Nobody would board a plane whose design had never flown. Yet most crypto bots are funded exactly that way — a strategy from a screenshot, a backtest with no fees, and hope. Backtesting is how you rehearse a strategy on history before it meets reality. Done honestly, it will not tell you what a bot will earn. It will tell you whether the idea survives contact with costs, volatility and your own worst assumptions.
What a backtest is — and is not
A backtest replays your exact rules — entries, exits, sizing, filters — on historical price data and records what would have happened. That makes it a filter, not a prophecy. It can disqualify a bad strategy quickly and cheaply, and it can rank several decent ones. What it cannot do is promise the future, because the future is not the past: regimes change, liquidity shifts, and every other market participant adapts. Treat a backtest the way a pilot treats a simulator: essential, and not the sky.
The four numbers that matter
Equity curves are marketing. These are diagnostics:
| Metric | What it tells you | What healthy looks like |
|---|---|---|
| Max drawdown | Worst peak-to-trough loss | Small enough that you would not disable the bot mid-way through it |
| Win rate | Share of trades closed in profit | Meaningless alone — read it with the next two |
| Profit factor | Gross gains ÷ gross losses | Above ~1.5 after costs; below 1.2 is noise |
| Expectancy | Average € per trade after costs | Positive, and stable across sub-periods |
A 94% win rate with tiny winners and rare catastrophic losers is a ticking premium-selling machine. A 35% win rate with strong expectancy can be excellent. Read the four together, in that order of priority: drawdown first, because everything else is theoretical if you cannot hold through the worst case.
Data quality decides everything
A backtest is only as honest as its data. Watch for:
- Missing candles and gaps — the strategy cannot trade through holes it never sees, and real exchanges have them.
- Wrong granularity — a strategy tuned on hourly closes may behave differently when intraday swings are visible.
- Survivorship bias — testing only on assets that still exist flatters every strategy. Include the dead ones.
- Short windows — a great 2021 backtest says nothing about 2022. Cover bull, bear and chop.
Fees and slippage: where paper profits go to die
The single most common backtest fraud is omission of costs. A grid bot completing 500 cycles at a 0.8% gross cell pays 0.1% per side in realistic fees — that is 0.2% of every cycle, a quarter of the profit, gone. Add slippage on entries and exits, and “consistent daily returns” routinely become consistent daily bleed. Any backtest you evaluate should state its fee and slippage assumptions explicitly, and they should mirror reality. If a vendor’s backtest has no fee line, the backtest is an advertisement.
Overfitting: designing a perfect past
Given enough free parameters, any strategy can be tuned to fit history perfectly and trade the future terribly. The warning signs: dozens of optimized inputs, gorgeous backtest, no out-of-sample results. The discipline: keep rules simple, freeze parameters, then validate on data the optimizer never saw — a walk-forward test where each segment is tuned on the past and scored on the future. If performance collapses out-of-sample, the strategy was memorizing, not generalizing.
A practical rule of thumb: a strategy you cannot explain to a competent stranger in two minutes has too many moving parts to trust with capital. Complexity is where overfitting hides, and simplicity is the only reliable defense that does not require you to outsmart your own optimizer.
A backtest is a rumor about the future, not a promise. Your job is to check the rumor’s sources.
From backtest to live, in order
- Backtest on long, complete data with honest costs. Kill what fails here.
- Paper trade the survivors on live prices. This catches implementation bugs no history file can.
- Trade small — a size where mistakes are tuition, not trauma.
- Review on a schedule against the backtest’s own drawdown and expectancy bands, not against your feelings.
How we backtest at NexoBot
Every NexoBot strategy — from DCA to momentum — ships with its history replayed against five years of market data, with a 0.1% fee per side — the same fee the live engine applies — and the results are presented as simulations with their drawdowns visible, not screenshots of winners. Before funding anything, you can run the same configurations yourself in the demo: $10,000 in virtual funds, live market prices, and the full reporting of trades, fees and equity. Simulated results, honestly labeled, are the only kind worth trusting.
The bottom line
Backtesting will not find you a perfect strategy — there is not one. It will find the strategies that are dead on arrival, quantify the pain you are signing up for, and force every assumption through a fee schedule before it can cost you money. Run the numbers honestly, respect the drawdown, and let the simulator do the dying for you.
Лише освітній контент — не є фінансовою, податковою чи юридичною порадою. Минулі результати, реальні чи історичні, не гарантують майбутніх.
Застосуйте це на практиці
Протестуйте ці стратегії на демо-рахунку — 10 000 віртуальних USDT на реальних ринкових даних.