Backtest & Forecast Report
New enrichment signals (FRED macro + Yahoo targets + insider) over 6m / 12m / 24m, with Monte‑Carlo 6m forward projection.
⚠ Read before interpreting
Look-ahead audit passed: all filters use data known at MOO (prior day’s close-based rsi14_lag1, pre-market pm_drift/pm_vol, open-known gap_pct). Yahoo targets snapshot dropped. Per-ticker daily signals (not macro-triggered) — fires on any day with the ticker-specific pattern. Liquidity gate: PM vol ≥ 50K. Ensemble deduped on (date, ticker, direction). Position $10k/trade, cost 0.10% round-trip.
The signals (S1–S6) were discovered on the full dataset (2023-01 … 2026-04), so the 6m / 12m / 24m backtests overlap the discovery window and should be treated as in-sample — real OOS edge will be lower.
The per-day Sharpe shown is inflated by intraday diversification across many concurrent trades; the more honest metric is WR-net per trade.
Cost assumption: 0.10% round-trip per trade (auction liquidity on both legs). Position size: $10k / trade, unlimited concurrent.
Forecast is a bootstrap from the last 12 months of aggregate daily P&L — extrapolates prior regime, does not predict regime shifts.
Filters
Equity Curve — Ensemble (12m)
6-Month Forecast — Monte Carlo fan
Bootstrap of paths from the last 12 months of daily aggregate P&L for the selected signal. Shaded bands = 5–95% and 25–75% percentiles; line = median.
All Signals × Windows
Side-by-side stats. Sort by clicking headers.
Signal recipes
Methodology & assumptions
Universe
Datum daily bars ≥ 2M avg daily volume, price ≥ $5.
Entry / Exit
Entry at MOO (9:30 auction). Exit at 9:55 market order. No intra-window stops, no discretion.
Costs
0.10% round-trip per trade (0.05% slippage + 0.05% commission). Applied at trade level before aggregation.
Position sizing
$10,000 per trade, unlimited concurrent. No leverage applied beyond gross.
New data sources
FRED macro (23+8 series, 10y daily) · Yahoo price targets (998 tickers) · Yahoo insider trades (69K rows, cleaned for parse bugs).
Forecast method
Bootstrap: resample last 12m daily P&L with replacement, 2000 paths × 126 trading days. Zero-P&L days included when no signal fires.