High-Volatility-Quintile 6-Month Momentum with Jegadeesh-Titman Overlapping Cohorts, Long-Short, Point-in-Time Top-60 Binance USD-M Perp Universe (DAILY bars, monthly cohort formation, 6-month cohort hold)

Hypothesis

Restated from first principles for crypto perps, not ported line by line, from QuantConnect Strategy Library #155 'Momentum and Reversal Combined with Volatility Effect in Stocks'. Kept from the source: momentum is concentrated in the highest-volatility names. Each month, rank a point-in-time top-60 universe of Binance USD-M perps by 30-day dollar volume. Keep the top vol quintile (12 names), using realized vol over days t-187..t-7 (the last week is skipped, as in the source). Rank those 12 by 180-day return. Go long the top 3 and short the bottom 3. Each monthly cohort gets 1/6 of gross, is held 6 months, and up to 6 cohorts overlap (Jegadeesh-Titman style). Book gross is 1.0x. Positions change only at month boundaries, and expected time-in-market is ~90% once all cohorts are live. The long-only half is pre-registered as a separately scored variant 2. The full text is in the outbox file.

Iteration 3 fixes the Layer-3 sandbox timeout. (1) Per-leg history moved from deques to plain lists with amortised chunk trimming. The old code sliced a deque (closes[a:b]), which raises TypeError at every rebalance attempt, and it ran bisect over deque indexing (O(n) in the middle). Lists make both O(1)/O(log n). (2) The instrument set is cut to 40 liquid, currently listed majors, all confirmed in the catalog with 1-DAY bars. That is ~96k sandbox bars instead of ~144k+ for a 60-leg book, and per-bar engine cost is the (c) per-span cost from L155. The top quintile of 40 is 8 names, so long 3 / short 3 is preserved. universe_size=60 can be restored when data and wall time allow. Signal logic, cohort math, the calendar-anchored scalar state, the L139 cutoff alignment rule (only bars with ts <= month-start cutoff, strictly older than the current primary bar) and 1.0x gross sizing are unchanged. Max gross = 1.0x equity, far under the 20x cap. Futures venue because the book shorts.

verification_loop: sandbox_timeout: Verification failed (Layer 3 — sandbox backtest): smoke test exceeded the 300s wall-clock limit on a ~20000-bar window (2400 days of BTCUSDT.BINANCE-1-DAY-LAST-EXTERNAL). This almost always means per-bar work that scales with history — e.g. rescanning the full funding/supplementary series, or rebuilding a list and calling min()/sorted() inside calculate_signal()/on_bar() on every bar. Precompute sorted timestamp arrays ONCE in __init__ and use bisect, or cache lookups keyed by timestamp, so per-bar cost is O(log n) not O(n).

Implementation

Each calendar month, rank a point-in-time universe of liquid Binance USD-M perps (candidate list of 40) by 30-day dollar volume. Keep the top vol quintile, using realized vol over days t-187..t-7. Rank that bucket by 180-day return: long the top 3, short the bottom 3. Each monthly cohort carries 1/6 of 1.0x gross and is held 6 months, with up to 6 cohorts overlapping (Jegadeesh-Titman). The book is netted per name and traded only at month boundaries. Daily bars.

Verification Results

sandbox_timeout: Verification failed (Layer 3 — sandbox backtest): smoke test exceeded the 300s wall-clock limit on a ~20000-bar window (2400 days of BTCUSDT.BINANCE-1-DAY-LAST-EXTERNAL). This almost always means per-bar work that scales with history — e.g. rescanning the full funding/supplementary series, or rebuilding a list and calling min()/sorted() inside calculate_signal()/on_bar() on every bar. Precompute sorted timestamp arrays ONCE in __init__ and use bisect, or cache lookups keyed by timestamp, so per-bar cost is O(log n) not O(n).

Iteration History

Verification failed (Layer 1 — static analysis): - Syntax error: invalid decimal literal (<unknown>, line 1) (line 1)

Outcome Summary

In multi-leg universe strategies with long lookbacks, precompute or cache cohort and ranking state and only recompute at month boundaries, so per-bar cost stays O(1) or O(log n) and the smoke test fits within the 300s sandbox limit.

After 3 iterations it was abandoned in a verification loop. The Layer 3 sandbox backtest went past the 300s wall-clock limit on a ~20,000-bar window (2,400 days of BTCUSDT.BINANCE 1-day bars), which points to per-bar work that grows with history length. Backtest review, analyst and risk stages never ran.

A long-short crypto adaptation of volatility-conditioned momentum: each month, take the highest-volatility quintile of a point-in-time top-60 Binance USD-M perp universe (ranked by 30-day dollar volume), rank those names by 180-day return, then go long the top 3 and short the bottom 3 in overlapping Jegadeesh-Titman monthly cohorts, each held 6 months at 1/6 of a 1.0x gross book.

There are no performance numbers: the strategy never finished verification, so no backtest report, optimization report, return, Sharpe or trade count exists.