Heteroscedasticity and Autocorrelation Efficient (HAE) Estimation and Pivots for Jointly Evolving Series

A new two-way map between time domain and numerical magnitudes or values domain (v-dom) provides a new solution to heteroscedasticity. Since sorted logs of squared fitted residuals are monotonic in the v-dom, we obtain a parsimonious fit there. Two theorems prove consistency, asymptotic normality, efficiency and specification-robustness, supplemented by a simulation. Since Dufour's (1997) impossibility theorems show how confidence intervals from Wald-type tests can have zero coverage, I suggest Godambe pivot functions (GPF) with good finite sample coverage and distribution-free robustness. I use the Frisch-Waugh theorem and the scalar GPF to construct new confidence intervals for regression parameters and apply Vinod's (2004, 2006) maximum entropy bootstrap. I use Irving Fisher's model for interest rates and Keynesian consumption function for illustration.