Results

How much do specification choices matter for crypto factors?

Every published crypto factor rests on many small choices: the data source, filters, weighting, the number of portfolios, the week, and where breakpoints come from. We built the factors under every combination of the ten design choices of Fieberg et al. (2024), using two of their four data sources (CoinGecko and CoinMarketCap). We added three breakpoint universes and eight week calendars, for 248,832 designs in total. This page shows how much the answers depend on those choices. Unless noted, the numbers are annualised Sharpe ratios from 2018 onward. The Standard explains which design we recommend.

The size premium is a breakpoint artefact

CSIZE (small minus big), 2018 onward.
Breakpoints Weights Designs Mean Sharpe Significantly positive Significantly negative
All coins Value 27648 -0.03 22% 42%
Top 100 Value 27648 -0.92 0% 79%
≥ $100M Value 27648 -1.04 0% 87%
All coins Capped value 27648 0.56 34% 13%
Top 100 Capped value 27648 -0.76 0% 58%
≥ $100M Capped value 27648 -0.74 0% 52%
All coins Equal 27648 0.92 39% 9%
Top 100 Equal 27648 -0.27 14% 35%
≥ $100M Equal 27648 -0.22 16% 33%

The size premium in CSIZE (small minus big), 2018 onward, is not robust. Its sign and significance depend heavily on how breakpoints and weights are chosen, and the premium is positive only when the small leg is allowed to include microcaps.

Positive mean Sharpes appear only under the All coins breakpoints, and only with capped-value or equal weighting. Those “small” coins are worth a few hundred thousand dollars, so they are not realistically investable. Once breakpoints are restricted to more investable universes — Top 100 or ≥$100M — the mean Sharpe turns negative in every weighting scheme, and significantly negative designs outnumber significantly positive ones by a wide margin. Under value weighting the contrast is sharpest: the All coins row is the only one that is not clearly negative, while the Top 100 and ≥$100M rows are dominated by significantly negative results.

In summary, a positive size premium emerges only when “small” includes microcaps worth a few hundred thousand dollars, defined by all-coin breakpoints. When small is instead defined relative to investable coins, small portfolios typically underperform large ones, often significantly. The size premium here looks less like a durable return and more like a product of breakpoint and weighting choices. It is not a premium an investor can reliably earn in these designs.

Momentum lives in value-weighted, investable portfolios

CMOM (winners minus losers, two-week momentum), 2018 onward.
Weights Lag (days) Designs Mean Sharpe Significantly positive Significantly negative
Value 0 41472 1.09 85% 0%
Value 1 41472 1.10 87% 0%
Capped value 0 41472 0.03 27% 21%
Capped value 1 41472 0.62 49% 2%
Equal 0 41472 -2.36 2% 64%
Equal 1 41472 -0.37 16% 31%

Momentum is robust in value-weighted designs and weak or negative elsewhere. For CMOM (winners minus losers, two-week momentum), 2018 onward, value-weighted momentum is strong and stable: it has the highest mean Sharpe of any weighting scheme, the large majority of designs are significantly positive, and almost none are significantly negative. The one-day implementation lag barely changes it, so value-weighted momentum here is not a lag artefact.

Capped-value weighting is weaker and more lag-sensitive: its mean Sharpe is close to zero at lag 0 and improves markedly at lag 1. Equal weighting is negative throughout, and its worst results are at lag 0. Equal-weighted “momentum” therefore looks less like momentum and more like microcap short-term reversal.

The one-day lag matters mainly where microcaps dominate the portfolio. It turns capped value positive and partially mitigates equal weighting, while leaving value-weighted momentum essentially unchanged. Taken together, the table supports a robust momentum premium in value-weighted, more investable designs, and little or no momentum in equal-weighted ones. Whether the equal-weighted loss is specifically microcap short-term reversal would need a direct decomposition, but the pattern is consistent with that interpretation.

Non-standard errors

Following Menkveld et al. (2024) and Fieberg et al. (2024), the non-standard error (NSE) is the standard deviation of a factor’s Sharpe ratio across designs. The standard error (SE) is the usual sampling uncertainty of one Sharpe ratio (Lo, 2002). Their ratio N/S tells how much more the result depends on the researcher’s choices than on the data.

Non-standard vs standard errors of annualised Sharpe ratios. 2014–2021 is the sample of Fieberg et al. (2024).
Period Factor Breakpoints Designs Mean Sharpe NSE SE N/S
2014-2021 CMKT All coins 82944 1.20 0.46 0.37 1.26
2014-2021 CMKT Top 100 82944 1.20 0.46 0.37 1.26
2014-2021 CMKT ≥ $100M 82944 1.20 0.46 0.37 1.26
2014-2021 CSIZE All coins 82944 1.20 1.48 0.42 3.48
2014-2021 CSIZE Top 100 82944 0.50 0.85 0.42 2.03
2014-2021 CSIZE ≥ $100M 82944 0.26 0.81 0.57 1.41
2014-2021 CMOM All coins 82944 -0.05 1.64 0.42 3.88
2014-2021 CMOM Top 100 82944 0.12 1.42 0.42 3.37
2014-2021 CMOM ≥ $100M 82944 0.20 1.51 0.45 3.33
2018+ CMKT All coins 82944 0.16 0.30 0.34 0.88
2018+ CMKT Top 100 82944 0.16 0.30 0.34 0.88
2018+ CMKT ≥ $100M 82944 0.16 0.30 0.34 0.88
2018+ CSIZE All coins 82944 0.48 1.56 0.34 4.52
2018+ CSIZE Top 100 82944 -0.65 0.61 0.34 1.78
2018+ CSIZE ≥ $100M 82944 -0.67 0.68 0.36 1.86
2018+ CMOM All coins 82944 -0.08 1.88 0.35 5.45
2018+ CMOM Top 100 82944 0.05 1.62 0.34 4.72
2018+ CMOM ≥ $100M 82944 0.08 1.64 0.34 4.77

For size and momentum, the design choices matter several times more than sampling noise. Investable breakpoints cut the non-standard error of the size factor by more than half.

N/S ratio when one design choice is held fixed (2018 onward). Lower is better: fixing that choice removes more of the dispersion.

The effects of individual choices match those in Fieberg et al. (2024). Value weights, a $1M floor, a $1 minimum price and terciles reduce the dispersion; stablecoins make no difference. For momentum, value weights (N/S 1.2 instead of 6.3 with equal weights) and a one-day lag (2.9 instead of 6.2) matter most. Of the two choices that they do not vary, where the breakpoints come from matters as much as any of theirs for size (N/S 4.5 with all-coin breakpoints, 1.9 with breakpoints from coins ≥ $100M). Which day the week ends shifts individual weekly numbers, but changes the overall dispersion only modestly.

Where the standard sits

The standard within the distribution of designs, 2018 onward.
Factor Source Weeks Sharpe t Percentile, all designs Percentile, investable VW designs
CMKT CMC 458 0.41 1.21 77% 47%
CMKT CG 458 0.41 1.22 77% 48%
CSIZE CMC 458 -0.68 -2.02 47% 90%
CSIZE CG 458 -0.83 -2.45 36% 76%
CMOM CMC 458 1.41 4.19 92% 77%
CMOM CG 458 1.36 4.04 90% 71%

We did not choose the standard for its Sharpe ratios; see The Standard for the criteria. Its momentum factor is nonetheless among the strongest designs. Its size factor is negative, like that of most investable designs.

References

Fieberg, C., Günther, S., Poddig, T., & Zaremba, A. (2024). Non-standard errors in the cryptocurrency world. International Review of Financial Analysis, 92, 103106.
Lo, A. W. (2002). The statistics of Sharpe ratios. Financial Analysts Journal, 58(4), 36–52.
Menkveld, A. J. et al. (2024). Nonstandard errors. The Journal of Finance, 79(3), 2339–2390.