Research Note 02

Momentum: Evidence from a Binance Replication

The research question: on a practically tradable Binance universe of USDⓈ-M perpetual futures, does a cross-sectional momentum spread remain statistically distinguishable from zero after realistic trading costs, rather than only on paper. Liu, Tsyvinski & Wu ("Common Risk Factors in Cryptocurrency," The Journal of Finance, 2022) document momentum as one of three significant factors across a broad sample of more than 1,700 coins, with a long-short spread significant across formation horizons of one to four weeks. A double sort in the same paper, and separately in an earlier conference presentation by Yukun Liu, shows the effect concentrated among larger, more liquid coins rather than small ones — a detail that matters directly for what universe should be tested. Starkiller Capital, an asset manager, published its own honest in-sample/out-of-sample backtest (SSRN 4322637) spanning the 2021–2022 bear market: the strategy lost less than the market in relative terms without producing an absolute profit, and the fund named a concrete cost threshold — 125 basis points — above which the strategy stops being viable.

Replication method and data

The project's own replication used 832 Binance USDⓈ-M perpetual symbols on a daily-close panel from 2020-01-01 through 2026-07-31, deliberately restricted to Binance perpetuals rather than the full CoinMarketCap universe to avoid exchange-level survivorship bias and the illiquid tail where momentum returns are known to evaporate once costs are accounted for. Methodology was pre-registered before the result was seen: Sunday-close ranking, Monday-open entry, one-week holding, quintile baskets, and a fixed $1,000,000 trailing-turnover liquidity threshold.

Gross and net results

Over 338 independent weeks, the gross spread averaged +0.573% per week, with 52.7% of weeks positive; one of three significance tests (a within-week permutation test) was formally significant, while a bootstrap confidence interval and a sign-flip test were not. Applying a Binance taker fee of 0.05% per side plus an assumed 0.05% per side of slippage — 0.40 percentage points round-trip — reduced the net spread to +0.173% per week, with none of the three significance tests passing and the share of profitable weeks falling to exactly 50.0%.

A second, literature-mandated specification (two-week formation and holding, non-overlapping periods) was run because two independent sources — Dobrynskaya and Li & Zhu — separately identified the two-week formation horizon as more robust than the one-week horizon the project started with. Its net result was also indistinguishable from zero, and its sign flipped depending on an arbitrary choice of which Monday the non-overlapping periods started counting from.

Verdict

The verdict separates three tiers of evidence. Multi-year peer-reviewed literature confirms cross-sectional momentum as a real, statistically significant anomaly on the broad crypto market, concentrated among larger coins. A published practitioner backtest by an actual asset manager beat the market in relative terms through a real bear market without turning a profit. The project's own replication on an actually tradable Binance universe found no net-of-costs spread distinguishable from zero in either of the two specifications the project's rules permitted testing.

Full note: GitHub ↗ · SSRN 7404139 ↗ · DOI ↗