Scoring methodology
Every strategy is backtested on the same 15 years of real daily data, then ranked by a single composite score that balances return against risk.
The composite score
A weighted blend of six components, each fed by your real backtest metrics (no random padding):
| Component | Weight | Rewards |
|---|---|---|
| Return (CAGR) | 30% | Higher compound annual growth (sigmoid-normalized so it can't run away) |
| Sharpe ratio | 25% | Return earned per unit of risk |
| Max drawdown | 20% | Capital preservation — deep losses are penalized exponentially |
| Consistency | 15% | Share of winning months |
| Stability | 5% | Durable long-run compounding |
| Volatility | 5% | Lower portfolio volatility |
The leaderboard re-computes this score on the server from your submitted metrics, so it can't be spoofed, and it stores your strategy definition so any entry is reproducible.
How the backtest works
- Data: daily adjusted close (splits + dividends) for 200+ US stocks & ETFs, ~15 years, via Yahoo Finance.
- Costs: a per-side transaction cost is charged on turnover. Taxes are not modeled.
- Benchmark: every strategy is compared to SPY (S&P 500) over the identical window.
- Determinism: the same strategy always produces the same result — fair comparison for everyone.
Honest limitations (read this)
- Overfitting: a strategy tuned to look great on history often fails live. High score ≠ good future bet.
- Survivorship bias: the ticker universe is today's well-known names; delisted failures aren't included.
- No slippage/liquidity modeling beyond a flat cost; real fills differ.
- Hypothetical results are not real trading and not a prediction.