Win Rate Analyzer

There's no universal 'good' win rate - it depends on your payoff ratio. A 35% win rate can be excellent with 3:1 payoff. A 70% win rate can be terrible with 1:4 payoff. Focus on positive expectancy, not win rate alone. That said, most successful traders have win rates between 35-65%. Extremely high (>80%) or low (<30%) win rates are rare and may indicate issues.

Use NET win rate (after all transaction costs) for realistic performance assessment. Gross win rate (before costs) can be misleading, especially for frequent traders or those trading illiquid instruments. The difference can be significant - a trade with a small gross profit might be a net loss after commissions (on options), regulatory fees (SEC/FINRA), and slippage / bid-ask spread.

Minimum 30 trades for basic analysis, 100+ trades for reasonable confidence, 500+ trades for statistical significance of small edges. With only 20 trades, a 60% win rate could easily be luck. With 500 trades, a 55% win rate is likely real. More trades = more confidence in your win rate estimate.

One month is usually too short to draw conclusions. Win rate naturally varies due to market conditions and randomness. Before changing strategy, consider: How does this month compare to long-term average? Has market regime changed? Is sample size sufficient? Check at least 3 months of data before making changes.

Common reasons: 1) Slippage not fully accounted in backtest. 2) Execution delays in live trading. 3) Emotional decisions deviating from system. 4) Market impact when trading real size. 5) Overfitting in backtest design. Expect 5-15% reduction from backtest to live. This is normal.

Use binomial test comparing your win rate to 50% (no edge baseline). Calculate p-value = P(observed or more extreme | true rate = 50%). If p-value < 0.05, your win rate is statistically significant. Online calculators or scipy.stats.binom_test() can compute this. Remember: more trades = easier to achieve significance.

This is valuable information! Calculate conditional win rates for different regimes (trending/ranging, high/low volatility, etc.). Build trading rules: trade full size in high-WR conditions, reduced size or skip trades in low-WR conditions. This adaptive approach can significantly improve overall performance.

Monitor rolling win rate (30-100 trades). Compare recent vs historical performance. Use statistical tests (Chow test) for structural breaks. Warning signs: steady multi-month decline, recent WR significantly below baseline, increased variance. If decay detected, investigate cause (market change, competition, overtrading) and adapt.

Lower win rate = higher variance = more potential for losing streaks = higher risk of ruin if sizing is aggressive. A 40% WR strategy needs smaller position sizes than 60% WR to achieve same risk of ruin. Use Kelly criterion or Monte Carlo simulation to determine safe position sizing for your specific win rate and payoff ratio.

Yes, for detecting changes quickly. Exponentially weighted moving average (EWMA) of trade outcomes gives more weight to recent trades. This is more responsive to regime changes and edge decay. However, also maintain long-term average for baseline comparison. Both perspectives are valuable.

Use Beta-Binomial model. Prior: Beta(α₀, β₀) where α₀/(α₀+β₀) is prior mean. After k wins in n trades, posterior is Beta(α₀+k, β₀+n-k). Posterior mean shrinks toward prior. Use informative prior (e.g., Beta(10,10) for skepticism) to avoid overconfidence in small samples. Posterior gives full distribution, not just point estimate.

Features: market conditions (VIX, trend), technicals (RSI, volume), timing (day, hour), strategy-specific. Use logistic regression or gradient boosting (XGBoost). Critical: time-series train/test split (not random), proper calibration (Platt scaling), validation of probability estimates. Use predicted probability for position sizing and trade selection.

Uncorrelated trades: portfolio (day/week) WR > individual trade WR due to diversification. Correlated trades: portfolio WR ≈ individual WR (no benefit). Calculate: portfolio daily WR depends on number of trades, individual WR, and correlation structure. Use simulation to estimate if analytical solution is complex.

Bonferroni: divide significance level by number of tests (α' = α/n). Conservative but simple. Benjamini-Hochberg: controls False Discovery Rate, less conservative. Use when testing many segments (strategies, instruments, time periods). Without correction, 5% of null segments will appear significant by chance.

Use regression-based attribution: regress win/loss outcomes on potential factors (entry signal, market condition, timing, etc.). Coefficients show marginal contribution. Alternatively, use decision trees to identify which factors most strongly predict wins. Compare WR in presence vs absence of each factor to quantify contribution.