Mapping the Safe-Haven Network: An Exhaustive Search for Lead-Lag Structures across Timeframes

This study maps the lead-lag structure of the safe-haven network using an exhaustive search framework. We analyze 30 instruments (29 currency pairs and XAUUSD) across three timeframes (D1, H1, M15), testing 1,305 pairs (be applied per timeframe) and 1,670,400 combinations of regime filters, trading sessions, and tolerance windows. Each candidate pair is subjected to a six-layer validation framework: binomial sign test, bootstrap resampling, non-parametric permutation test, robustness across tolerance windows, chronological out-of-sample validation, and Benjamini-Hochberg false discovery rate correction.The results reveal a sparse structure. Of 1,305 pairs tested, only 50 pass FDR correction and only 5 survive full validation (0.38% conversion rate). Lead-lag structure is timeframe-dependent: H1 produces 4 validated pairs, M15 produces 1, and D1 produces none. It is also session-dependent: New York produces the most FDR-passing pairs (27 at H1). Five pairs are validated: AUDJPY-CHFDKK, CADCHF-XAUUSD, AUDJPY-GBPUSD, and CADCHF-GBPUSD at H1, and CHFDKK-USDCNH at M15. GBPUSD emerges as the most frequent leader, leading two of the five validated pairs. XAUUSD leads CADCHF, providing further evidence that XAUUSD can act as a lead instrument under the conditions identified in this study. USDCNH leads CHFDKK at M15.