Figure 1

Mean absolute error (a), rms error (b), and exchange probability (c) as a function of $k$ for one-step prediction with $N=1500$. Dotted curves reflect error when keeping all observations. The error in the traditional method is denoted by the horizontal dash-dotted line. Solid, dotted, and dash-dotted curves are means over 40 realizations; dashed curves capture the central $95\%$ of the distribution. rms error (d) with $2\%$ observational noise on the data stream.Reuse & Permissions

Figure 2

Cumulative probability distribution of one-step absolute prediction errors for $N=1500$ with $k=0$ (solid line), $k=1500$ (dashed line), $k=15000$ (dash-dotted line), and $k=30000$ (dotted line). Each curve represents an average over 40 realizations.Reuse & Permissions

Figure 3

Ratio of prediction errors of the refined method to the traditional method as a function of $N$ for one-step (a) and two-step iterated (b) prediction showing mean absolute error (solid line), rms error (dashed line), and $90\%$ quantile of absolute error (dotted line).Reuse & Permissions