Hyperliquid Signed Regression-Intercept Reversal: TRX, ADA and ICP 4H

Hypothesis

Derived by the Research Lead from ChaoZhang/HPotter, Quant Strategy Based on Linear Regression Intercept: test its published signed-intercept long/short rule independently on TRX, ADA and ICP perpetuals using 4-HOUR bars. Source: https://www.fmz.com/strategy/436997 . The complete locally archived Pine source was read; the live page exposes the formula but gates the full code. Preserve the formula literally, including its negative denominator: replacing it with a conventional least-squares endpoint would test another rule. Research Lead choices are the three-name panel, four-hour interval and conservative position sizing. This is a disclosed source adaptation, not evidence of profitability. Orders execute at the signal-bar close on these continuous venues; independent opposing positions on the same instrument are never required.

Measured negative portfolio expectancy after costs across 858 closed positions warrants declining optimization of this three-name, four-hour operationalization. This does not reject the source author's complete strategy, establish that every length fails, or draw conclusions about the withheld period after 2026-04-07.

Implementation

Three independent TRX, ADA and ICP 4H signed-intercept reversal sleeves. Market orders execute at the signal-bar close, the continuous-venue equivalent of next-bar-open execution. Each entry uses the prescribed MTM-equity volatility sizing and remains fixed until an opposite signal closes it.

Preserves the literal negative-denominator source formula and all fixed and tunable research parameters. Addresses QA's independent-entry failure by adding TRX's existing bar stream to extra_bar_types so the framework routes it through on_extra_bar, exactly like ADA and ICP; the primary entry path is disabled. Entries require confirmed flat state and no outstanding or in-flight entry, and raw-bar callbacks submit exits only. Missing or invalid own bars immediately clear sizing eligibility and the daily-return window, including missing midnight closes. Entry eligibility returns only after 60 fresh contiguous completed daily returns; direction state and opposite-signal exits remain active. No strategy evaluation-date guards or additional trading filters are introduced.

Backtest Review

Sharpe

-0.33

Total return

-15.71%

Max drawdown

24.79%

Trades

1169

Win rate

28.2%

Profit factor

0.92

Literal negative-denominator formula is preserved. Trades include 432 longs and 426 shorts, capital-relative sizing, and ICP activation after its own warmup.

858 closed positions provide a substantial development-period sample.

Closed-position P&L is -$8,066.17, profit factor 0.943, portfolio return -7.725%, and Sharpe -0.158.

Mean net trade return is 0.00075%, below the Hyperliquid 0.15% fee-aware floor and predicted 0.40%.

TRX loses $14,593.57, outweighing ADA's $884.74 and ICP's $5,642.66 gains. Longs gain $2,772.84; shorts lose $10,839.01.

Commissions total $8,419.14 and funding contributes -$2,215.63, already included in reported P&L. No specific supported implementation repair is established.

Iteration History

Provide a framework-supported own-bar entry path for TRX as well as ADA and ICP, with confirmed-flat and outstanding-order checks, and prevent duplicate primary-path entries. Do not submit entries from on_raw_bar, whose framework contract permits exits only.

TRX entries are handled only through the base primary signal path, which waits for same-timestamp ADA and ICP bars. on_raw_bar updates TRX and submits exits only. If either listed secondary sleeve lacks a bar, a valid flat TRX sleeve cannot enter or reopen after reversal, despite having its own valid signal and sizing history. The hypothesis requires independently traded sleeves on their own completed bars; the config rationale acknowledges this dependency but it is not an authorized hypothesis amendment.

Track sizing-window validity and freshness explicitly. Invalidate entry eligibility immediately when a gap or invalid close makes the required daily history incomplete, including gaps across midnight; restore eligibility only after the required valid daily-return window exists. Continue retaining direction and allowing signal exits.

Sizing history remains eligible after a missing or invalid bar. _advance_leg marks day_complete false for an intraday gap but leaves ret_count and vol unchanged until a midnight close. If the midnight bar itself is missing, the next day's first bar resets day_complete without clearing that stale window. _leg_size checks only ret_count, so it can open new positions using the old 60-return estimate even after a required completed daily return is unavailable. This violates the instruction to skip entries when bars or sizing history are invalid.

Outcome Summary

This three-name, four-hour implementation did not establish a cost-covering edge; its abandonment does not establish that every regression length or the source author’s complete strategy fails.

After two iterations, backtest review recommended abandonment because portfolio expectancy after costs was negative and mean net trade return fell below the 0.15% fee-aware floor; no supported implementation repair was established. Optimization, post-optimization analysis and risk review were not reached.

Test the literal signed regression-intercept long/short rule, preserving its negative denominator, on TRX, ADA and ICP Hyperliquid perpetuals using four-hour bars and capital-relative sizing.

The supplied backtest reports a -15.71% return, -0.329 Sharpe, 1,169 closed positions and 24.79% maximum drawdown. The decisive development-period review separately cited -7.725% return, -0.158 Sharpe and 858 closed positions, with mean net trade return of 0.00075%.