Jesse Community RSI2 by FengkieJ: Faithful Long-Short Port on DOT Hyperliquid Daily Bars

Hypothesis

Reproduce the supplied RSI2 trading strategy, originally implemented by FengkieJ on 2020-06-27 from Larry Connors' RSI2 concept. Preserve every strategy rule and all five defaults: fast_sma_period=5, slow_sma_period=200, rsi_period=2, rsi_ob_threshold=90, rsi_os_threshold=10. Test DOTUSD.HYPERLIQUID on 1-DAY bars with a margin account, fixed 1.0 leverage and the source's entire-balance, fee-adjusted sizing. The source specifies neither instrument nor timeframe; these are platform research choices, not original defaults. No stop-loss, profit target, maximum holding time, funding filter or signal-strength sizing is present. Performance is unverified. Source: [pinned RSI2 implementation](https://github.com/jesse-ai/example-strategies/blob/7c91e0a37bf62165790120d730442e4f6eb00364/RSI2/__init__.py).

The explicit faithful-reproduction request takes precedence over the generic avoidance of textbook RSI mean reversion. This is a reproduction experiment, not a claim of a novel crypto mechanism. Hyperliquid is underrepresented at 8.0% versus its 20% target, and retaining shorts addresses the 82.8% long-only concentration. DOT avoids another BTC/ETH experiment, although the empty corpus_concentration object prevents verifying DOT's historical share. Single-instrument scope is retained because adding hedges or rotation would change the source. Daily bars offer substantially longer Hyperliquid validation history than intraday bars.

Negative expectancy in this fixed-default DOT daily reproduction: 55 closed trades produce PF 0.839, -25.68% net return and -0.1811% average notional return versus the 0.15% review floor. Both long and short books lose after modeled costs and credited funding. Optimization cannot tune the frozen specification; this finding does not reject RSI2 on other targets.

Implementation

Faithful daily DOT Hyperliquid RSI2 port: buy above SMA200 when RSI2 is at most 10; short below SMA200 when RSI2 is at least 90. Exit longs above SMA5 and shorts below SMA5. Use the entire equity balance with fee-adjusted sizing at fixed 1.0 leverage.

The consistency validator checks fixed research values inside parameters, including leverage. Explicitly declaring parameters.leverage as the float 1.0 alongside config.leverage=1.0 addresses the reported Layer 1.5 failure. Indicator calculations, entry conditions, exits and sizing remain unchanged; only a sizing comment changes the source hash. Market orders at the daily decision point are the platform adaptation of the source's price-specified entries. No additional risk rules or filters are introduced. The required outbox file could not be written because this session's filesystem is read-only; the complete bound JSON is returned here.

Verification Results

Acceptable as a faithful port. If rejections appear in the full backtest, add a small buffer to the reserve rather than changing the rule.

Sizing commits about 99.9% of equity at 1.0x (equity * (1 - 3*taker_fee) / close). The order fills at the next bar's open plus modelled impact, so realised notional can exceed equity; the sandbox reports avg_position_pct 107%. This is the source's own entire-balance sizing, as the hypothesis states, but margin rejections are possible on gap-ups.

Leave as is for the faithful port. Any stop belongs in a separate, explicitly labelled variant hypothesis.

The strategy has no stop-loss or maximum holding time, and exit relies only on a close crossing SMA5. This matches the source and the hypothesis says so explicitly, so it is not a defect. Combined with full-equity sizing, it still lets a single trade lose a large share of equity (sandbox DD 60.6%).

Make sure the full backtest folds HYPERLIQUID DOT funding. This is not a strategy-code change.

Funding was not folded into the sandbox (funding_folded=false, funding_events_available=0). Average hold is about 4.9 days on a HYPERLIQUID perp, so hourly funding is a real carry that this PnL leaves out.

Sandbox (one fixed draw, 1347 daily bars): -28.0% return, PF 0.83 on 55 trades, avg_trade_return_pct -0.23%, max DD 60.6%, which is above the plan's own 40% max_drawdown_pct constraint. Longs win only 37.5% (16 trades). Shorts win 59% (39 trades). The two sides behave very differently. With no stop and full-equity 1x sizing, one adverse trade cost about $33.8k (largest loss). Check first: drawdown versus the pre-registered 40% cap, and long-side versus short-side PnL. Funding was not folded into the result (funding_events_available=0), so multi-day Hyperliquid holds carry an unmodelled cost or credit.

Backtest Review

Sharpe

0.02

Total return

-25.68%

Max drawdown

58.96%

Trades

55

Win rate

52.7%

Profit factor

0.84

55 closed trades over 1,347 days; entries function and position quantities vary with capital.

Funding is credited: 3,048 available events and approximately $2,290 net funding income, superseding QA's missing-funding concern.

QA's economic concern is borne out: net return -25.68%, PF 0.839, and average trade return -0.1811%, below the supplied Hyperliquid 0.15% review floor.

Both sides lose: longs -$11,370 across 16 trades; shorts -$14,311 across 39 trades.

Drawdown 58.96% exceeds the declared 40% risk constraint; this supports, but does not independently determine, rejection.

All parameters are frozen, leaving no authorized tuning path. No specific implementation defect was established.

Buy-and-hold returned -83.48%, with approximately 93.22% drawdown on the stored sampled curve; relative outperformance does not establish positive expectancy.

Iteration History

Verification failed (Layer 1.5 — hypothesis/config consistency) [class=hypothesis_mismatch]: - optimization_plan fixes leverage at 1.0

Outcome Summary

This fixed-default DOT daily reproduction showed negative expectancy despite outperforming buy-and-hold, without establishing that RSI2 fails on other targets.

Backtest review returned abandon because both long and short books lost money and average trade return fell below the 0.15% review floor. All parameters were frozen, leaving no authorized tuning path; optimization and subsequent analyst, risk, and promotion stages were not reached.

Faithfully reproduce the fixed-default Connors RSI2 long-short strategy on DOT Hyperliquid daily bars at 1.0 leverage, using a 200-day SMA trend filter, RSI2 extremes for entries, and 5-day SMA exits.

The backtest produced 55 closed trades over 1,347 days, returning -25.68% net with Sharpe 0.024 and profit factor 0.839. Maximum drawdown was 58.96%, and average trade return was -0.1811% of notional after modeled costs and credited funding.

Analysis

RSI2