Research Note 03

Trend: When Published Edge Ages

The research question: does time-series momentum, one of the most repeatedly confirmed strategies in traditional-asset finance, still hold as a real edge in crypto today, or did the early favorable crypto evidence describe a market regime that has since passed — and separately, where the literature does show a positive number, can a retail trader on a real exchange actually reach it. Moskowitz, Ooi & Pedersen (2012) tested time-series momentum across 58 liquid futures and forward contracts over more than two decades, finding the effect present across nearly every asset class examined, using inverse-volatility position sizing that later matters for interpreting crypto-specific volatility-scaled results. Liu & Tsyvinski (2021, Review of Financial Studies) found a strong Bitcoin time-series momentum effect on data running only through May 2018 — a market that predates deep Binance perpetual futures, meaningful institutional participation, and the post-2020 structural changes the current market operates under.

The decisive test

The decisive test is Grobys, Kolari, Sandretto, Shahzad & Äijö (2025, Financial Markets and Portfolio Management), using 30 large-cap coins over 416 weekly observations from January 2016 through December 2023 — the longest and most methodologically recent available crypto-specific test. Split by regime, the full-sample result is 0.90% per week and not significant; the pre-July-2020 subsample is 1.74% per week, significant only at the 10% level; the post-July-2020 subsample — the regime closest to today's market — is negative and not statistically significant. A single extreme short-leg event (the coin Mindol, a roughly 1,400% one-day move in December 2020) drove much of the raw sample's insignificance; trimming that one outlier recovers a significant 1.51% per week. A volatility-managed version of the strategy reaches 1.86% per week on the full sample, but this figure is weighted toward the older, more favorable regime and does not independently establish that a volatility overlay revives the strategy specifically after July 2020. A broader robustness check on 2,500 coins, value-weighted, is significantly negative at −3.40% per week and stays negative under every trimming approach tested, and — separately from the regime question — was never a reproducible single-exchange retail portfolio to begin with, since no exchange has ever listed anything close to 2,500 coins with tradable perpetual futures depth on both sides.

Why this note closes without a Binance replication

This note is the only one in the series so far to close without running the project's own Binance replication, and states why directly: when the most recent, longest, and methodologically strongest available literature already answers the question of statistical significance in the regime a trader would actually be trading today, running a private replication would only re-derive a result the literature has already established with more data than the project could practically collect.

Verdict

The verdict: real academic standing on traditional assets, but a regime-dependent crypto picture — evidence that appears real in an earlier data window is not established, and is not currently reachable, in the market as it exists now. This is a calendar problem, not a fee-tier or capital problem: no amount of retail capital or execution skill can access a period of market history that has already passed.

Formal write-up: SSRN 7405060 ↗ (working paper, DOI 10.2139/ssrn.7405060 ↗)

Full note: GitHub ↗