Exact pruned dynamic programming tests the documented periodisation of the Turkey–PKK conflict in geocoded event data, 1989–2016

Abstract

Periodisations of the conflict between Turkey and the Kurdistan Workers’ Party rest on qualitative turning points. Periodisation is cast as penalised optimal segmentation, a shortest-path problem, solved exactly by pruned dynamic programming on 1,461 weeks of event data. The optimum recovers five of six documented turning points, but a pre-specified circular-shift test cannot distinguish this from chance (p = 0.123). Post hoc, simulations show low power, lower with the seasonal cycle than without it; modelling or removing that cycle lowers p to 0.006–0.010, and a greedy solver would have reported a valid positive; none survives correction for multiple testing.