Sign in

Julian Walterskirchen

@jwalterskirchen.bsky.social
307 followers 744 following 16 posts

PhD Student, Dep. PolSci University of Gothenburg Interested in Conflict, Security, Cooperation, Terrorism, Violent Extremism, and quant. Methods. ORCID: 0000-0002-5587-7075

PostsRepliesMedia
Julian Walterskirchen @jwalterskirchen.bsky.social · 27/04/2026
Grateful to my co-authors (@christianoswald.bsky.social , Marco Binetti, Sonja Häffner), the Center for Crisis Early Warning (CCEW), and the organizers of the prediction challenge. Open access here: doi.org/10.1080/0305...
doi.org
Taking time seriously: Predicting conflict fatalities using temporal fusion transformers
Previous conflict forecasting efforts identified three areas for improvement: the importance of spatiotemporal dependencies and nonlinearities and the further exploitation of latent information in ...
010
Julian Walterskirchen @jwalterskirchen.bsky.social · 27/04/2026
- Out-of-sample / true future evaluation is critical—no single model dominates across all contexts - Interpretability remains important alongside predictive accuracy
TFT interpretability feature showing importance of different variable sets over time
100
Julian Walterskirchen @jwalterskirchen.bsky.social · 27/04/2026
A few takeaways from the project: - Methodological development remains essential, especially for temporal dynamics and uncertainty - Collaborative initiatives like VIEWS enable cumulative and transparent progress
100
Julian Walterskirchen @jwalterskirchen.bsky.social · 27/04/2026
Empirically, the model performs competitively with established benchmark models from the VIEWS early warning system across multiple evaluation metrics.
100
Julian Walterskirchen @jwalterskirchen.bsky.social · 27/04/2026
In the paper, we apply Temporal Fusion Transformers (TFTs) to predict conflict fatalities. The approach addresses several recurring challenges in forecasting: - nonlinear and spatiotemporal dependencies - probabilistic (not just point) forecasts - interpretability despite model complexity
Temporal Fusion Transformer model architecture diagram
100
Julian Walterskirchen @jwalterskirchen.bsky.social · 27/04/2026
More broadly, prediction is becoming increasingly central to political science and policy. In conflict research, forecasts can help clarify likely trajectories, quantify uncertainty, and support early warning and prevention efforts.
100
Julian Walterskirchen @jwalterskirchen.bsky.social · 27/04/2026
The challenge brought together a wide range of approaches to improve conflict prediction. You can explore other contributions here: viewsforecasting.org/projects/gff... It’s a great example of how collective efforts can move the field forward.
viewsforecasting.org
GFFO: Prediction Challenge 2023/2024
With the goal of improving the accuracy and certainty of our forecasts, we launched our second prediction competition in 2023. It challenged participants to forecast conflict intensity as a probabilit...
100
Julian Walterskirchen @jwalterskirchen.bsky.social · 27/04/2026
Our recent paper has been published in @intlinteractions.bsky.social Taking time seriously: Predicting conflict fatalities using temporal fusion transformers This paper documents our contribution to the GFFO / VIEWS Prediction Challenge on forecasting organized violence.
142
Reposted by Julian Walterskirchen
Institute for Replication @i4replication.bsky.social · 01/04/2026
🧵1/ Our first meta-science paper (with 350+ coauthors) is published today in Nature. It presents one of the largest-ever reproducibility projects in economics & political science. Here’s what we found 👇
217089
Julian Walterskirchen @jwalterskirchen.bsky.social · 09/03/2026
The paper is open access here: journals.sagepub.com/doi/10.1177/... Very grateful to the many colleagues and collaborators who supported this work! #PoliSciSky #Conflictsky
journals.sagepub.com
Sage Journals: Discover world-class research
Subscription and open access journals from Sage, the world's leading independent academic publisher.
000
Julian Walterskirchen @jwalterskirchen.bsky.social · 09/03/2026
Key takeaways: • Text features can improve conflict prediction, but gains vary widely across methods • More complex models (e.g., transformers) are not consistently better than simpler approaches • The usefulness of NLP features depends on text sources and the availability/quality of other features
Lolliplot of predictive performance of different feature combinations across different timepoints.
111
Julian Walterskirchen @jwalterskirchen.bsky.social · 09/03/2026
Happy to share that my new paper is out in @conflictmanagement.bsky.social I compare several NLP approaches for conflict prediction: • conflict dictionaries • sentiment dictionaries • word scaling • dynamic topic models • specialized transformer models
Text features and baseline models' predictive performance line charts over 6 time horizons (BBC data, left; CrisisWatch data, right).
142
Reposted by Julian Walterskirchen
Moritz Marbach @moritz-marbach.com · 27/02/2026
🧺 Paper Picnic 2.0 is here! More journals. New features. An easier way to keep up with the latest research in political science and adjacent fields. 🧵👇
27530
Julian Walterskirchen @jwalterskirchen.bsky.social · 23/07/2025
Happy Day! More excited than I expected about reaching 100 citations in Google Scholar! Immensely grateful to my amazing co-authors who made it possible to reach this (mini) milestone!
Screenshot of a Google Scholar profile reaching 100 citations.
040
Julian Walterskirchen @jwalterskirchen.bsky.social · 07/05/2025
The true future forecast is still being evaluated, but you can check out our and the other teams' contributions already here: predcomp.viewsforecasting.org
predcomp.viewsforecasting.org
Dashboard | Challenge 23/24
Views Data Dashboard
010
Julian Walterskirchen @jwalterskirchen.bsky.social · 07/05/2025
The article introducing the 2023/24 @viewsforecasting.org challenge just got published open access. This was a lot of fun to contribute to. doi.org/10.1177/0022...
doi.org
The 2023/24 VIEWS Prediction challenge: Predicting the number of fatalities in armed conflict, with uncertainty - Håvard Hegre, Paola Vesco, Michael Colaresi, Jonas Vestby, Alexa Timlick, Noorain Syed...
Governmental and nongovernmental organizations have increasingly relied on early-warning systems of conflict to support their decisionmaking. Predictions of war...
120
Reposted by Julian Walterskirchen
Political Analysis @polanalysis.bsky.social · 11/03/2025
Currently in FirstView: “Exponential Random Graph Models for Dynamic Signed Networks: An Application to International Relations” by @corneliusfritz.bsky.social, @mehrl.bsky.social, Paul Thurner, and Göran Kauermann. The authors propose the Signed Exponential Random Graph Model (SERGM).
113
Reposted by Julian Walterskirchen
Institute for Replication @i4replication.bsky.social · 08/04/2024
Our first meta paper is out!! This paper combines our first 110 completed reproductions/replications. This is joint work with 350+ amazing coauthors. We summarize our findings below: econpapers.repec.org/paper/zbwi4r...
15031
Julian Walterskirchen @jwalterskirchen.bsky.social · 17/01/2024
I always dream of "knorke" becoming cool again, no luck so far though.
110
Julian Walterskirchen @jwalterskirchen.bsky.social · 22/10/2023
Our Political Analysis article got an issue number and is now fully officially published. :) Check it out here www.cambridge.org/core/journal... also check out the neat web-app we built to play around with it. kfeapps-conflict-intensity-streamlitapp-auon5d.streamlit.app
0123