QuantConnect HistoricalReturnsAlphaModel: Daily LTC/BCH Directional Signals on Hyperliquid

Hypothesis

Derived by the Research Lead from QuantConnect's HistoricalReturnsAlphaModel (Apache-2.0): preserve its default lookback=1, resolution=DAILY and one-day prediction interval. Source: https://github.com/QuantConnect/Lean/blob/856327ff33869cba98393781000c4ab1d2af423d/Algorithm.Framework/Alphas/HistoricalReturnsAlphaModel.py. This is an alpha model with actionable signals, but it does not specify portfolio construction, orders, stops or position sizing. Complete those omissions with independent equity-relative positions and signal-reversal/expiry exits; these additions are platform choices, not QuantConnect's rules. Trade LTC and BCH perpetuals independently using daily close-stamped bars. Preserve signed insight magnitude, including negative magnitudes for DOWN insights. No thresholds, trend filters, funding vetoes or added technical indicators. Market orders execute at the signal bar's close. Maintain same-direction exposure when a fresh daily insight renews it, rather than manufacturing daily round trips.

Hyperliquid is under its 20% venue target at 10.2%, while long-only strategies and single-symbol experiments are overrepresented. Two independently traded long-short altcoin signals preserve the source's per-symbol model without introducing an unsupported pairs mechanism. Daily resolution is the source default and avoids Hyperliquid's short-interval history limitation. Corpus concentration is empty, so no specific instrument-share claim is possible. This source port is intentionally tested despite prior momentum failures; historical outcomes do not justify rejecting a concrete signal model.

negative_expectancy: Decline optimization of this frozen daily LTC/BCH directional specification. Its measured net losses, PF 0.775 and -0.470% average trade return establish poor economics on the evaluated sample; this is not abandonment on drawdown or a predicted optimization outcome. Changing lookback, filters or prediction horizon would create a new hypothesis. This verdict covers the Research Lead's operationalization with independent 15% equity allocations and renewal/reversal exits, not QuantConnect's alpha model across other portfolio constructions or markets.

Implementation

Independent LTC and BCH daily signed one-bar return signals on Hyperliquid. Each position uses 15% of settled account equity at entry. Same-direction insights renew exposure without resizing; flat signals and expiry close it. Reversals close first and then enter the opposite direction while the latest insight remains valid.

Changed only on_position_closed to address the Layer 4 execution defect: delayed closing fills no longer require exact equality with the insight timestamp. Pending reversals survive temporary submission guards and are cleared when invalidated or an entry is established or in flight. Latest-insight direction and expiry govern reopening, with explicit shutdown, terminal-flattening, position and order guards. Signal definitions, independent streams, sizing and frozen research parameters remain unchanged. Sandbox economics remain concerning, but the identified latency defect warrants this correction before judging the mechanism. No verification was run. The requested outbox file could not be written because this session's filesystem is read-only; the complete bound JSON is returned here.

Verification Results

The sandbox reports -0.461% average net return per trade across 1,730 closed positions. Daily directional renewal should be assessed against Hyperliquid's approximately 0.09% round-trip taker commissions plus impact in the full backtest; these results do not establish a correctness failure.

Backtest Review

Sharpe

-1.41

Total return

-76.95%

Max drawdown

78.56%

Trades

1732

Win rate

32.6%

Profit factor

0.78

Reliable measured sample: 1,732 closed positions across 1,633 days, with no open positions at the end.

Trade spot-checks support the intended mechanism: LTC reverses long to short on the negative October 6 signal, holds that short through October 7, and subsequently maintains a six-day long rather than manufacturing daily round trips.

Net average trade return is -0.470%, versus the supplied Hyperliquid review floor of +0.15%; profit factor is 0.775 and total return is -76.95%. QA's economic concern is borne out.

All four complete calendar years and all three volatility regimes lose money. Sharpe is -1.415, with its reported confidence interval entirely negative.

The execution plan is frozen with no tunable parameters, so optimization has nothing to search.

Benchmark comparison is explicitly marked nonmeaningful; it does not support an alpha or beta conclusion.

Iteration History

Disable the cross-instrument synchronization barrier for this independent strategy through a deliberate framework-supported setting or initialization override. Keep each instrument's signal and lifecycle driven by its own bars. Verify that LTC continues to enter, reverse, and renew when a BCH bar is absent, and vice versa.

The inherited framework places same-timeframe BCH bars in its primary synchronization barrier, but this strategy leaves that barrier enabled. FactoryStrategy._maybe_process_primary() requires the latest BCH timestamp to equal the LTC timestamp before calculating or trading LTC. Consequently, a missing BCH daily bar suppresses a valid LTC insight, reversal, or renewal; an existing LTC position can then expire despite a fresh same-direction LTC bar. This contradicts the hypothesis's independent instrument signals and exposure renewal.

Verification failed (Layer 4 — QA review) [class=code_defect]: - [edge_concern] The sandbox reports -0.461% average net trade return and PF 0.778 across 1,730 closed positions. These are advisory performance concerns, not grounds for this failure. - [critical] on_position_closed clears pending_direction before requiring event.ts_event to equal the originating insight timestamp exactly. Any closing fill occurring later than the signal timestamp discards the reversal even when the insight remains valid. Consequently, normal execution latency leaves LTC or BCH flat until another daily bar instead of reopening the opposite exposure. Zero-latency sandbox fills do not verify this path. (line 157) — fix: Retain the pending reversal until it is completed or invalidated. After the close fills, reopen when the latest insight still has the pending direction and has not expired, with guards against shutdown, duplicate entries and terminal flattening. Use insight validity rather than exact fill-timestamp equality.

Retain the pending reversal until it is completed or invalidated. After the close fills, reopen when the latest insight still has the pending direction and has not expired, with guards against shutdown, duplicate entries and terminal flattening. Use insight validity rather than exact fill-timestamp equality.

on_position_closed clears pending_direction before requiring event.ts_event to equal the originating insight timestamp exactly. Any closing fill occurring later than the signal timestamp discards the reversal even when the insight remains valid. Consequently, normal execution latency leaves LTC or BCH flat until another daily bar instead of reopening the opposite exposure. Zero-latency sandbox fills do not verify this path.

Outcome Summary

This specific portfolio and execution implementation lacked a profitable edge on the evaluated sample; changing lookback, filters, or prediction horizon would require a new hypothesis.

Backtest review abandoned the strategy for negative expectancy and poor economics; its frozen specification had no tunable parameters. Optimization, post-optimization analysis, and risk review were not reached.

It used one-day historical returns to trade LTC and BCH perpetuals independently on Hyperliquid, with 15% equity allocations per instrument and daily signal renewal, reversal, or expiry exits.

The backtest returned -76.95% with a Sharpe ratio of -1.415 across 1,732 closed positions over 1,633 days. Average net trade return was -0.470%, profit factor was 0.775, and maximum drawdown was 78.56%.

Analysis

HistoricalReturnsAlphaModel