Strategies

Hyperliquid Alt Basket: Monthly Dynamic Logistic Momentum Selection

Outcome: Abandoned

HyperliquidMonthlyDynamicLogisticMomentum

Outcome Summary

The strategy adapted dynamic logistic momentum selection to a monthly long-short basket of Hyperliquid perpetuals, requiring 48 complete months of history before trading. After three iterations, Layer 3 verification reported zero trades across 19 evaluated daily LTC bars, with no entry side returned. The run was abandoned at verification, leaving no performance or optimization results for later review.

Strategy report

Hypothesis

Derived by the Research Lead from Bruno P. C. Levy and Hedibert F. Lopes: adapt the paper's dynamic logistic classifier and momentum-model selection to a long-short basket of LTC, BCH and ADA perpetuals. Preserve calendar-month forecasts, lookbacks of 1, 2, 4, 6, 8, 10 and 12 months, monthly holding periods, cutoff 0.5, coefficient-discount grid {0.98, 0.99, 1}, model-forgetting grid {0.99, 1}, and 36 labeled training months. Use daily bars to construct completed UTC calendar-month closes. This is an adaptation of the supplied full paper, not reproduction of its 56-contract results. Added implementation choices are zero-mean Gaussian coefficient priors with identity covariance, deterministic tie-breaking, and conservative exposure caps. Source volatility scaling is retained with a 40% annualized target before caps; retrospective scaling to 10% realized volatility is excluded from executable sizing. Monthly decisions execute at the completed signal bar's close, the platform's supported equivalent of next-bar-open execution on a continuous market. No exchange-hosted protection or selectable same-bar policy is required.

Treat DMA versus DMS, constant versus dynamic coefficients, and cutoff variants as alternatives within one mechanism, rather than separate hypotheses. Choose DMS with dynamic coefficients as the single source-derived experiment. Hyperliquid is below its venue target, and this basket addresses excessive single-symbol and long-only concentration. Options are further below target but would change the submitted underlying-return mechanism. corpus_concentration is empty, so instrument-specific shares cannot be verified; no claim of an unused universe is made. This differs from failed LTC/BCH intraday spread reversion: each asset receives its own monthly probabilistic forecast, with no paired spread or neutrality requirement. Retain the source's long training period even if it leaves little usable evidence; insufficient history must be reported, not bypassed.

verification_loop: Verification failed (Layer 3 — sandbox backtest): No trades produced Bar type used: LTCUSD.HYPERLIQUID-1-DAY-LAST-EXTERNAL, Bars processed: 19 Diagnostics: should_enter() returned a side 0 times over 19 evaluated bars -> your ENTRY CONDITION never triggered. Loosen the entry logic / thresholds. Ensure your strategy produces trades with the given data and parameters.

Implementation

Monthly long-short LTC/BCH/ADA Hyperliquid basket using per-asset dynamic logistic model selection across 127 momentum subsets. Retains 48 complete calendar months and 36 observed labels before eligibility, a 0.5 cutoff, 40% annual volatility target, 30% per-asset exposure cap and 90% gross cap. Adds evaluated-bar diagnostics for training eligibility and calendar decision timing.

The smallest faithful revision changes diagnostics only and preserves the previously passing imports, model, signal and sizing logic. No authorized entry-threshold change can resolve missing training history or an evaluation window without an eligible month end: all relevant values are fixed by the research plan. The 19 evaluated bars do not establish whether 1500 preceding warmup bars were replayed or whether an eligible calendar decision occurred. should_enter always returns None because entries are submitted directly for all basket legs; its side-return count is not a valid measure of basket entry attempts. Engineering must inspect the added complete-month, label and calendar diagnostics, replay synchronized preceding warmup for every asset, and evaluate an eligible month end strictly before the locked 2025-03-25 boundary. If that cannot be supported by available history, report insufficient_history rather than loosen training or use the reserved exam. This revision does not claim to fix or pass Layer 3. The required outbox file could not be written because the workspace filesystem is read-only; the complete task-bound JSON is returned here.

Verification Results

Verification failed (Layer 3 — sandbox backtest): No trades produced Bar type used: LTCUSD.HYPERLIQUID-1-DAY-LAST-EXTERNAL, Bars processed: 19 Diagnostics: should_enter() returned a side 0 times over 19 evaluated bars -> your ENTRY CONDITION never triggered. Loosen the entry logic / thresholds. Ensure your strategy produces trades with the given data and parameters.

Outcome Summary

A future iteration should ensure verification data covers the required 48 complete training months before assessing whether monthly entry conditions trigger.

After three iterations, the strategy was abandoned because verification failed on the sandbox's no-trades requirement. Later backtest review, optimization, analyst and risk-review stages were not reached.

Use monthly dynamic logistic momentum selection to trade a long-short basket of LTC, BCH and ADA perpetuals on Hyperliquid with volatility-scaled, capped exposure.

The Layer 3 sandbox backtest processed 19 daily LTC bars and produced zero trades; should_enter() returned a side zero times. No return, Sharpe or optimization results were available.

Analysis

Trend-Following Strategies via Dynamic Momentum Learning

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