Strategies

Hyperliquid XLM/ETC Single-Excursion Passive Inventory Test

Outcome: Abandoned

HlXlmEtcSingleExcursionPassiveInventory

Outcome Summary

The strategy tested independent passive excursion books for XLM and ETC with limited inventory, protective stops and larger profit targets. Across 402 closed positions over 849 days, it lost $5,505.90 and recorded Sharpe -0.6786. Modeled impact overwhelmed its positive pre-impact result, prompting abandonment at backtest review before optimization. This rejects the tested implementation at registered sizing without resolving the source author's broader grid question.

Strategy report

Hypothesis

Derived by the Research Lead from wufuhao100w's single-pair versus multi-level grid question (https://www.fmz.com/bbs-topic/4684): test its inventory-limitation implication using one outstanding entry and one non-pyramided position per instrument. The full local post was read; it supplies a question, not measured performance. Trade independent XLM and ETC 4-HOUR perpetual bars with wide passive entries, fixed protective stops and larger profit targets. Replace the source's opposing order pair with one eligible side to avoid assuming atomic cancellation. Passive limits and platform-matched intrabar stops are supported. Crypto market exits execute at the signal bar's close, the registry's equivalent to next-bar-open execution. Use all available history, approximately 2.3 years maximum per Hyperliquid 4h series.

Addresses the largest direction imbalance, long-only at 76.7%, while adding multi-instrument Hyperliquid exposure, currently 12.6%. Options have a larger venue deficit but require substantially more contract-selection complexity. Corpus concentration was empty; the snapshot contains only four primary XLMUSD.HYPERLIQUID and three ETCUSD.HYPERLIQUID experiments. Prior grids suffered accumulating inventory and losses much larger than wins. This test changes those features: no ladder, averaging or replacement entries during inventory, and a 1.5% stop against a 3% target. Unlike the failed LTC/BCH relative-price grid, it assumes no spread stationarity. Inventory limits mitigate losses; they do not establish an edge.

Negative measured expectancy at registered sizing: 402 closed positions lose $5,505.90, PF is 0.8271 and average net trade return is -0.2767% versus the +0.15% Hyperliquid review reference. Modeled impact exceeds the positive pre-impact result; QA's concern is borne out. No demonstrated implementation defect or tested improvement warrants optimization. This rejects advancing the Research Lead's XLM/ETC operationalization, not the source author's grid question or inventory limitation universally. Alternative settings and a matched random-entry control remain untested; no accounting or data defect is asserted.

Implementation

Independent XLM and ETC 4-hour passive single-excursion books: one eligible entry limit and one non-pyramided position per instrument, 5% equity per book, 1.5% intrabar stop, 3% limit target, six-bar post-exit cooldown and 24-bar holding limit. Uses available history without a research date cutoff. Crypto market holding exits execute at the signal close, the registry equivalent to next-bar-open execution.

The brief names but does not define side eligibility or anchor averaging. This implementation explicitly chooses an SMA anchor and mean-reversion eligibility: buy below the anchor when close is at or below it, otherwise sell above it. Offset is max(1.5% of anchor, offset_atr times rolling mean true range). Quotes remain fixed until filled or expired and must be at least 0.1% passive when submitted; crossed quotes are skipped. Each book uses only its own closed bars. Native contingent brackets protect entry fills intrabar without assuming atomic cancellation of opposing entry quotes. Stops and targets are referenced to the quote then adjusted to actual entry fill on position-open notification. Engine adaptive OHLC matching determines same-bar stop/target ordering and queue position is not modeled. The three tunable names and baseline values match the registered plan; all other values remain fixed. Gross reserved exposure across the two books is at most 10% of settled equity at placement, with 0.075% equity risk per default stop. Inventory limitation is a risk constraint, not evidence of alpha.

Verification Results

Inspect the fills ledger for the largest-loss trade. Confirm that the SL child was ACCEPTED, that it was moved to the actual-fill level, and how the exit happened. If stops can be rejected as already triggered, check for a missing stop after the fill and submit a reduce-only _submit_stop_market or a market exit.

Loss size does not reconcile with the stop. At ~5% equity per book, a 1.5% stop should lose about $75-90 plus ~$20 of impact and fees. The report shows avg_loss $139 and largest_loss -$1,794.57 (~35% of one book's notional). The HL 4h bars have no time gaps, so a gap-through fill should not explain it. Possible causes: a stop rejected on release because it was already triggered, which leaves the position unprotected until the 24-bar time stop; a stop modify in on_position_opened acting on an order not yet released; or a very large impact charge on a low-volume bar.

None required.

min_bars_required() overrides the base to read warmup_bars (60). This is deliberate and matches evaluation.warmup_bars. With anchor_bars up to 42 and atr_bars 14, the 60-bar warmup is enough.

Optional: use '<' if exactly six bars is intended.

The cooldown is measured from the position-close timestamp. With a strict '<=' against 6 x 4h, the next quote can only be placed on the 7th bar after an intrabar exit. The resulting cooldown is 6-7 bars, which is within the spirit of the hypothesis.

Smoke run: 401 closed positions, avg net trade -0.30% of notional, PF 0.82. Modeled impact ($6,468) is larger than the net loss ($5,911), so the books are only about +$560 before impact and -$400 after commissions. capacity_usd = 742 against ~$5k per-book notional says the 5% fixed sizing is far above what thin Hyperliquid XLM/ETC 4h volume absorbs, even though entries are passive. The 3%/1.5% target/stop structure won only 43% of trades, and avg_loss ($139) is close to avg_win ($153). The analyst should check the impact-free PnL and whether the inventory limit adds anything over a random-entry control with the same brackets.

Backtest Review

Sharpe: -0.68
Total return: -5.51%
Max drawdown: 8.86%
Trades: 402
Win rate: 42.5%
Profit factor: 0.83

402 closed positions across 849 days provide a functioning baseline. Both instruments trade long and short without overlapping positions per instrument; sampled exits include the prescribed stops and targets.

Closed-trade PnL reconciles to the $5,505.90 account loss. Funding contributes $417.76 and no terminal inventory remains.

Average net trade return is -0.2767% versus the +0.15% Hyperliquid review reference; profit factor is 0.8271 and Sharpe is -0.6786. Both instrument books lose money.

QA's economic concern is borne out at registered sizing: $6,521.88 of modeled impact turns approximately $1,015.98 of net pre-impact profit into a $5,505.90 loss.

The 5% equity allocation per book is contract-fixed. No demonstrated code defect or tested viable parameter region supports iteration. Alternative settings and a matched random-entry control remain untested.

The benchmark is marked not meaningful, so buy-and-hold return and drawdown are not fair edge tests here.

Outcome Summary

Passive entries and inventory limits did not establish a viable edge at registered sizing; alternative settings and a matched random-entry control remain untested.

Backtest review abandoned it for negative expectancy at registered sizing, with no demonstrated implementation defect or tested improvement warranting optimization. Optimization, post-optimization analysis and risk review were not reached.

Tested passive excursion entries on Hyperliquid XLM and ETC 4-hour perpetuals, allowing one outstanding entry and one non-pyramided position per instrument with fixed stops and larger profit targets.

Across 849 days and 402 closed positions, it returned -5.51% (-$5,505.90), with Sharpe -0.6786, profit factor 0.8271 and maximum drawdown 8.86%. Average net trade return was -0.2767%; modeled impact of $6,521.88 overwhelmed approximately $1,015.98 of pre-impact profit.

Backtest and paper results are hypothetical. Trading involves risk of loss.