Forecasting Conflict Intensity: Afghanistan
Forecasting Conflict Intensity: Afghanistan
Author(s): Robert Bryce; Kevin Sprague
No pages: 7
Year: 2012
Article ID: 15-3-4
Keywords: Afghanistan, analysis, conflict modelling, forecasting conflict, training and analysis
Format: Electronic (PDF)
Abstract: Aggregate violence data in Afghanistan from 1 January 2005 to 30 June 2011 is considered. The time series is characterized by correlated stochastic fluctuations around a smoothly growing and seasonally varying mean intensity level, as well as sudden and large spikes in intensity. These spiky outliers are directly correlated with election day violence and appear decoupled from the underlying dynamics. Key results are as follows: (1) approximate exponential growth (λ = (144.7 ± 15.5) × 10–5 /day) in long-term conflict intensity, (2) election day violence exhibiting the same growth rate (λ = (145.2 ± 7.6) × 10–5 /day) as the day-to-day conflict growth – these growth rates are comparable to that of coalition fatalities (λ = (91.6 ± 0.3) × 10–5 /day), and (3) fluctuations displaying nominal power-law scaling (α = 0.63 ± 0.01) with the conflict intensity. Smoothing allows forecasting of the expected (mean) intensity while the observed power-law scaling enables usage of surrogate time series to forecast the stochastic fluctuations enveloping the mean. One crucial finding arising from the nominal exponential growth in election day violence is an expectation of a more than doubling in violence in the next election day over the 2010 elections.