FAQ
Quick answers to common questions about our methods, forecasts, and data.
Are the patterns specific to a country, region, or epoch?
No. We test out‑of‑country, out‑of‑region, and out‑of‑period generalization and find that core shapes recur across contexts. See our Predictability working paper.
Do you make subnational predictions?
Yes. We forecast at PRIO‑GRID (0.5°) and country levels. See the 3D shapes working paper in Working Papers.
How does the autoregressive baseline compare?
At short horizons, purely autoregressive shape‑matching performs on par with covariate‑augmented versions. See EPJ Data Science and the JPR paper on variability.
What do you explain particularly well?
Variation over time — especially the onset, escalation, and decay of episodes. By borrowing futures from the closest analogs, the model captures bursts and plateaus that standard linear baselines often smooth out. See our Journal of Peace Research paper on accounting for variability.
Is this causal or predictive?
Predictive. We focus on anticipating outcomes given current trajectories and historical regularities. The analog set provides interpretable narratives — “this looks like these past episodes” — to support diagnostic and scenario discussions.
How do you quantify uncertainty?
We construct similarity‑weighted mixtures of analog futures to produce prediction intervals (50/80/95%), quantiles, and exceedance risks; calibration is monitored with rolling coverage, PIT, and CRPS. Details on the Methodology page.
What horizons do you support?
Typically 1–6 months; weekly or quarterly on request.
Does this work beyond conflict fatalities?
Yes. We use similar shape‑based methods for protest intensity and migration flows. The approach is generic to time‑series with recurring motifs and regime shifts.
Can I use the data and forecasts?
Yes. See Downloads for datasets and formats, and Publications for citation details.