*Patrick Kiefer and Michael Nowotny*
Asset managers are adding cryptocurrencies to traditional portfolios, a trend spurred by client demand and the deployments of spot Bitcoin ETFs in January 2024 and spot ETH ETFs six months later. By December 2025, Bank of America private wealth CIO Chris Hyzy recommended an "…allocation of 1% to 4% in digital assets," for clients who have “comfort with elevated volatility.”
To understand a cryptocurrency's contribution to portfolio risk, asset managers need to understand the connection between the cryptocurrency returns and systematic risk in equity markets. One way to quantify this is to measure the relationship between the cryptocurrency return time series and the time series of equity market risk factor replicating portfolio returns. Below, we conduct time-series regressions of crypto returns on the Fama-French 5-factor portfolio returns. Over a daily sample beginning in 2021, returns on major cryptocurrencies load positively on the market factor and negatively on the Fama-French (FF) profitability (RMW) factor. None of the other factors (size, value, CMA) are significant.
The connection is interesting because while the RMW-replicating portfolio is comprised of stocks formed based on public profitability and payout data, cryptocurrencies aren’t typically valued based on cash flows. Nonetheless, and perhaps unsurprisingly, cryptocurrencies trade like early-stage tech stocks, which also tend to have high market betas and negative loadings on profitability. This relationship is useful to asset managers who are already familiar with risk and return behavior of early stage tech, and who already work with the FF 5-factor model to understand systematic risk in equity markets. We address some possible interpretations of this connection after describing the data and findings.
We select the 13 largest cryptocurrencies by market capitalization excluding stable coins that have continuous Binance trading history from January 4, 2021 to October 31, 2025. We use daily factor returns for the Fama-French 5 factor asset pricing model, available here.
Daily cryptocurrency returns are constructed from Binance trade data featuring an aggressor flag as well as an indicator of whether the trade took place at the top of the order book or whether it walked the book. The aggressor flag identifies which party initiated each trade (buyer or seller), allowing us to sign order flow. The book-walking indicator captures trades that consume multiple price levels, which we use to compute execution-weighted prices. Together, these allow us to estimate the bid-ask spread’s components and recover unobserved mid-prices that are free of bid-ask bounce. Specifically, we perform price impact regressions following the specification and procedure developed in Glosten and Harris (1988) based on the theory in Glosten and Milgrom (1985). The resulting midprices are used to compute the time series of cryptocurrency returns.
To handle incongruencies in trading times, for each trading day, we compute cryptocurrency returns from 4:00 PM EST on the prior trading day to 4:00 PM EST on the current trading day, matching the timestamp boundaries used in CRSP close-to-close equity returns from which Fama-French factors are constructed. This means Friday-to-Monday cryptocurrency returns (spanning approximately 72 hours of continuous trading) are paired with the corresponding Friday-to-Monday factor returns, ensuring both series reflect the same calendar interval despite differences in market operating hours. We discuss this construction further in the data appendix.
We run time series regressions of returns of 13 major cryptocurrencies onto Fama-French equity market factor returns. Coefficient estimates are reported. p-values are computed using Newey-West for heteroskedasticity and autocorrelation corrected covariance estimation. We use a maximum lag parameter of 6 days for the autocorrelation.
Over the full sample, the market factor loadings are positive and significant for every cryptocurrency and the RMW factor loadings are negative and significant for all but SOL and XRP. The results are significant over multiple subsamples. Restricting attention to the previous two years, the SOL RMW loading becomes significant, possibly due to overwhelming idiosyncratic price action in SOL stemming from the FTX scandal and subsequent crash in the earlier part of the sample. The findings are robust to multiple approaches for covariance estimation and trading time alignment computations.
We estimate the full five-factor model; coefficients on SMB, HML, and CMA were statistically indistinguishable from zero for all cryptocurrencies and are omitted for brevity. Full results available upon request.
| Crypto | R² | Mkt-RF (β) | (p-value) | RMW (β) | (p-value) | α | (p-value) |
|---|---|---|---|---|---|---|---|
| BTC | 18.6% | 1.05** | 0.000 | -0.95** | 0.000 | 0.0013 | 0.165 |
| ETH | 18.2% | 1.48** | 0.000 | -0.93** | 0.001 | 0.0017 | 0.175 |
| ADA | 14.0% | 1.49** | 0.000 | -1.11** | 0.001 | 0.0016 | 0.315 |
| LINK | 13.3% | 1.66** | 0.000 | -0.85* | 0.018 | 8.00e-04 | 0.617 |
| AVAX | 12.2% | 1.75** | 0.000 | -1.28** | 0.000 | 0.0033 | 0.157 |
| SOL | 11.9% | 1.81** | 0.000 | -0.82 | 0.070 | 0.0046* | 0.021 |
| LTC | 11.0% | 1.30** | 0.000 | -0.88** | 0.003 | 3.00e-04 | 0.820 |
| BCH | 9.9% | 1.32** | 0.000 | -0.99** | 0.002 | 0.0010 | 0.521 |
| ZEC | 9.5% | 1.47** | 0.000 | -1.78** | 0.000 | 0.0025 | 0.205 |
| BNB | 9.3% | 1.04** | 0.000 | -0.87** | 0.002 | 0.0034* | 0.034 |
| XLM | 8.8% | 1.31** | 0.000 | -0.87** | 0.004 | 0.0014 | 0.468 |
| XRP | 7.7% | 1.33** | 0.000 | -0.68 | 0.081 | 0.0028 | 0.122 |
| TRX | 5.0% | 0.75** | 0.000 | -0.67* | 0.037 | 0.0026* | 0.047 |
Note: * indicates statistical significance at p ≤ 0.05, ** indicates significance at p ≤ 0.01
The regression shows highly significant positive market betas and highly significant negative loadings on RMW. The RMW factor contributes meaningfully to explanatory power beyond the market factor alone, with t-statistics on RMW loadings ranging from 2.1 to 4.2 across cryptocurrencies with significant coefficients.
The negative RMW loading indicates that cryptocurrencies perform well precisely when unprofitable, speculative equities outperform profitable ones—i.e., during ‘risk-on’ sentiment regimes. This co-movement suggests a common speculation or sentiment factor drives both asset classes, even though cryptocurrencies lack traditional cash-flow fundamentals. One interesting potential economic explanation for the connection is that cryptocurrencies implicitly contain claims to cash flows in the distant future. Early stage tech stocks with this property are explored in Pastor, L., and Veronesi, P. (2009).