# Beyond Anonymous Order Flow: Wallet Identity as a Feature for Short-Horizon Return Prediction

Published 2026-11-06 · Crypto.world Team · Crypto.world

> An arXiv preprint tests whether activity by previously scored Hyperliquid wallets improves short-horizon return forecasts beyond anonymous order-book features. The authors report higher out-of-sample scores when wallet features are added, while noting that retrospective choices and data limitations constrain the results.

An [arXiv](https://arxiv.org/html/2611.05218) preprint by Jingqiu Ding and Daojing Zhai examines whether persistent wallet identities can improve short-horizon return predictions on [Hyperliquid](https://crypto.world/companies/hyperliquid). The researchers reconstruct full-depth order books from consensus-ordered messages and compare models using anonymous market features with models that also include wallet-specific activity.

## A July record of trading activity

The study covers Hyperliquid records from July 1 to 27, 2026: 17.1 billion messages, 14.3 million aggressive orders by 147,113 wallets, and $84.3 billion in taker notional across ten perpetual markets. The authors estimate their models on bitcoin, ether and solana markets.

Public order-book data typically summarize trading without identifying the accounts behind it. Hyperliquid’s Level-4 record includes persistent pseudonymous wallet addresses, allowing the researchers to examine how trading by particular wallets relates to subsequent price movements.

The analysis separates wallet scoring, model fitting and evaluation into chronological periods. Wallets are scored using aggressive-order markouts from July 1–10; models are fitted on July 11–20 and evaluated on July 21–27. The anonymous feature block contains ten variables summarizing prices, quotes and order flow. An additional eleven variables describe activity by wallets that had previously ranked in the top decile.

## Wallet rankings and forecast results

Across the three model markets, rankings based on notional-weighted ten-second markouts persist across adjacent ten-day windows, with a Spearman correlation of 0.52. After covariate adjustment, validation markouts reach 3.11 basis points in the top ventile, or top 5% of wallets.

On a 100-millisecond grid, adding the eleven wallet features raises the one-second out-of-sample R² of the ridge model from 10.88% to 12.31%. The preprint reports this as a 13.2% relative increase, with a t-statistic of 9.2. A gradient-boosted tree model also improves: its score rises from 19.48% with anonymous features to 20.65% with wallet identity, a 6.0% relative increase.

The authors report that the relative gains are largest at shorter prediction horizons and remain positive through ten seconds. A peer-adjusted wallet score, which controls for markouts in the same market and minute, produces a smaller 6.8% one-second increase.

## Controls and limits

As a comparison, the researchers test 200 cohorts matched on scoring-window notional and order count. The one-second ridge increment for the wallet features is 1.6 times the largest increment among those cohorts. The authors caution that donor shortages made the comparison cohorts smaller than the top-decile group, so the ranking is descriptive.

Delaying all features by 200 or 300 milliseconds leaves relative ridge gains of 10.2% and 9.1%, respectively. An earlier public sample from December 2025 also produces a positive increment under the same volatility-scaled specification.

The authors describe the results as a retrospective comparison with the tested anonymous feature set, not a demonstration of executable profit. They note that specification choices were made retrospectively, that states were carried forward during data gaps, and that the analysis uses consensus time rather than receipt time. Those limits mean the study does not establish that wallet activity reflects private information or that the reported forecasts could be traded profitably.

## Sources

1. Original source · Preprint [arXiv](https://arxiv.org/html/2611.05218) arXiv:2611.05218

## Related stories

- [Hyperliquid Strategies Files for U.S. Futures Broker Status](https://crypto.world/news/hyperliquid-strategies-futures-commission-merchant-application)
- [Trade.xyz Opens Anthropic’s 10x Pre-IPO Perp on Hyperliquid](https://crypto.world/news/trade-xyz-anthropic-pre-ipo-perp-hyperliquid)
- [ICE and TradeXYZ Plan U.S. Venue for Hyperliquid Perpetuals](https://crypto.world/news/ice-tradexyz-hyperliquid-perpetuals-us-venue)
- [Former Robinhood engineer admits using listing data in Hyperliquid trades](https://crypto.world/news/former-robinhood-engineer-guilty-hyperliquid-listing-trades)
- [CFTC's Selig says onchain derivatives venues must register or partner with an exchange to reach U.S. retail](https://crypto.world/news/cftc-chair-onchain-derivatives-registered-exchanges)

## Explore

[More articles](https://crypto.world/news)

---

Canonical: https://crypto.world/news/study-wallet-identity-hyperliquid-return-signals
