FIGURE 3.

Dependence of the Pearson correlation coefficient, $\rho$, on embedding dimension, $E$, for the example time series from Figure 1. The best fitting model is given by the highest $\rho$, and corresponds to $E=3$ which is what would be used to forecast ${\widehat{N}}_{T+1}$. We have shown high enough values of $E$ to show a clear decline in $\rho$, but in general the maximum $E$ considered should be about $\sqrt{T}$ (Munch et al., 2020), which is 10 here.