"Sell in May." "Uptober." "Sunday dumps." Crypto Twitter repeats calendar folklore every year like it's settled science, but almost nobody who shares these claims has actually tested them against multi-year OHLCV data with a defined hypothesis and an out-of-sample split. Most of what passes for seasonality analysis is a chart screenshot from one specific year, stretched into a rule.
A losing manual trade costs you one bad decision. A losing algorithm costs you the same bad decision, repeated automatically, at every signal, for as long as it stays deployed. That asymmetry is why the mistakes below matter more in algorithmic trading than in discretionary trading — a bot doesn't hesitate, doesn't have a bad feeling about the setup, and doesn't stop until you tell it to.
Most indicators do one job — RSI measures momentum, ATR measures volatility, ADX measures trend strength. Parabolic SAR does two at once: it generates entry/exit signals when its dots flip sides, and while a trend is intact, those same dots trail behind price as an accelerating stop-loss. J. Welles Wilder — the same analyst behind RSI and ATR — designed it to answer one question: "if I'm in this trade, where's the level that says I'm wrong?"
Most crypto portfolios are diversified in name only. A trader runs five bots across BTC, ETH, SOL, and a handful of large-cap alts and calls it "spread risk" — but if all five pairs fall 20% in the same afternoon, there was never any real diversification to begin with. The number that would have told you this in advance is correlation, and almost nobody checks it before allocating capital.
Every strategy covered so far on this blog — RSI mean reversion, EMA crossovers, breakouts — shares one assumption: the asset you're trading will move in a predictable direction. Pairs trading drops that assumption entirely. Instead of betting BTC goes up or ETH goes down, you bet on the relationship between the two — and that relationship can be far more stable than either asset's price on its own.
On March 11, 2023, USDC traded as low as $0.87 on some venues after Circle disclosed that $3.3 billion of its reserves sat at the failed Silicon Valley Bank. It recovered within 48 hours once federal regulators guaranteed deposits — but for two days, every bot quoting trades in USDC was pricing its own P&L against a moving target. If your bot held a "flat" position in USDC during that window, you weren't flat. You were long a stablecoin that had just lost 13% of its dollar value.
A backtest that only shows in-sample performance is a strategy that hasn't been tested yet — it's been fitted. The gap between "looks great on the data I optimized against" and "still works on data it never saw" is where most algorithmic strategies quietly die. VolatiCloud's backtest detail page now scores that gap directly: a 0-100 overfit-risk score, a Low/Moderate/High band, and the specific factors that produced it, computed automatically whenever you run an out-of-sample split or a 3-window walk-forward validation.
Every backtest report leads with total profit, and total profit is the number least worth trusting. Two strategies can post the same +60% over two years — one on a steady climb, the other spending fourteen months underwater before a single lucky quarter. Reading a backtest well means knowing which metrics carry real information, which ones routinely mislead, and in what order to check them.
A 0.1% taker fee sounds trivial. On a $1,000 position, that's $1 out the door. Easy to dismiss. But a bot cycling in and out of positions 500 times a year pays $500 in fees on that same $1,000 — a full 50% drag before the market has moved a single tick in your favor. Fee optimization is one of the fastest ways to improve a live strategy's returns without touching the signal logic.
Your backtest just returned a 47% annual return with a Sharpe ratio of 1.8. Before you fund the bot, ask one question: how many trades is that result based on? If the answer is fewer than 100, the number might mean nothing.