated in survival-conditioned samples, while momentum and market beta no longer price the cross-section. A non-trivial share of the published crypto anomaly evidence is an artefact of data truncation rather than genuine return predictability — a methodological wake-up call for the empirical crypto asset-pricing literature.

Behaviour and market structure

Cryptocurrencies: Herding and the Transfer Currency examines herding behaviour on the full, survivorship-bias-free cross-section of coins. Against prior evidence — which the paper attributes to sample bias — it documents statistically significant herding, reinforced under a beta-herding robustness check. It also introduces Bitcoin as a transfer currency: herding measures centred on Bitcoin rather than on a value-weighted market portfolio better capture the dispersion of investor beliefs across the crypto cross-section.

Contagion to traditional finance

Estimating Crypto-Related Risk uses the November 2022 failure of FTX as a natural experiment to measure crypto-related risk in U.S. banks. A market-based sensitivity measure — the historical covariance between bank stock returns and bitcoin returns — explains the cross-section of returns on 219 U.S.-listed financial institutions on the FTX announcement day. The measure is unrelated to standard proxies for operational risk (corporate governance, business complexity) but is significantly related to the Tier 1 capital adequacy ratio: on average, it is banks with sufficient liquidity reserves that venture into the crypto sphere.

Infrastructure: the crypto2 R package

The crypto2 R package provides survivorship-bias-free access to cryptocurrency market-cap data from coinmarketcap.com, including coins that have since been delisted. It is hosted on CRAN with documentation at sstoeckl.github.io/crypto2. The package underpins the bias-correction work in the Liebi et al. paper and is used by outside researchers running survivorship-aware crypto studies.

Ongoing work

Cross-coin diversification effects, crypto-related risk in non-U.S. banks, and extensions of the FTX contagion methodology to other large failures (Terra-Luna, 3AC).

Collaborators

Lars Kaiser (formerly University of Liechtenstein), Manuel Ammann, Luca Liebi, Tom Burdorf (University of St.Gallen), Lukas Müller, Johanna Müller, Dirk Schiereck (TU Darmstadt).