On the invalidity of the ordinary least squares estimate of the equilibrium climate sensitivity

Abstract

The equilibrium climate sensitivity is often estimated by the ordinary least squares applied to annual data of observed/calculated temperature and forcing series. One of the conditions under which the ordinary least squares estimator is consistent is the uncorrelatedness of the regressor and regression error. However, this condition can fail in a regression using historical data of temperature and forcing. Alternative estimators established in econometrics are shown to mitigate the impact of the correlated regressor and regression error and deliver a more reliable estimate of the equilibrium climate sensitivity.