20 Years of Backtest: -7% in Crashes, +112% in Strong Years — Where the Alpha Comes From
20 Years of Data, Spread Open
2008 financial crisis. S&P 500: -37%. Our system: -7%.
2022 rate-hike bear. S&P: -19%. We: -11%.
2025 strong year. S&P: ~+10%. We: +112%.
These aren't hand-picked samples. They're the output of running our signal system across the full 2006-2026 window, on the US universe of roughly 1,800 stocks, with a $300k fresh account cold-started at the beginning of every test window.
Here is what that looks like.
The 20-Year Picture
| Scenario | Period | Win rate | Strategy CAGR | Max drawdown | S&P | Alpha |
|---|---|---|---|---|---|---|
| Full baseline | 2006-04 → 2026-04 (20y) | 48.0% | 30.82% | 34.00% | ~+8%/yr | +22%/yr |
| 2008 crash | Full year | 37.9% | -6.84% | 26.12% | -37% | +30% |
| 2022 bear | Full year | 31.1% | -11.28% | 16.04% | -19% | +8% |
| 2024 bull | Full year | 51.7% | +54.00% | 14.03% | +25% | +29% |
| 2025 strong | Full year | 52.4% | +112.13% | 19.27% | ~+10% | +102% |
| Last 12 months | 2025-04 → 2026-04 | 59.5% | +141.97% | 12.33% | ~+10% | +132% |
| Last 4 months | 2025-12 → 2026-04 | 57.5% | +67.09% | 10.38% | ~+5% | +62% |
Before reading these numbers: a backtest is not a forecast. It's a reconstruction of the past. Treating backtest numbers as expected returns is irresponsible. What follows are facts, not promises.
Counterintuitive #1: The Alpha Lives in the Bear
Look at rows 2 and 3. The system lost 7% in 2008 and 11% in 2022.
That doesn't sound sexy. But put it next to the market:
| Year | Market | System | Excess alpha |
|---|---|---|---|
| 2008 | -37% | -7% | +30% |
| 2022 | -19% | -11% | +8% |
| 2024 | +25% | +54% | +29% |
| 2025 | +10% | +112% | +102% |
In bulls the system can multiply the market's return. In bears, it cuts the loss to a fraction. The bear alpha and the bull alpha both came from the same set of rules running unchanged across two decades.
This is the core thesis of systematic rules. Not "we can pick the next Tesla." It's "when the market caves, we get cut much less than everyone else, and when the market runs, we let our winners run further."
How the system achieves that is our moat — we don't spell out the mechanism. But the 2008 and 2022 pressure tests gave the same answer in real dollars: the account didn't get washed out in either year.
Counterintuitive #2: Win Rate Is 48% — How Does It Make Money?
3,784 trades over 20 years, 52% of them lose money.
Sounds like a coin flip. But look at the top 5 winners by contribution:
| Symbol | Contribution |
|---|---|
| SNDK | +$20.8M |
| UAMY | +$6.2M |
| RGTI | +$5.1M |
| DAVE | +$3.5M |
| AMPX | +$2.5M |
SNDK alone produced 70x the starting capital.
The system is a right-tail hunter. It accepts a large number of small losses for the right to not miss the one or two stocks per year that run 2x, 5x, 10x. When one of those emerges, the system lets the winner grow.
Below 50% win rate, net profitable. That is a design feature, not a bug. If you want 80% win-rate comfort, this system is not for you.
Counterintuitive #3: Short Windows Lie to You — Both Ways
Most people ask "how is it performing lately?" We ran those windows too:
| Window | Annualized | Reading |
|---|---|---|
| Last 3 months | 55.49% | Recent strength — could revert |
| Last 4 months | 67.09% | Multi-baggers carrying the slate |
| Last 12 months | 141.97% | Multi-baggers held all year |
| 2024 full year | 54.00% | Textbook |
| 2025 full year | 112.13% | Outlier strong year |
Same system. Last 3 months annualized 55%. Looks like 2025 will compound forever — it won't.
A short window can flatter a system as easily as it can flatter the market. Our multi-bagger holds typically run 6-18 months; many of the eye-catching trailing numbers are still carrying open positions whose fate is not yet decided. A full year remains the minimum credible evaluation window, and even that is one sample.
Alpha Distribution (Backtested, observed only — not a forecast)
From 20 years of data plus 7 independent yearly/sub-year windows, here is the observed distribution:
| Scenario | Observed frequency | Alpha range |
|---|---|---|
| Normal year (no multi-baggers caught) | ~60% | +15% to +25% /yr |
| Good year (one multi-bagger caught) | ~25% | +30% to +50% /yr |
| Big year (2+ multi-baggers) | ~10% | +70% or more |
| Degraded / failed year | Not yet observed | Worst recorded: ~+5% |
A "20% alpha floor" holds in every sample we have run. But we won't claim "90% probability of holding 20%" — that requires more independent yearly samples than we currently have.
Methodology (Full Transparency)
- Data range: 2006-01 through 2026-04, full US market
- Validation protocol: each window is cold-started independently with $300k capital, max 20 concurrent positions
- No overfitting: In-sample (2006-2019) CAGR 15.99%; out-of-sample (2020-2026) CAGR 43.69%. OOS materially outperforms IS — the opposite of what overfitting produces.
- Benchmark: SPY buy-and-hold, including 2008/2020/2022 drawdowns
- Internally auditable: all backtest scripts and data artifacts are archived; live vs. backtest deviation will be published continuously.
Closing Thought
What the 20-year data actually says is not "you'll make 30% a year." It is this: across every real stress test the US market threw at us over two decades, this system never cut the account in half. Worst drawdown was 34%. S&P's worst over the same period was 55%.
That's the deliverable. Not a promise of wealth. The spine to stay in the market when everyone else is getting washed out.
Disclosure: All data in this post comes from our internal backtest scripts, which are auditable. Investing involves risk. Historical backtest does not represent future returns. Nothing here constitutes investment advice; signals are reference tools.