Factor investing in digital assets.

Why the analogy is useful

Cross-sectional factor investing rests on a small number of empirical regularities that have persisted across decades, geographies and asset classes: assets with higher recent returns tend to continue outperforming over short horizons, assets priced cheaply relative to fundamentals tend to outperform over long horizons, assets with stronger balance sheets and higher-quality cash flows tend to deliver better risk-adjusted returns, and positions that earn income while held often add to total return even when price appreciation is modest. The underlying thesis is that markets are efficient enough to punish naive mistakes, but inefficient enough that disciplined, systematic exposure to persistent return drivers pays off over full cycles.

Digital asset markets meet the first condition (they are efficient enough, in the liquid part of the universe, to punish naive mistakes) and the second: there is enough structural dispersion across assets and venues to reward disciplined cross-sectional construction. This is the reason factor thinking travels.

Where the analogy breaks

It travels with important caveats.

• Fundamentals are young. The equity "value" factor leans on decades of accounting data and a well-understood concept of intrinsic value. Digital assets have neither the history nor an agreed-upon concept of fundamental value. On-chain metrics (active addresses, fees, revenue, tokenholder flows) are useful but noisy, and their economic interpretation evolves faster than the literature can keep up.
• Carry is regime-dependent. In equities and currencies, carry is a first-order driver with reasonable persistence. In digital asset markets, the yield surface available to institutional participants changes materially with the macro cycle and venue structure. A carry model that does not adapt to regime is a carry model that will eventually blow up.
• Quality is hard to define. The equity-style "quality" factor aggregates profitability, leverage, and stability. The digital asset analogue has to be assembled from uneven data (developer activity, protocol revenue, governance concentration, custodial resilience), each with its own measurement error. The composite is real, but it is not a one-line formula.
• Momentum is fast. Cross-sectional momentum in digital assets operates on materially shorter horizons than in equities. Classical twelve-minus-one construction is, in this market, almost an anti-signal over long horizons. Shorter, regime-aware windows are more defensible, at the cost of higher turnover.
• Capacity is finite. Tokens have narrower borrow availability, shallower order books outside the top of the universe, and greater venue fragmentation than a typical equity universe. Capacity must be an input to the factor, not an afterthought.

How we approach it

One market-neutral research specification combines a small number of well-defined factors (momentum, carry, dispersion, liquidity and quality) inside a hypothetical exposure-design layer that is explicit about capacity and turnover assumptions, with a neutralisation step that removes sector and market-beta exposure. Every factor is specified ex ante, validated out-of-sample and stress-tested across historical regimes before it is retained as a research output. Factors that fail out-of-sample are discarded.

We also maintain a clear separation between effects supported by repeatable evidence (including cross-sectional dispersion, microstructural effects and certain flow-driven effects) and narrative-driven signals, which remain research hypotheses rather than model inputs. Narrative is a context layer, not an alpha source.

How counterparties can assess a research model

For authorised asset managers and institutional counterparties, a systematic digital-asset research model should be evaluated with the same rigour as a traditional factor model:

• Does the model have an articulable thesis per factor, or does it rely on an opaque composite?
• Are factors validated out-of-sample, or are they anchored in a visually attractive back-test?
• Are hypothetical exposure assumptions explicit about capacity, turnover and cost, or are costs treated as a back-test detail?
• Is the neutralisation step defensible, and are the model's intended and unintended exposures documented?

Those questions do not predict returns. They help an authorised manager decide whether a research output merits its own independent analysis. DFI Labs does not determine or monitor positions in client accounts.