The trading study that shows its
Mercurio is a research project, not a product. We built an algorithmic trading bot, tested it to destruction, and proved — with our own data — that it did not beat the S&P 500 risk-adjusted. Every gate, trade, bug and drawdown is open to inspect. Backtested and paper-traded, never live.
Here's the whole method — gates, trades, drawdowns and all.
An algorithmic trading study, taken apart gate by gate.
No black box. Trend-following on 1-hour bars, gated by the market regime, screened by a language-model veto, capped by hard risk limits. We replayed it across two years of real history — the -29.3% drawdown included — and it returned +53.1%. That looks good alone. The honest question this study set out to answer was harder: does it beat simply holding the S&P 500? Keep reading.
Four gates between a chart and a trade.
Every session runs the same checklist. If the index is below its average, Mercurio holds cash - and holding cash was the decision that kept the leverage survivable.
Read the 200-day signal
One rule, one instrument. Each session Mercurio checks whether the S&P 500 closed above its 200-day simple moving average. Above: hold the 3x fund. Below: hold cash. Nothing else is predicted.
Execute at the daily close
The signal is read on the daily close and the rotation is placed in that same session. In the backtest, acting one day late cut the return below the index, so latency past 30 minutes raises an alarm.
3x via UPRO, never margin
Exposure comes from UPRO, a 3x daily S&P 500 fund - not from borrowed money, so there is no margin call. The price is daily leverage decay and a drawdown roughly twice the index's.
55% breaker and paper lock
A portfolio drawdown breaker at 55% is the only hard stop; there is no per-trade stop on a single leveraged holding. Paper trading is locked on until the rule is validated against the live index.
Two years, drawn in full.
Mercurio's account equity across the whole two-year backtest — every figure computed straight from the real equity curve, the -29.3% drawdown included.
Account equity, month by month, over the full backtest window.
Account equity from a historical simulation on 102 symbols · $25,000 paper capital · session-gap stop fills, 0.15% per-side slippage. Not live. Past performance does not guarantee future results.
The rule that replaced it
Diversification was the original plan. We built and backtested every variant - mean reversion, momentum, gap, pairs, crypto, trend following - and none beat simply holding the index on a risk-adjusted basis. What replaced them is one rule: hold a 3x S&P 500 fund while the index closes above its 200-day average, otherwise hold cash.
Leveraged Index Timing
Backtested total return -- - more dollars than the index, but with a lower Sharpe and roughly twice the drawdown. That trade-off is the whole product: amplified exposure, not alpha.
Figures from the 5-year backtest (2021-06-17 to 2026-06-17). Paper / simulated. A single overnight crash can make the drawdown far worse. Past performance does not guarantee future results.
Risk management is inviolable, not optional.
Position sizing, stop-losses, loss limits, and a drawdown circuit breaker are enforced in code, not suggestions. The engine moves through clearly defined states - and sits in cash whenever the odds are not there.
- Risk per trade
- 1.5%
- Daily loss limit
- 5%
- Weekly loss limit
- 7%
- Drawdown breaker
- 15%
- Max positions
- 15
- All systems live
Full allocation while the S&P uptrend is confirmed.
- Fast-drop guard
Position sizes cut as portfolio drawdown builds.
- Fast-rally
Trailing stops tighten to lock in gains on strong runs.
- Manual pause
Trading halted on command; positions keep their stops.
- Cooldown
Flat for five days after the drawdown breaker trips, then resume.
- Out of regime
Flat in cash whenever the S&P uptrend is not confirmed - no new longs.
Where the +53.1% came from
55.9% win rate, 1.38 profit factor, 381 trades. The exact configuration Mercurio runs today, broken down every way that matters — win rate, returns by period, and the names that carried the book.
- Capital base
- $25,000
- Net P&L
- +$13,278
- ROI
- +53.1%
- Win rate
- 56%
- Trades closed
- 381
- Profit factor
- 1.38
- Expectancy
- +$35
- Commissions
- $0.00
Return by calendar period
2024 and 2026 are partial years within the test window.
Top contributors
Net profit by symbol over the window.
- MU+$2,470
- CRWD+$2,195
- PANW+$1,956
- ASML+$1,737
- ARM+$1,680
These results are a historical simulation of Mercurio's live configuration on 102 symbols with $25,000 of paper capital, session-gap stop fills and 0.15% per-side slippage. They are not live trading results and not a forecast. Over a full five-year cycle that includes a bear market, the same strategy was net negative — shown openly on the backtesting page.
The study ended. The account kept trading — on the successor rule.
The trend-following strategy above was closed in June 2026: it did not beat the S&P 500. The same Alpaca paper account now runs Mercurio v3.0 — hold UPRO (3x S&P 500) while SPY closes above its 200-day average, otherwise short-term treasuries — updated straight from the engine. No real capital. The backtest was the thesis; this is what came after the honest verdict.
What other trading products bury, we lead with
Mercurio is a research project run entirely in paper trading — not a get-rich scheme. Trust the method, not the marketing. So here are the four things you deserve to know before anything else.
- It is paper trading. No real capital is at risk. Paper validation must finish before any live decision.
- It draws down. Even in the strong 2-year window, equity fell 29.3% from its peak before recovering.
- It is a bull-market strategy. Over a full 5-year cycle with a bear market, the same approach was net negative. We do not pretend otherwise.
- The edge is modest. A realistic expectation is in the low-double-digit percent per year — not the triple-digit fantasies sold elsewhere.
Every decision, written down and inspectable
See the method. Check the proof.
Dig into the engine, the risk model, and the full backtest — including the five-year cycle where this same strategy lost money. Or read the source on GitHub.