backtest
Macro proxy vs calendar judge
Deterministic SPY>SMA200 and FOMC±1 filters as stand-in for news judgment — no headline archive.
Past “news scrape then judge” cannot be replayed without a headline corpus. This run uses rule proxies for what news often encodes.
Base: momentum 2R · ATR×2 · RS>SPY · sector>SPY · liquidity floor (no pennies) · 2019→now · n=170
Pooling: all passing signals, time-ordered 1R compound (not live top-1 picker)
| Variant | Trades | Win% | E[R] | stop/target/time | CAGR% | MaxDD% |
|---|---|---|---|---|---|---|
| Chart + RS + sector | 3286 | 43.88 | 0.127 | 1628/868/790 | 66.84 | −70.38 |
| + SPY > SMA200 | 2903 | 44.06 | 0.1298 | 1450/782/671 | 59.68 | −72.80 |
| + FOMC ±1 no entry | 2985 | 43.55 | 0.1198 | 1483/787/715 | 54.75 | −69.97 |
| + SPY>200 and FOMC window | 2625 | 43.85 | 0.1302 | 1307/713/605 | 52.89 | −68.40 |
Honest claim
Calendar/regime filters are point-in-time deterministic — not an LLM news desk test. E[R] bumps slightly when macro gates trim trade count; CAGR falls because fewer trades compound.
Source
Summarized from state/reports/macro_proxy_bt.md.