Abstract
<jats:p>Motivated by the presence of a strong cyclical component in paleoclimate data, this chapter considers the problem of conducting cointegration inference when the data contains very large and persistent cycles. Our first contribution is to show, analytically and through Monte Carlo simulations, that while point estimation remains consistent, commonly applied tests are no longer valid when the data contains persistent cycles. Our second contribution is empirical: the authors propose the use of the long-run covariability approach of Müller and Watson (2018) to quantify low-frequency comovement amongst a range of paleoclimate times series over the last 800,000 years. These methods allow us to focus on the long run properties of the data, bypassing short and medium run fluctuations, while being agnostic regarding the order of integration of the time series. The authors provide new estimates for the long-run relationship between temperatures and CO2 atmospheric concentration, concluding that in the long-run a 100 ppm increase in CO2 concentration levels would raise temperatures by around 1°C. Finally, the authors illustrate how joint modeling of this set of paleoclimate time series can be carried out by factor analysis and how long-term projections about temperature increases and ice-sheet retreat can be constructed.</jats:p>