The fixed-b asymptotic framework provides refinements in the use of heteroskedasticity and autocorrelation consistent variance estimators. We show however that the fixed-b limiting distributions of t-statistics are not pivotal when the variance of the underlying data generating process changes...