Backtesting Before Real Money: Proving the Edge Exists

The Gate You Can't Skip

You have a rule document. You've run the fabrication audit. You've defined your Do-Not-Trade conditions. Now the question is whether any of it works.

This is the gate that separates traders who build systems from traders who have opinions about systems. The gate is backtesting.

Most traders want to skip this step. The reasons are predictable: backtesting is tedious, the results are uncertain, you can't perfectly replicate historical conditions, and the real money is sitting there waiting to be deployed. These rationalizations are expensive. They're the reason most discretionary traders underperform the strategies they believe in — because they never verified the strategy, they just believed in it.

A rule document without a backtest is a hypothesis. A backtest is the experiment that determines whether the hypothesis is worth trading. Skipping the experiment means running the experiment on your live capital — which has a very different cost structure than running it on historical data.

What Backtesting Measures

A backtest applies your rules to historical data, simulates the trades they would have generated, and produces a performance profile. The profile has several components — each tells you something different about the system.

Win Rate: The percentage of trades that hit their target before their stop. A system with a 60% win rate produces winners 60% of the time, measured at the defined target and stop, not at some arbitrary point.

Average Winner (in R): How many R-multiples does the average winning trade produce? If you risk 1R to make 2R, your average winner is 2.0. This matters as much as win rate — a system with 45% win rate and 3R average winner can be more profitable than a 65% win rate system with 1.2R average winner.

Average Loser (in R): This should be close to -1.0 if your stops are consistent. If your average loser is -1.7R, your stops are being violated — either by slippage, by manual override, or by inconsistent stop placement.

Expectancy: The expected return per trade, calculated as: (Win Rate × Avg Winner) + (Loss Rate × Avg Loser). A positive expectancy means the system makes money over time. A negative expectancy means it doesn't, regardless of how good the entries feel.

Maximum Drawdown: The largest peak-to-trough decline in the account during the backtest period. This tells you the worst period the system has experienced and gives you a baseline for psychological preparation.

Profit Factor: Total gross profit divided by total gross loss. Above 1.0 = profitable. Above 1.5 = solid. Above 2.0 = excellent. Below 1.0 = losing system regardless of win rate.

Sample Size: Why 20 Trades Means Nothing

This is where most backtests fail. A trader extracts 20 historical signals, sees a 70% win rate, and concludes their edge is proven. It isn't. At 20 trades, a 70% win rate is statistically compatible with a system that's actually 50-55% at scale.

The math: with 20 binary outcomes and a true 55% win rate, you'll observe 70%+ approximately 12% of the time. That's a 1-in-8 chance your "70% win rate" is a 55% system you just happened to catch on a good run.

At 100 trades with a true 55% win rate, the probability of observing 70%+ drops below 1%. At 200 trades, it's essentially zero.

The practical implication: extend your historical dataset until you have at least 100 clean signal occurrences. If your system generates 3 signals per week, you need at least 33 weeks of data — and ideally more than a year so the sample spans multiple market regimes.

Regime Coverage: The Test Within the Test

A backtest that only covers a bull market proves you have a bull market strategy. That's not the same as having a strategy.

The minimum regime coverage for a credible backtest:

• At least one sustained bull trend period
• At least one sustained bear trend period
• At least one extended choppy/ranging period
• At least one high-volatility shock event (COVID crash, FTX collapse, LUNA implosion)

If your system's win rate collapses in any of these regimes, that's critical information. It means the strategy is regime-dependent and needs a regime filter to turn itself off in hostile conditions. That's not a failure — that's a finding. You add the regime filter to your DNT conditions and the system improves.

If the win rate is reasonably consistent across regimes, you have a genuinely robust system. Those are rare and worth protecting.

Paper Trading as the Bridge

After backtesting on historical data, the next step is paper trading — forward testing with simulated capital but real-time market conditions.

Paper trading is different from backtesting in an important way: backtesting uses data you can see in full. Paper trading uses data that's arriving in real time, which means you're applying your rules under actual uncertainty — not under the false certainty of hindsight.

The paper trading mandate: run your system for a minimum of 30 trades in live market conditions before committing real capital. This accomplishes several things:

Execution gap identification. The difference between what your rules say and what you actually do in real time. You'll discover you hesitate on certain setups, exit early on others, and miss entries because of latency in the checklist. Paper trading surfaces these gaps safely.

Emotional behavior calibration. Even with simulated capital, the psychological experience of watching a setup move against you before recovering is different from seeing it on historical charts. You learn how you actually behave, not how you think you'll behave.

System edge confirmation. If the paper trading results are roughly consistent with the backtest results, the system is performing as expected. If they diverge significantly, investigate why — the divergence is information about either execution gaps or changed market conditions.

What Good Backtest Results Look Like

A legitimate edge doesn't produce perfect results. If your backtest shows an equity curve that goes up smoothly with no drawdowns, you've made a mistake — either in the methodology or in the data.

Real system characteristics:

• Losing streaks of 4-8 trades are normal at 60% win rate. If your backtest never shows 5 consecutive losses, the historical data might be cherry-picked or you have lookahead bias.
• Monthly losing periods occur even in excellent systems. A system profitable over 12 months might lose 3-4 of those months.
• Drawdown periods of weeks are expected. The 4-6 week drawdown is how you discover whether your system's edge is real or temporary.

The validation framework from the InDecision Framework backtesting methodology is built around honesty with these numbers — including the ugly months, the drawdown periods, and the losing streaks. The 67% historical win rate includes all of them.

The Build Order: Why This Sequence Matters

There's a reason this lesson comes after rule extraction, regime detection, and DNT conditions:

1. Extract rules → define what you're testing
2. Regime detection → know what conditions to test across
3. DNT conditions → define what to exclude from the backtest
4. Backtest → measure the rule set with proper context and filters applied

Running a backtest without the DNT conditions produces inflated results — you're measuring the system without its quality filter. Running it without regime classification produces regime-blind results. The sequence matters because each step's output is an input to the next.

This is the build order for any systematic trading approach. The shortcut of going directly to backtesting without the preceding structure is why most "tested" strategies fail in live trading — the test was run on an incomplete system.

Build the system first. Prove the edge second. Trade it third. In that order, every time.