Alpha Asymmetry in Foreign Exchange Markets: An Investigation of Exploitability

Abstract

Skewness-based trading rules presuppose asymmetrically distributed signals. We test that premise, and whether a rule built on it earns money, for five technical transformations of the EUR/JPY price (“alpha signals”), using weekly indicative Yahoo Finance mid-quotes from January 2016 to August 2025 (504 weeks). Under block-bootstrap inference only the volatility-expansion signal is robustly skewed (1.75, 95% CI [1.18, 2.16]). The tail signal’s skewness changes sign across three defensible weekly aggregations of one daily rule, and the positive tail skewness reported in an earlier version described an unsigned magnitude series, right-skewed by construction. A skewness-threshold rule executed at the first trading-session open after each Friday signal returns −0.73% cumulative gross over 15 holding episodes, with a Sharpe ratio indistinguishable from zero. No break-even cost exists. The result is not stable across specifications: all four execution timings lose money, and four pre-specified readings of the entry rule span 9.04 percentage points and change sign. Walk-forward selection leaves the rule nearly dormant, and proxy factor regressions show no positive residual alpha (intercept t = −0.45 on in-position weeks). Neither White’s Reality Check nor Hansen’s SPA rejects the null that no candidate rule beats a zero-return benchmark. Earlier versions, now superseded, misapplied the holding rule and priced at the Friday close. The evidence covers one currency pair and does not model capacity, financing or per-order costs. Within that scope the evidence does not establish an exploitable edge, and the results depend on construction choices the original specification did not justify.