Trend-Following Backtester · Guide · Concepts 한국어
Concepts
What is backtesting? How simulated trading works and when not to believe it
Backtesting answers “what if I had followed this rule for the last five years?” It sounds simple. There are more ways to inflate the answer than to get it right.
How it works: walk forward one day at a time
Start at the first day of history and step forward. On each day, evaluate the rule using only data available up to that day, act on the result, and accumulate the balance.
That constraint is the whole game. If tomorrow's price leaks into today's decision, the backtest is void — you were trading with knowledge of the future, and the result will not reproduce.
Four numbers to read
- CAGR — annualised return. Comparable across strategies with different run lengths.
- Max drawdown — the deepest fall from a peak. This, not CAGR, is what makes people quit.
- Sharpe — return per unit of volatility. Same return, steadier path, higher number.
- Trade count — how many times costs are charged. This is where backtested edges quietly die.
Ethereum, 2017-09-25 to 2026-07-30. Buy and hold returned 27.1% CAGR with a -95.6% drawdown.
| Strategy | CAGR | Max DD | Sharpe | Trades |
|---|---|---|---|---|
| Supertrend | 97.2% | -48.5% | 1.54 | 93 |
| Parabolic SAR | 95.0% | -47.8% | 1.54 | 149 |
| Bollinger band breakout | 94.8% | -48.6% | 1.56 | 71 |
| SMA crossover | 89.6% | -49.0% | 1.42 | 107 |
| EMA crossover | 85.9% | -51.4% | 1.39 | 121 |
Five ways results get inflated
1. Look-ahead bias
Unconfirmed information entering the decision. The most common and most fatal.
2. Survivorship bias
Testing only on what still exists today. Drop the delisted names and every strategy looks good.
3. Ignoring costs
Worse the more often you trade. This site charges 0.5% round trip by default.
4. Overfitting
Fitting parameters to the past. Measured directly in optimization and overfitting.
5. Cherry-picking
Showing only the windows that worked. Which is why the asset pages here publish the unflattering results too.