5.4 Variance and Sample Size

Understand why your first 20 trades tell you almost nothing about whether your strategy works, and why traders who change strategies after 5 losses are guaranteed to fail.

Layer 5: The Meta Game — Chapter 4 Goal: Understand why your first 20 trades tell you almost nothing about whether your strategy works, and why traders who change strategies after 5 losses are guaranteed to fail.

The Core Idea

Trading outcomes are a mix of skill and luck. In the short run, luck dominates. In the long run, skill emerges.

A losing strategy can win 10 in a row. A winning strategy can lose 10 in a row. If you change strategies based on the last 10 trades, you are surfing noise, not signal.

The number of trades it takes to reliably distinguish skill from luck is far larger than you intuit. Most retail traders quit, switch, or revamp their strategy long before the math has had time to express itself.

The Coin Flip Analogy

Take a fair coin (50/50). Flip it 10 times. What's the chance of getting 7 or more heads?

About 17%. Not crazy. Now imagine you've designed a "trading system" that won 7 out of 10 trades. How confident should you be in your edge?

Almost none. A 70% win rate over 10 trades is well within the noise band of a 50/50 coin flip. You have no evidence of edge yet.

Flip it 100 times

Probability of 70+ heads on a fair coin: about 0.004%. Now if you got 70 wins out of 100 trades, you have strong evidence of an edge.

Sample size is the only way to separate luck from skill.

The Standard Error of Win Rate

The statistical formula for how uncertain your measured win rate is:

Standard Error = sqrt(p × (1-p) / n)

Where p = win rate, n = number of trades.

Example: 50% win rate, 20 trades

SE = sqrt(0.5 × 0.5 / 20) = sqrt(0.0125) ≈ 0.112 = 11.2%

That means a 95% confidence interval for your "true" win rate is roughly:

Measured Rate ± 2 × SE = 50% ± 22.4% = [27.6%, 72.4%]

After 20 trades with a 50% win rate, your actual win rate could be anywhere from ~28% (terrible) to ~72% (elite). You have learned essentially nothing.

Example: 50% win rate, 100 trades

SE = sqrt(0.5 × 0.5 / 100) = 0.05 = 5%
95% CI = 50% ± 10% = [40%, 60%]

Still wide. You've narrowed it but you don't know if you're a 40% trader or a 60% trader.

Example: 50% win rate, 500 trades

SE = sqrt(0.5 × 0.5 / 500) ≈ 0.022 = 2.2%
95% CI = 50% ± 4.4% = [45.6%, 54.4%]

Now you have a real signal. It took 500 trades to confirm a 10% confidence band.

Translation to Real Life

You will not have statistical certainty about your edge for at least 100-200 trades.

For a swing trader taking ~3-5 trades per week, that's 6-12 months. For a swing trader taking 1-2 trades per week, that's 1-2 years.

During that time you must trust the process, not the outcomes. The math is brutal: if you change strategies every 20 trades because you "didn't see results," you'll never give any strategy enough sample size to prove itself.

Streak Math (Why Bad Streaks Are Normal)

The probability of a losing streak of length k in a row, given win rate p:

P(streak = k) ≈ (1-p)^k

But the probability of seeing at least one streak of length k over n trades is much higher:

P(at least one streak of length k in n trades) ≈ 1 - (1 - (1-p)^k)^n

Example: 50% win rate, 100 trades

Streak length	Probability of seeing it at least once in 100 trades
4 losses in a row	~95% (near certain)
5 losses in a row	~85%
6 losses in a row	~62%
7 losses in a row	~41%
8 losses in a row	~25%
10 losses in a row	~9%

A 6-loss streak in 100 trades has 62% probability of happening to a 50% win rate trader. It's not a sign you've lost your edge. It's expected noise.

For a 40% win rate trader (which is fine if your R:R is 2:1 or better)

Streak length	Probability in 100 trades
5 losses	~99%
7 losses	~83%
10 losses	~37%

A 7-loss streak is essentially guaranteed for a 40% win rate trader over 100 trades. Plan for it.

P&L Variance, Not Just Win Rate Variance

Win rate variance is one source of noise. P&L variance is bigger.

A 50/50 strategy with 2R wins and 1R losses can have a wildly varying month-to-month P&L based purely on randomness, even with a real positive edge.

Simulated example

Win rate: 50%
Win: +2R = +$200 (with $100 risk)
Loss: -1R = -$100
Expectancy: +$50 per trade
Trades per month: 20

Over 100 simulated months:

Best month: +$2,500 (highly profitable)
Worst month: -$700 (clearly losing)
Average month: +$1,000 (matches expected value)
Months with negative P&L: ~15%

Even with a clear positive edge, 1 in 6 months will be a losing month. A trader who panics and changes strategy after one losing month will never compound an edge.

The Monte Carlo Simulation

A Monte Carlo simulation runs your strategy thousands of times against random sequences to map the distribution of possible outcomes.

A simple version, conceptually:

Define your strategy parameters: win rate, R:R, trades per period.
Generate 10,000 random sequences of trades.
Plot the resulting account equity curves.
Look at the worst 5% (your "tail risk") and the median.

What you'll see

Even strategies with clear positive edge have wide P&L distributions
Many "winning strategies" have a 5% chance of significant drawdown
Some "losing strategies" still produce 30 months of winning streaks before reverting

You can build a Monte Carlo in Excel with RAND() or in Python with numpy.random. Or use online calculators. Run one for your strategy before you trade it live.

The Backtester's Illusion

A common trap: you backtest a strategy on 5 years of data, see it returned 80% with a 50% win rate, and conclude you have a great edge.

But:

That 5 years is one sample path out of millions of possible histories
The strategy might have a much wider distribution of outcomes than that one path shows
If you change one parameter and the result changes wildly, you've curve-fit to noise

One historical run is one data point. It doesn't tell you the distribution. Always Monte Carlo your strategy before trusting one backtest.

How Many Trades Until You Can Trust Your Edge?

Approximate guideline:

Sample Size	Confidence in Win Rate (assuming 50% true rate)
10 trades	None. Pure noise.
20 trades	Still mostly noise.
50 trades	Beginning to see a signal. ±14% confidence band.
100 trades	Useful but still noisy. ±10% band.
200 trades	Reasonably reliable for win rate. ±7% band.
500 trades	Strong confidence. ±4.4% band.
1,000 trades	High confidence. ±3% band.

For a typical swing trader at 3-5 trades/week, 100 trades = 5-8 months of consistent trading. Don't make major strategy changes before you have 100 trades.

The Behavioral Trap

Here is what destroys most retail traders:

Backtest a strategy. Looks good.
Trade it live for 10 trades. Loses 6 of them. Down 5%.
Conclude "this doesn't work in this market."
Tweak strategy. Trade new version 10 trades. Loses 5.
Conclude "still not working."
Switch to entirely new approach. Trade 10 trades. Loses 6.
Conclude "trading is impossible."

What actually happened: each of those strategies might have been profitable. They never got enough sample size to express their edge. The trader kept paying for "first 10 trades" of strategies and never got to see the long-run mean.

The fix

Predefine your evaluation period: 50 trades minimum, 100 preferred.
Predefine your kill criteria: 25% drawdown OR clear violation of strategy thesis.
Trade through normal variance. Trust the math.

Common Mistakes

Declaring victory or defeat after 10 trades. Pure noise.
Comparing your performance to YouTube traders who claim 90% win rates. They're lying, lucky, or showing one week. The math doesn't support sustained 90% over hundreds of trades.
Changing position size based on recent results. "Hot hand" thinking. Each trade is independent.
Adding new indicators after losses. Looking for a "fix" because variance hit. Your strategy isn't broken; you got unlucky.
Abandoning during normal drawdown. A 15% drawdown is uncomfortable but expected. Predefine when you'll stop. Don't decide in the moment.
Believing in "this market is different." Sometimes it is. Usually it isn't. Always reason from data, not feeling.
Sample sizing the wrong thing. "I've traded 6 months." Total time matters less than total trades. 6 months of 1 trade/week = 26 trades = noise.

A Mental Model: The Sports Analogy

A baseball hitter with a true talent of .300 (30% hits) will have weeks where they hit .150 and weeks where they hit .450. Over a season (600 at-bats), their true talent emerges.

If a coach benched a .300 hitter after 10 bad at-bats, that coach would never have any players. Real coaches let their hitters work through slumps because they know the math.

You are the coach AND the hitter. Don't bench your strategy after 10 trades. Don't fall in love with it after 10 wins either. Wait for the sample size.

Practical Takeaways

Don't judge a strategy on fewer than 50 trades. 100 is better.
Plan for losing streaks of 5-8 trades. They're statistically normal.
Plan for losing months. ~15-20% of months will be losing months even with a real edge.
Run a Monte Carlo on your strategy before going live. Understand the distribution, not just the mean.
Build a stat-tracking spreadsheet from trade #1. Win rate, average win, average loss, P&L per month. Update after every trade.
Don't change strategy parameters mid-stream. Let it run for the predefined sample.
Don't increase size on a "hot streak." Don't decrease size on a cold streak that's within expected variance.
Define your kill criteria in advance. "I'll stop if drawdown exceeds 20%" or "if win rate drops below 35% over 100 trades." Decide while calm.
Compare your stats to your backtest, not to your feelings. If live matches backtest distribution, your strategy is working as designed even if recent P&L is bad.
Time is on your side IF you survive the early sample. Position size conservatively so variance doesn't ruin you.

Quick Self-Check

I understand why 10 trades tells me nothing
I can explain the difference between win rate variance and P&L variance
I know that a 5-7 loss streak is normal even for winning strategies
I have a predefined sample size (50+, ideally 100+) before I judge a strategy
I have predefined kill criteria for when to stop trading a strategy
I know to run a Monte Carlo before going live with a backtested strategy
I will not increase size on hot streaks or decrease on cold streaks within normal variance
I understand that losing months are expected, not abnormal

Previous: 5.3 Position Sizing Next: 5.5 Backtesting Properly