Julian Walterskirchen’s research while affiliated with University of the Bundeswehr Munich and other places

When do autocrats replace their ministers? Research suggests that failed coup attempts lead autocrats to remove disloyal ruling elites. While this is an important and intuitive finding, failed coup attempts are very rare events and hence leave large amounts of variation in minister removal unexplained. Autocrats are reactionary when a rare window of opportunity, such as a failed coup attempt, occurs. However, previous work also suggests that they are anticipatory and survival-seeking. We argue that autocratic leaders are strategic actors who take preemptive action against potentially disloyal elites, rather than risking a coup attempt being realized. Using structural coup risk as a lower estimate of a leader's own perception of coup risk, we find that it significantly contributes to the prediction of minister removals. Compared to failed coup attempts, we find that it substantially improves predictive performance across a number of metrics. Our findings contribute to nascent research on government stability and the survival of autocratic leaders.

Previous conflict forecasting efforts identified three areas for improvement: the importance of spatiotemporal dependencies and nonlinearities and the further exploitation of latent information in conflict variables, a lack of interpretability in return for high accuracy of complex algorithms, and the need to quantify prediction uncertainty. We predict conflict fatalities with temporal fusion transformers which have several desirable features for forecasting, addressing all these points. First, they can produce multi-horizon forecasts and probabilistic predictions, offering a flexible and nonparametric approach. Second, they can incorporate time-invariant covariates, known future inputs, and other exogenous time series which allows to identify globally important variables, persistent temporal patterns, and significant events. Third, this approach puts a strong focus on interpretability so that we can investigate temporal dynamics more thoroughly. Our approach outperforms benchmarks from an award-winning early warning system over several metrics and test windows and is a valuable addition to the forecaster's toolkit.

Governmental and nongovernmental organizations have increasingly relied on early-warning systems of conflict to support their decisionmaking. Predictions of war intensity as probability distributions prove closer to what policymakers need than point estimates, as they encompass useful representations of both the most likely outcome and the lower-probability risk that conflicts escalate catastrophically. Point-estimate predictions, by contrast, fail to represent the inherent uncertainty in the distribution of conflict fatalities. Yet, current early warning systems are preponderantly focused on providing point estimates, while efforts to forecast conflict fatalities as a probability distribution remain sparse. Building on the predecessor VIEWS competition, we organize a prediction challenge to encourage endeavours in this direction. We invite researchers across multiple disciplinary fields, from conflict studies to computer science, to forecast the number of fatalities in state-based armed conflicts, in the form of the UCDP ‘best’ estimates aggregated to two units of analysis (country-months and PRIO-GRID-months), with estimates of uncertainty. This article introduces the goal and motivation behind the prediction challenge, presents a set of evaluation metrics to assess the performance of the forecasting models, describes the benchmark models which the contributions are evaluated against, and summarizes the salient features of the submitted contributions.

When do autocrats replace their ministers? Research suggests that failed coup attempts lead autocrats to remove disloyal ruling elites. While this is an important and intuitive finding, failed coup attempts are very rare events and hence leave large amounts of variation in minister removal unexplained. It also portrays autocrats as reactionary, whereas previous work suggests that they are anticipatory and survival-seeking. We argue that autocratic leaders are strategic actors who take preemptive action against potentially disloyal elites, rather than risking a coup attempt being realized. Using structural coup risk as a lower estimate of a leader's own perception of coup risk, we find that it significantly contributes to the prediction of minister removals. Contrasting its effect to that of failed coup attempts, we find that it substantially improves predictive performance across a number of metrics. Our findings contribute to nascent research on government stability and autocratic leader survival.

Blair et al. (2023) examine the effect of UN peacekeeping on democratization in conflict-affected countries. They use fixed effects and instrumental variable estimators and find evidence that ”UN missions with democracy promotion mandates are strongly positively correlated with the quality of democracy in host countries but that the magnitude of the relationship is larger for civilian than for uniformed personnel, stronger when peacekeepers engage rather than bypass host governments when implementing reforms, driven in particular by UN election administration and oversight, and more robust during periods of peace than during periods of civil war”. Since the authors provide an impressive list of robustness checks, we focus on computational and robustness reproducibility. We replicate the findings using the Stata code provided in the replication material and reproduce all main analyses in R. We add year fixed effects to country fixed effects, cluster standard errors, use fixed and random panel regression estimators and ordered Beta regression estimators. We furthermore reproduce instrumental variable estimators with two different packages. We find that the original findings were reproducible and robust.

Natural language processing (NLP) tools have been applied successfully in improving predictions in a wide range of research areas. However, what works in one area may not work in conflict research. We therefore seek to offer an initial assessment of the most prominent NLP methods for conflict prediction tasks. We evaluate the performance of features extracted from a conflict dictionary, two sentiment dictionaries, a word-scaling approach, dynamic topic models, and a transformer model on a classical conflict prediction task. The results highlight the importance of considering different NLP approaches, depending on the availability of text sources and other predictor variables.

Previous conflict forecasting efforts identified three areas for improvement: the importance of spatiotemporal dependencies and nonlinearities and the further exploitation of latent information in conflict variables, a lack of interpretability in return for high accuracy of complex algorithms, and the need to quantify prediction uncertainty. We predict conflict fatalities with temporal fusion transformers which have several desirable features for forecasting, addressing all these points. First, they can produce multi-horizon forecasts and probabilistic predictions, offering a flexible and non-parametric approach. Second, they can incorporate time-invariant covariates, known future inputs, and other exogenous time series which allows to identify globally important variables, persistent temporal patterns, and significant events. Third, this approach puts a strong focus on interpretability such that we can investigate temporal dynamics more thoroughly. Our approach outperforms benchmarks from an award-winning early warning system over several metrics and test windows and is thus a valuable addition to the forecaster's toolkit.

This draft article outlines a prediction challenge where the target is to forecast the number of fatalities in armed conflicts, in the form of the UCDP `best' estimates, aggregated to the VIEWS units of analysis. It presents the format of the contributions, the evaluation metric, and the procedures, and a brief summary of the contributions. The article serves a function analogous to a pre-analysis plan: a statement of the forecasting models made publicly available before the true future prediction window commences. More information on the challenge, and all data referred to in this document, can be found at https://viewsforecasting.org/research/prediction-challenge-2023.

We replicate the analysis provided in Bokobza et al. (2022). They identify a causal effect of failed coup attempts on cabinet minister removals in autocracies on both the country and individual minister level and show that higher-ranking ministers and those holding strategic positions are more likely to be purged than more loyal and veteran ministers using fixed effects panel models. We focus on computational reproducibility and robustness replicability. In addition to reproducing the original results using Stata and R, we replicate analyses using random effects panel models and ordered beta regression models, reproduced analyses performed in R using different packages, replaced the main independent variable, clustered standard errors on a different level, and added independent variables related to coup-proofing. We find that the original findings were reproducible and robust.

Recent advancements in natural language processing (NLP) methods have significantly improved their performance. However, more complex NLP models are more difficult to interpret and computationally expensive. Therefore, we propose an approach to dictionary creation that carefully balances the trade-off between complexity and interpretability. This approach combines a deep neural network architecture with techniques to improve model explainability to automatically build a domain-specific dictionary. As an illustrative use case of our approach, we create an objective dictionary that can infer conflict intensity from text data. We train the neural networks on a corpus of conflict reports and match them with conflict event data. This corpus consists of over 14,000 expert-written International Crisis Group (ICG) CrisisWatch reports between 2003 and 2021. Sensitivity analysis is used to extract the weighted words from the neural network to build the dictionary. In order to evaluate our approach, we compare our results to state-of-the-art deep learning language models, text-scaling methods, as well as standard, nonspecialized, and conflict event dictionary approaches. We are able to show that our approach outperforms other approaches while retaining interpretability.