Strategies

It reads RSI, MACD, Bollinger Bands, and Volume on each candle. RSI below 30 signals oversold (a long bias) and above 70 overbought (a short bias); MACD crossing its signal line indicates momentum direction; Bollinger %B near the lower band is oversold and near the upper band overbought; a volume spike adds confirmation. When at least 2 of the 4 agree, it opens a position, and closes when RSI returns toward neutral (around 50) or a stop-loss is hit.

When the fast EMA(21) crosses above the slow EMA(55) it's a long signal; crossing below is a short signal. A higher-timeframe filter only allows longs when price is above EMA(200) and shorts when below, so trades go with the broader trend. It exits on the opposite crossover.

lsr_agg_long_frac (dirBearHigh) measures what fraction of CEX accounts are net long. When the crowd is positioned most short (signal below the 20th percentile), a structural bid from AF/staking tends to push HYPE up — the strategy goes long. When the crowd is most long (above the 80th percentile), that bid is absorbed and the strategy goes short. Position closes when the signal reverts through the neutral midpoint or after maxHoldBars bars.

No rolling window needed: thresholds are absolute, derived from params. Enter short when funding ≥ defaultRate × entryShortMultiple (longs overpaying carry). Enter long when funding ≤ −(defaultRate × entryLongMultiple) (longs paying shorts at k× default magnitude). The symmetric multiple design filters shallow-negative noise (e.g. −1e-6 ≈ 0.08× default) that any-negative-funding rules misfire on. Exit when |funding − defaultRate| < exitBand (normalised), or after maxHoldBars bars, or on the engine stop-loss. Because ≥80% of funding values sit at exactly the default, percentile-based entries are degenerate (p10 = p90); this strategy uses absolute deviation instead.

Each hour the strategy reads the aggregate long fraction (lsr_agg_long_frac) — the share of accounts at major CEXs that are net long HYPE. It computes the 20th and 80th percentiles from the available signal window. When the crowd is extremely short (signal below the 20th percentile), a structural bid from the Alligator Fund and staking rewards tends to push price up against the crowd: the strategy goes long. When the crowd is extremely long (above the 80th percentile), it shorts, fading the crowded side. The position exits when positioning reverts past the median (50th percentile), or unconditionally after a configurable max-hold cap to limit exposure in persistent positioning regimes.

It measures realised volatility on a rolling window and compares to a 30-day baseline. In low-vol periods (range-bound markets), it allocates more capital to mean-reversion logic; in high-vol periods (trending markets), it shifts to trend-following. Rather than a binary regime toggle, the allocation is weighted continuously by the volatility percentile, avoiding whipsaws at regime boundaries.

When Hyperliquid's liquidation volume spikes far above its rolling baseline, the strategy waits for the first candle close that prints opposite the cascade direction (a confirmation candle), then enters a counter-trend position with a tight stop. The confirmation filter sacrifices the first part of the bounce in exchange for not catching falling knives mid-cascade.

Hyperliquid publishes both the current funding rate and a predicted next rate. Persistent positive deviation (actual exceeds predicted) signals fresh long crowding — historically a precursor to reversals. Negative deviation signals short crowding. The raw deviation is normalised as a 7-day z-score so the signal is scale-invariant across calm and volatile funding regimes.