Signal over speed.
HFT needs co-location and microseconds. Mid-frequency crypto — minutes to days — is playable from a laptop. What matters is signal quality and risk control, not latency. The sweet spot for predictive edge sits at a 0.2–0.4 hit-rate, not the highest-probability bets.
Three strategies that survive contact with reality.
The trend is your friend.
Short SMA crosses long SMA — go with the direction. Rebalance daily or weekly on hourly bars of liquid pairs like BTC/USDT and ETH/USDT. Boring, cheap, works when markets move.
What stretches, snaps back.
Bollinger touch on the lower band → buy; upper band → sell. Best in ranging regimes, worst in one-way trends — so gate it with a regime filter.
Bet the relationship, not the direction.
Find two cointegrated assets (BTC/ETH, SOL/AVAX). Model the spread, z-score it, short the winner and long the loser when it stretches past ±2. Close near zero. Cointegration — not correlation — is the test that matters (ADF on the spread).
| Strategy | Net Sharpe | Ann. return | Turnover / yr |
|---|---|---|---|
| Short-term reversal | 1.77 | +36.6% | ~89x |
| Long-term momentum | 1.32 | +29.1% | ~18x |
| Mixed (equal-vol) | 2.52 | +40.1% | — |
Pairs trading, live.
A synthetic BTC/ETH spread with rolling z-score. Move the entry threshold — watch trade count and simulated PnL respond. Same mechanic Intellitrade runs on real pairs.
The four things that kill quant books.
Relationships break.
In crashes everything correlates to 1. A stat-arb book without a kill switch is a directional bet in disguise.
Parameters are fragile.
Lookbacks and thresholds swing PnL. Walk-forward validate. Refuse a strategy that only works on one seed.
Costs eat everything.
A +21% strategy at zero fees turned −14% with 2 bps maker fees. Model realistic costs before you get excited.
Concentration cuts both ways.
Concentrated strategies beat diversified ones 4× — and blow up harder. Size like you might be wrong.
Run these signals, not screenshots.
Intellitrade wires crypto data, cointegration tests, and analyst books into one terminal. Paper-trade for six months, then keep what earned its cost.