Irregular leadership changes in 2014: Forecasts using ensemble, split-population duration models

详细信息    查看全文

• 作者：Andreas BegerAuthor Vitae ; Cassy L. DorffAuthor Vitae ; Michael D. Ward ; ^{michael.d.ward@duke.edu" class="auth_mail" title="E-mail the corresponding author}Author Vitae
• 关键词：Government forecasting ; Political Instability Task Force
• 刊名：International Journal of Forecasting
• 出版年：2016
• 出版时间：January-March 2016
• 年：2016
• 卷：32
• 期：1
• 页码：98-111
• 全文大小：1064 K

文摘

We forecast Irregular Leadership Changes (ILCs)–unexpected leadership changes in contravention of a state’s established laws and conventions–for six months in mid-2014 using predictions from an ensemble of seven split-population duration regression models. The original forecasts were made in May 2014. Our approach allows us to aggregate models for different mechanisms leading to ILCs in one ensemble forecast, is sensitive to the overwhelming number of non-events (zeros) in the data, and allows us to make real-world forecasts with a lag of approximately five weeks. The data are based on 45 ILCs recorded for the period from March 2001 to March 2014, with monthly observations for up to 168 countries worldwide. The ensemble achieves in- and out-of-sample AUCs of ∼0.85, and we present the 10 highest forecasts, which include Thailand.