Strategies

LINK Hyperliquid Daily Pin-Bar Reversal with Stop-Entry Confirmation

Outcome: Abandoned

LinkHyperliquidDailyPinBarStopEntry

Outcome Summary

LinkHyperliquidDailyPinBarStopEntry tested an operationalized daily pin-bar reversal thesis with stop-entry confirmation and equity-relative sizing. In one iteration, its backtest produced 55 closed positions, a -7.83% return, and a -0.799 Sharpe ratio. The reviewer abandoned it before optimization because measured economics were materially negative after costs and credited funding, and no tested profitable parameter region supported further iteration.

Strategy report

Hypothesis

Derived by the Research Lead from MQL5 Code Base's PinBar indicator by PunkBASSter: test completed daily rejection candles using the published recognition defaults and stop-entry direction. This is a partial-rule source, completed with explicit order expiry, profit-taking, time exits and equity-relative sizing. Trade LINKUSD.HYPERLIQUID long and short on 1-DAY bars. Preserve MinBarSize=200, PriceOffset=50, BarRatio=2.2, TailRatio=1.3, UseBodyDirection=false, UseCloseThirds=true and ExtremumOfBars=10. Adapt broker-specific points into proportional research points: one point equals point_unit_pct percent of the close preceding the signal candle, baseline 0.01%; therefore the baseline minimum range is 2% and the entry/stop offset is 0.5%. This normalization is our adaptation, not the author's rule. Entry and protective stops trigger intrabar through platform matching; no exchange-hosted protection is required. The article was read, but the downloadable indicator code could not be retrieved, so this is not claimed as a byte-faithful implementation. No preliminary performance test was performed.

The submitted daily pin-bar mechanism requires neither BTC nor a momentum crossover. Hyperliquid is underrepresented at 11% versus its 20% target, and short-capable trading addresses the larger long-only concentration of 79.8%. Daily bars avoid Hyperliquid's short intraday history. Corpus concentration was empty, so LINK's individual research share cannot be verified. Keep the single-instrument source mechanism rather than inventing a hedge solely to satisfy diversity quotas.

negative_expectancy: decline optimization of this Research Lead operationalization of PunkBASSter's PinBar recognition and stop-entry direction on LINK daily bars. Across 55 closed positions over 1,659 days, measured economics lose materially after costs and credited funding. This rejects the tested proportional-point adaptation and completed exit rules, not the unavailable original indicator code or pin-bar trading generally. Alternative specifications remain untested.

Implementation

LINKUSD.HYPERLIQUID daily pin-bar reversal, long and short. Each completed daily bar is checked against the PinBar defaults. Its range must be at least MinBarSize points, where one point is point_unit_pct% of the previous close. The range must be at least BarRatio times the body, and the rejection tail at least TailRatio times the opposite tail. The close must be in the favourable third of the range (UseCloseThirds), body colour is ignored, and the pin's low (or high) must be the 10-bar extreme (ExtremumOfBars). A bullish pin places a BUY STOP PriceOffset points above its high, with the protective stop PriceOffset points below its low; a bearish pin is the mirror image. The entry stop rests for entry_expiry_bars bars, and a new pin replaces a pending one. Once the entry fills, a reduce-only STOP_MARKET protective stop and a reduce-only LIMIT target at target_r_multiple x planned risk rest intrabar. A time exit closes the position after max_holding_bars bars. Each trade risks risk_pct of equity on the planned stop distance, capped at max_exposure_pct x leverage notional. No pyramiding.

Implements the Research Lead's distilled hypothesis. It keeps the published PinBar defaults and the stop-entry direction, and adds the specified proportional point units, order expiry, R-multiple target, time exit and equity-relative risk sizing, with parameter names exactly as in the optimization plan (fixed values unchanged; the 4 tunables use their baselines). The source indicator code could not be retrieved, so these comparison conventions are mine and are written in the class docstring: 'range >= BarRatio x body', 'nose >= TailRatio x opposite tail', close-thirds as above, and the extremum taken over the prior ExtremumOfBars-1 bars, inclusive. Entries are true resting STOP_MARKET orders, so the trigger is matched intrabar (a gap fills at the open). The stop and target are placed from on_order_filled, anchored to the planned trigger and stop levels. All per-trade state is scalar attributes; expiry and holding are counted in bars from the bar timestamps. Daily LINK ranges usually exceed the 2% minimum, and the close-third and 10-bar-extreme gates are the binding filters, so I expect roughly 10-20 qualifying pins a year, with only some of them triggering.

Verification Results

Cancel only the reduce-only orders (cancel_order on each), not every order on the instrument, and skip the order-management calls while self._in_warmup is set.

In calculate_signal, while flat, any leftover reduce-only order makes the code call cancel_all_orders(instrument_id). That also cancels a pending entry stop that has not expired, and pend_* is left set. This should not happen in practice: on_order_filled and _submit_exit both clear the protective legs when the position goes flat. The one exception is the paper warmup path, which runs calculate_signal and can cancel orders that were carried across a restart. That is restart behaviour only, so it is a warning.

Optional: re-anchor the target to the fill price (event.last_px), or recompute the R from the fill.

The target and stop are anchored to the planned trigger, not the actual fill. When a daily gap fills the entry well beyond the trigger, the realised R differs from target_r_multiple x planned risk. The size is also computed from the planned distance, so the realised risk can be above risk_pct. The class docstring discloses this, and the 30% notional cap bounds it.

None needed.

min_bars_required() is overridden to ExtremumOfBars+1 (11). This is deliberate and correct: the base gate skips calculate_signal until 11 bars exist, and _detect_pin also requires 9 prior bars. The static-analysis base_shadow warning is benign.

The sandbox lost money: 55 closed positions over 1,659 days, PF 0.545, win rate 31%, avg_trade_return_pct -2.64% against the +0.15% HYPERLIQUID floor, and every volatility tercile negative. With a 2R target, a 31% win rate should be close to breakeven. The measured avg win/avg loss is only about 1.2, so most winners seem to leave through the time exit before they reach the target, or the gap fills move the entry away from the planned trigger. The analyst should first split exits by reason (stop / target / time). The sample is thin: roughly 13 positions a year at baseline. Also, the report shows funding_folded=false with 0 funding events, so Hyperliquid funding on 5-6 day holds is not in this PnL yet.

Backtest Review

Sharpe: -0.80
Total return: -7.83%
Max drawdown: 9.59%
Trades: 55
Win rate: 30.9%
Profit factor: 0.56

The strategy functions: 74 submitted entries produced 55 closed positions, balanced between 27 longs and 28 shorts, with variable equity-relative sizing.

Net return is -7.83%, PF 0.556 and average trade return -2.565%, below the +0.15% HYPERLIQUID review floor. All three volatility terciles lose.

QA's economic concern is borne out. Average winner/loss is 1.24 at a 30.9% win rate, giving negative expectancy of $142 per position.

The detail file contains no explicit exit reasons. Sixteen positions lasting exactly ten days net -$361; the other 39 net -$7,468. Losses therefore are not confined to likely time exits.

QA's funding caveat is superseded: this report folds funding into PnL, and trade funding totals +$226.53. Funding does not rescue the result.

No verified implementation repair or measured profitable parameter region supports developer iteration; no parameter sweep has been evaluated.

positive net expectancy

pre-optimization fee viability

Outcome Summary

The tested proportional-point adaptation and completed exit rules need evidence of positive expectancy before further optimization; the result does not establish how alternative specifications or the unavailable original indicator code would perform.

The backtest reviewer issued an abandon verdict because expectancy was negative after costs and credited funding, with losses across all three volatility regimes and no measured profitable parameter region supporting iteration. Optimization, post-optimization analysis, and risk review were not reached.

It tested long and short LINK daily pin-bar reversals on Hyperliquid using stop-entry confirmation, proportional-point normalization, and explicit protective stops, targets, and time exits.

Across 55 closed positions over 1,659 days, it returned -7.83% with a Sharpe ratio of -0.799 and maximum drawdown of 9.59%. Profit factor was 0.556, win rate was 30.9%, and average trade return was -2.565% of notional.

Analysis

PinBar

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