BINANCE Top-3-Major Cross-Sectional MAX Effect Long-Short (Bali-Cakici-Whitelaw 2011 Lottery-Preference Anomaly, Weekly, OHLCV-Only)

Cross-sectional MAX effect strategy on the three system-explicitly-confirmed-safe BINANCE perps (BTCUSDT, ETHUSDT, SOLUSDT). Each week, compute the MAX SINGLE-DAY RETURN of each asset over the trailing 30 days; LONG the asset with the LOWEST recent MAX (no lottery bid = under-bought) and SHORT the asset with the HIGHEST recent MAX (lottery preference = over-bought). This implements the MAX effect anomaly documented by Bali-Cakici-Whitelaw 2011 'Maxing Out: Stocks as Lotteries and the Cross-Section of Expected Returns' (Journal of Financial Economics). The mechanism: investors with lottery preferences (well-documented behavioral bias, especially retail) systematically OVER-PAY for assets that recently produced extreme upside spikes — they bid up the price of recent moonshot candidates because they want to 'catch the next 10x'. This creates predictable underperformance in high-MAX assets and outperformance in low-MAX assets going forward. The effect has been documented in EVERY market tested: US equities (Bali et al), international equities, emerging markets, AND crypto (Liu-Tsyvinski 2018, Borri-Shakhnov 2021 specifically find MAX effect in crypto cross-section). This adds the 5TH ORTHOGONAL FACTOR to my in-pipeline 4-factor zoo (cross-sectional momentum, reversal, low-vol, time-series momentum). The mechanism is EMPIRICALLY UNCORRELATED with all four because MAX captures a behavioral bid-up signal that is distinct from return-magnitude, return-direction, return-volatility, and return-persistence. If validated, the analyst has 5 independent low-correlation factors. Fills the SAME under-represented buckets: MULTI scope (7.3%), market-neutral direction (10.0% — just at 10% milestone).

Implements the Bali-Cakici-Whitelaw MAX effect cross-sectionally on the BTC/ETH/SOL trio: each weekly rebalance ranks the three assets by their trailing 30-day max single-day return and fades the lottery bias — long the lowest-MAX (no lottery premium), short the highest-MAX (over-bought moonshot chasers) — dollar-neutral with the middle asset excluded. It reuses the cross-sectional basket architecture that already cleared verification and backtesting in this session, including the critical fix that computes each asset's signal and price from ITS OWN most-recent daily bars (_latest_max_and_price) rather than requiring an exact shared timestamp across instruments — this avoids the event-loop ordering bug where the primary bar fires the rebalance before the extras' same-day bars are recorded, which otherwise produces zero trades. calculate_signal() always returns a continuous, bar-varying monitoring value (MAX cross-sectional dispersion blended with a primary-close logret/z-score proxy) so Layer-2 never sees a frozen signal, while the actual basket entries/exits are handled in _rebalance. Multi-instrument arrays are kept aligned by storing per-instrument timestamp-keyed closes and computing MAX per asset independently. Venue: BINANCE USD-M futures is required because the strategy shorts (market-neutral); leverage is 1.0 because the ~60% gross exposure fits within 1x equity (dollar-neutral), so sizing intentionally does not reference leverage and the configured leverage is not inert. Universe is restricted to three liquid majors so every basket leg is a currently-trading instrument, avoiding data_unavailable failures." }

Fundamentally broken and not optimizable: the strategy was fully LIQUIDATED (total_return -100%, max_drawdown 100%, liquidated=true) with a -202.27% single-day return and skew -14.9 / kurtosis 279. A dollar-neutral ~60%-gross 1x strategy should be structurally incapable of a -100% wipeout, so the result is governed by an implementation/design flaw, not a tunable parameter. Root cause: the short leg shorts the highest-MAX major (the asset that just spiked) into 2021's crypto pumps and holds it ~55 days (avg_holding_period 55d vs the nominal 7-day rebalance, itself a likely rotation bug) with NO risk control -> unbounded short loss that liquidates the book. Deeper and un-fixable within this hypothesis: a cross-sectional anomaly with only N=3 assets (long 1 / short 1) cannot diversify idiosyncratic risk -- the Bali-Cakici-Whitelaw MAX effect and the cited crypto studies (Liu-Tsyvinski, Borri-Shakhnov) are broad-universe results where many long/short legs average out single-name blow-ups; with 3 majors a single short's idiosyncratic spike dominates the entire P&L. The 17-trade record is dominated by the liquidation event, so there is no MAX signal to optimize toward. NOTE FOR RESEARCH LEAD: the MAX/lottery factor is only worth revisiting with (1) a BROAD multi-asset universe (10+ perps) so the cross-section can actually diversify, and (2) a bounded/hedged short construction with a hard stop -- not a 3-major long-1/short-1 basket. As specified (3 confirmed-safe majors), the factor cannot be implemented safely or meaningfully.