Walk-Forward v2 · 5y ZAPAS + 2y CSE
Applying FRED macro (23 series, as-of prior session) and insider trading (rolling 30d/90d ending strictly before trade) to two PM-MOO strategies. Zero look-ahead. Tested across 6, 12, and 24-month windows with 70/30 train/test splits.
Overview
Window Summary
| Strategy | Window | Date range | N trades | Train WR | Test WR | SA winners | 2-way survivors |
|---|
OOS Win Rate by Config
Top WF-Validated Setups
Survivors require train dWR ≥ +3pp (SA) / +5pp (combo) AND OOS dWR ≥ +1pp. Sorted by OOS WR lift.
| Strategy | Win | Type | Setup | Train N | Train WR | OOS N | OOS WR | Exp ret/trade | Sharpe ann |
|---|
6-Month Forecast
Projection = (trades/month in OOS) × 6 months × mean OOS return per trade. Assumes regime persistence. Not annualized P&L — per-unit position %.
Projected trade volume
Projected cumulative P&L (%)
Key Findings
ZAPAS (5yr, base 73% WR)
- ✦ NFCI_chg_q4 (tightening fin. conditions top quartile) — most consistent winner across 12m & 24m
- ✦ OIL_q1/q3 + rates combos survive all three windows — commodity-rate interaction robust
- ✦ ins_net30=mod_sell & ins_cluster_buy30=0 → OOS WR 85-87% in 12m & 24m
- ✦ T10Y2Y_q4 (steepest curve quartile) in 6m window → OOS WR 78-86%
- ✦ VIX_high + HY_q4 or VIX_chg_q4 → 81-85% OOS in 6m window
CSE Bridge (6m, base 61% WR)
- ✦ T10YIE_q3 (10y breakeven mid-range) — only SA survivor, +8.6pp OOS on 1,294 trades
- ✦ HY_q3 × DXY_q1 → train 67% / OOS 83.5% (+23.8pp!)
- ✦ HY_q3 × T10YIE_q3 → 74% / 82.4% OOS on 187 trades
- ✦ Insider mod_sell + T10YIE_q3 → 78% OOS on 127 trades
- ✦ TRAP: DXY_chg_5d_q=q2 → OOS WR 51% (−9pp vs base)
Caveats
CSE Bridge test window is 6 weeks (Feb 27 → Apr 17, 2026). Small OOS sample for low-N combos. Re-validate monthly.
Price targets dropped entirely — snapshot = look-ahead risk. Kept only FRED (as-of prior session) and insider (rolling windows strictly ending before trade).
ZAPAS test window (2025 bull run) has elevated base WR (77-78%). Macro-factor edges stack on top but absolute numbers may not persist in regime change.
Forecast P&L is per-unit position size. To convert to $ P&L, multiply by average position size and account for slippage/commission (not modeled here).