I Ran an Automated SOXL Trading Bot for 7 Months ??Here's Every Trade, Win and Loss
I Ran an Automated SOXL Trading Bot for 7 Months ??Here's Every Trade, Win and Loss
Most articles about leveraged ETF trading strategies share backtests. Backtests are useful, but they hide something: they never suffer the emotional cost of watching a 3x leveraged fund drop 15% in a single overnight session, and they quietly assume every fill happens at the perfect price. This is not a backtest. This is a log of what actually happened when I let an automated bot trade SOXL ??the Direxion Daily Semiconductor Bull 3x ETF ??with real money for roughly seven months.
Below is the full record: how the strategy was built, all 18 completed trades, the numbers that came out the other side, and the specific mistakes the data exposed. My hope is that it gives you something the usual "top 5 strategies" list can't: honest, position-by-position evidence of what a rules-based system does inside one of the most volatile products retail investors can buy.
Disclosure: This is a personal experiment run on a small live account, shared for educational purposes. It is not investment advice. Leveraged ETFs like SOXL carry extreme risk and can lose value rapidly. Past results do not predict future performance.
Why SOXL, and Why a Bot
SOXL aims to deliver three times the daily return of the NYSE Semiconductor Index. On a good day, that's exhilarating. On a bad one, it's brutal ??and because of the way daily compounding works, a choppy sideways market slowly bleeds a 3x ETF even when the underlying index ends flat. This is exactly the kind of instrument where human emotion does the most damage: people hold losers too long hoping to break even, and sell winners too early out of fear.
A bot doesn't feel any of that. It follows the same rules on trade 1 and trade 50. That consistency was the entire point of the experiment. The strategy itself was intentionally simple:
- Entry: buy on pullbacks during uptrends, using RSI to time dips (typically RSI in the high-40s to low-50s after a retracement).
- Scaling: add to the position in tranches rather than committing all capital at once.
- Exit: a combination of a trailing stop (lock in gains once price falls a set percentage from its peak), an RSI-based profit-take in strongly overbought conditions (around RSI 80), and a hard stop-loss near -15% as a final backstop.
Everything ran on a schedule against a live brokerage API. No discretionary overrides once a rule triggered.
The Headline Numbers
Across the full period, the bot completed 18 round-trip trades. Here's what those 18 trades produced:
| Metric | Result |
|---|---|
| Completed trades (round trips) | 18 |
| Wins / Losses | 11 / 7 |
| Win rate | 61.1% |
| Average winning trade | +23.19% |
| Average losing trade | -12.10% |
| Reward-to-risk ratio | 1.92 |
| Best single trade | +76.94% |
| Worst single trade | -15.47% |
Two things jump out. First, the win rate of 61% is solid but not spectacular ??this was never a strategy that's "right" almost every time. Second, and more importantly, the reward-to-risk ratio of 1.92 is what carried the results: winners were, on average, nearly twice the size of losers. In a system like this, that asymmetry matters far more than raw accuracy. You can lose 40% of your trades and still come out ahead if your winners are big enough and your losers are cut short.
The Trades That Mattered
Rather than dump all 18 rows, it's more useful to look at the trades that shaped the outcome ??because the distribution was highly uneven.
The big winners
Three trades did most of the heavy lifting, all of them exits triggered by the RSI overbought profit-take in a strong uptrend:
- +76.94% ??the single best round, closed when RSI pushed into extreme overbought territory during a powerful rally.
- +47.03% and +44.97% ??two more RSI-80 profit-takes during the same broad uptrend.
There was also a +32.60% trailing-stop exit early in the period and a +27.74% exit that fired as the bot escaped a weakening market. These five trades alone defined the profitable side of the ledger.
The losers
The losses clustered in a recognizable and, in hindsight, avoidable pattern. Several came from buying near local tops and then getting stopped out as price rolled over:
- -15.47% ??hit the hard stop-loss.
- -14.97%, -14.52%, -13.99%, -13.51% ??a run of trailing-stop exits after entries that were, frankly, too close to a peak.
- -11.76% ??another trailing stop after a high-level entry.
Notice something: no single loss exceeded 15.5%. That's the stop-loss doing ??job. The system was never going to hand back a full -50% leveraged drawdown, because the exit rules force it out well before that. The losses were annoying but bounded.
The Uncomfortable Lesson: Where the Bot Failed
The most valuable insight from seven months of real data isn't the +77% trade. It's the failure pattern. When I grouped the losing trades, the story was unambiguous: the bot's worst results came from entries made near price peaks. The early rounds bought around a high of roughly $70 and produced both a +32.6% win and, in nearby attempts, -13.5% and -14.5% losses. Later, a cluster of entries near the $213??229 zone produced three straight losses of roughly -12% to -15%.
In other words, the entry logic ??"buy the pullback in an uptrend" ??worked beautifully when the uptrend was genuine and early, and failed when it was chasing a move that had already run too far. The exit rules protected the downside, but they couldn't fix a bad entry. That's a real, specific weakness, and it's the kind of thing you only learn by watching actual fills rather than a smoothed backtest curve.
The trailing stop, in particular, finished roughly break-even on a win/loss count ??it saved profits in some rounds and got whipsawed out in others. The clearly profitable exits were the RSI-based profit-takes in strong trends. If I were rebuilding the system today, that's where I'd focus: making the entry filter stricter in extended markets, and leaning more on trend-confirmed exits than on trailing stops alone.
What This Actually Tells a Long-Term Investor
If you're a buy-and-hold dividend investor, the takeaway isn't "go build a SOXL bot." It's something more general and more useful:
- Asymmetry beats accuracy. A 61% win rate paired with a ~1.9 reward-to-risk ratio does more work than chasing a higher hit rate. The same logic applies to a long-term portfolio: cutting losers and letting winners run is worth more than being right most of the time.
- Rules remove emotion ??but bad rules still lose money. Automation guarantees discipline, not profit. The stop-loss protected capital; the entry logic still made avoidable mistakes.
- Leverage magnifies your process, not just the market. A 3x ETF made the good entries great and the bad entries painful. Whatever edge or flaw exists in your approach, leverage will amplify it. That's a sobering argument for keeping leveraged products to a small, defined slice of any portfolio ??if you use them at all.
The Bottom Line
Seven months, 18 completed trades, a 61% win rate, and a reward-to-risk ratio of 1.92 ??with one +77% round and one -15% stop-loss to show both ends of the spectrum. The system worked not because it was clever, but because it was consistent and because its winners were structurally larger than its losers. Its real weakness was entry timing in extended markets, a flaw the exit rules could contain but not cure.
I'll keep publishing these logs as the account continues, including the trades that don't go my way. If there's one thing worth taking from a real, unedited record like this, it's that a trading system's honesty lives in its losing trades ??and those are exactly the ones most write-ups leave out.