Abstract. We use a statistical model, the cointegrated vector autoregressive model, to assess the degree to which variations in Earth's orbit and endogenous climate dynamics can be used to simulate glacial cycles during the late Quaternary (390 kyr–present). To do so, we estimate models of varying complexity and compare the accuracy of their in-sample simulations. Results indicate that strong statistical associations between endogenous climate variables are not enough for statistical models to reproduce glacial cycles. Rather, changes in solar insolation associated with changes in Earth's orbit are needed to simulate glacial cycles accurately. Also, results suggest that non-linear dynamics, threshold effects, and/or free oscillations may not play an overriding role in glacial cycles.