Corrected Inference for the CPE Portfolio Tilt Strategy: Newey-West HAC Standard Errors and Robustness Checks

Paper 4 of the CPE series reported statistically significant portfolio tilt alpha at 63-day (t=4.07, p<0.001) and
126-day (t=5.51, p<0.000001) horizons using a standard one-sample t-test, but flagged overlapping weekly
observations as a key limitation requiring Newey-West HAC-corrected inference. This research letter closes
that gap. Applying HAC-robust standard errors with both rule-of-thumb (3 lags) and horizon-matched (13/26
lags) specifications, the 63-day result survives with HAC t=3.53 (p=0.001) under the rule-of-thumb
specification and HAC t=3.30 (p=0.002) under the horizon-matched specification. The 126-day result
survives with HAC t=4.71 (p<0.001) under the rule-of-thumb specification and HAC t=3.60 (p=0.001) under
the horizon-matched specification — the most conservative correction applied. We supplement with four
robustness checks: block bootstrap with block size equal to the horizon length (p<0.001 at both horizons);
subperiod analysis confirming both halves of the evaluation window are independently significant at 63d and
126d; non-overlapping observations showing consistent direction but insufficient power at n=3–6; and Wilson
score confidence intervals for calibration showing the 97.0% bullish hit rate lies above the stated 93.0% CPE
even after Bitcoin ETF concentration adjustment (effective n»36,142). The 21-day horizon shows no
significant edge under any specification. The conclusion of Paper 4 stands under corrected inference.