Performance deep dive
Streaks, drawdowns, profit factor, regime sensitivity, time-of-exit clustering, and autocorrelation across the five profitable strategies. The detail behind the headline numbers.
Strategies analysed on this page
The deep-dive metrics are computed for the five profitable strategy/bucket combinations identified elsewhere on the site:
- Iron Fly V1 at +30m
- Iron Fly V2 at +30m
- Iron Fly V3 at +60m
- Iron Fly V4 at +60m
- Dynamic 0DTE at +30m
1. Streak analysis
How long winning and losing runs typically last for each strategy. Useful for psychological preparation: how many losses in a row should you expect to see?
| Strategy | Avg win streak | Avg loss streak | Longest win | Longest loss | Win streaks | Loss streaks |
|---|
Key takeaway. Iron Fly V1 has never had more than two consecutive losses across 72 trades. Dynamic 0DTE is similar (max 2 losses in a row) and has the longest winning streak in the dataset at 14 trades. Iron Fly V3 and V4 have meaningfully worse loss-streak risk (4–5 in a row at worst), reflecting their fatter loss distribution.
2. Risk-adjusted performance
Total profit alone doesn't tell you how much risk you took to get it. These ratios normalise by volatility and drawdown.
| Strategy | Profit factor | Calmar | Sharpe | Sortino | Skewness | Excess kurtosis |
|---|
What each metric tells you
- Profit factor = gross profit ÷ gross loss. > 2 is solid, > 3 is excellent. Iron Fly V1 leads at 5.94.
- Calmar ratio = total return ÷ |max drawdown|. How much profit per dollar of peak-to-trough drawdown. V1 dominates at 17:1; V3, V4, and Dynamic are all in the 3–4:1 range.
- Sharpe (per-trade) = avg P/L ÷ std dev. All strategies are tightly clustered around 0.20.
- Sortino (per-trade) = avg P/L ÷ downside std. Penalises only volatility on the loss side. V2 leads at 1.18 — its variance is mostly to the upside.
- Skewness measures distribution asymmetry. With the 2026-03-20 outliers removed, skewness compresses substantially across all strategies. Values close to zero mean the per-trade P/L distribution is roughly symmetric.
- Excess kurtosis measures fat-tailness. Normal distributions have kurtosis ≈ 0. Values above 0 indicate fatter tails than a normal distribution — i.e. extreme outcomes are more likely than a Gaussian model would predict.
3. Drawdown distribution
Beyond the worst single drawdown, what does the typical drawdown episode look like?
| Strategy | Episodes | Median DD | 90th-pctile DD | Worst DD | Median length | DDs ≥ $5 | DDs ≥ $10 |
|---|
Iron Fly V1 and V2 have small, frequent drawdowns that recover quickly. V3, V4, and Dynamic 0DTE have fewer drawdown episodes but each one is much deeper. Dynamic 0DTE in particular has only had 3 drawdowns total — but one of them is its current $-27.89 hole. With only 3 episodes, there is very little data to characterise its true tail behaviour.
4. Win rate by IV regime
The CSV records IV Rank (current implied volatility relative to its trailing 1-year range) at trade entry. Splitting trades by regime shows which conditions each strategy was actually exposed to.
Major finding — and a major blind spot. Approximately 95% of all trades fired in the high IV regime (IV Rank ≥ 0.50). The bot has effectively no exposure to mid or low-IV markets in this dataset. This means every conclusion on this site is conditional on "high implied volatility regime" and we have almost no data on how these strategies behave when IV Rank drops below 0.35. If your actual trading environment shifts to a low-IV regime, the historical performance shown here is not reliable guidance.
5. Performance by SPX day-move size
Iron condors and iron flies are nominally "neutral" strategies, expected to perform best on quiet days and worst on big-move days. The data tells a different story.
Counterintuitive finding. The 1.0–2.0% day-move bucket carries the bulk of the per-trade profit across the Iron Fly family — average P/L of $9–15 per trade in that bucket. Quiet days (<0.25% move) are not where the money is made and in some cases (V2) are actually negative. Dynamic 0DTE is dangerous on the 0.5–1.0% bucket specifically: 17% win rate, $-4.31 avg P/L. Its single worst trade (−$15.92) lives there.
6. Time-of-exit clustering
When in the trading day do winners and losers actually close? The pattern is consistent across all five strategies. Iron Fly V1 shown as the representative case.
Iron Fly V1 +30m — exit timing
| Exit window after open | Wins | Losses | Win avg P/L | Loss avg P/L |
|---|
The pattern
- Profit targets fire predominantly in the 30m–2h window after open. This is where most winners close.
- Almost zero losses happen in the first 2 hours. Trades that go bad take time to do so.
- The 3–4h window is where losses cluster — wins and losses are split roughly equally there. This is likely the time-stop firing on positions that haven't reached their profit target.
- Held to settle is a high-variance exit path — low win rate when held (the strategy isn't designed to hold to expiry), but provides a different risk profile than the time-stop bucket.
- 0–30m exits are almost always losses — the rare immediate stop-out.
7. Sequential autocorrelation
Does today's trade predict tomorrow's? If wins or losses cluster sequentially, that information could change sizing decisions.
| Strategy | Trade-level lag-1 corr | Win % after winner | Win % after loser |
|---|
Trades are statistically independent. All correlations are within rounding distance of zero. Win rate after a winner is essentially identical to win rate after a loser. There is no momentum signal in trade history and no mean-reversion signal either. Practical implication: do not size up after wins or sit out after losses based on perceived "streaks". The next trade has the same expected value either way.
8. Day of week × month
The Day of Week page shows that Wednesday is the most consistent day overall. But is that a real cross-temporal pattern, or is it driven by one strong month? This grid shows total P/L per cell so you can see the pattern stability month by month.
Iron Fly V1 +30m
Dynamic 0DTE +30m
Wednesday is the only day-of-week pattern that survives the month-by-month test. Iron Fly V1 Wednesday was positive in every single month traded — never a loss. Dynamic 0DTE's Tuesday and Thursday weakness is concentrated in April — both days were fine in March but delivered meaningful losses in April. With the 2026-03-20 anomalous trades removed, Friday no longer looks dominant in either strategy.
What stands out across all eight analyses
- Almost no data on low-IV regimes. ~95% of trades fired with IV Rank ≥ 0.50. The biggest blind spot in the dataset.
- The strategies make most of their money on bigger directional moves, not on quiet days. The 1–2% move bucket carries per-trade profit. If volatility regime shifts, profit profile may invert.
- Trades are completely independent of each other. No streak signal. Sizing decisions should be based only on prior-trade math, not recent emotional momentum.
- The "held to settle" exit path is a high-variance bucket — low win rate when trades are held to expiry. Profit-targeting before settle is doing the heavy lifting on consistency.
- Wednesday is the only day-of-week pattern that survives the month-by-month test. Friday's apparent strength in earlier versions of this analysis was driven by anomalous trades that have now been removed.
- Iron Fly V2 has a hidden weakness on quiet days (low win rate on <0.25% moves) that the headline win rate hides.