Asylum Acceptance Rates – An Economic Perspective

Ruben Schenk, Damola Agbelese, Yifan Bao, Daria Borodulina

Project Overview

The asylum acceptance rates project aimed to investigate the influence of various economic factors on asylum acceptance rates across six European countries – Germany, Austria, Belgium, Italy, Spain, and France. The project was conducted as part of the ‘Data Science in Techno-Socio-Economic Systems’ course at ETH Zurich in May 2023.

Motivation and Background

Understanding the determinants of asylum acceptance rates is critical in shaping policies and assessing the socio-economic impact of asylum seekers on host countries. The motivation stemmed from the ongoing global refugee crises, including conflicts in Syria, Ukraine, and Afghanistan, which have collectively displaced millions of individuals. The economic pressures associated with refugee influxes and the varying asylum acceptance rates across European countries prompted the exploration of economic factors as potential predictors of asylum decisions.

Data Collection and Processing

Data was sourced from three primary repositories:

The dataset included quarterly data spanning from 2005 to 2020, focusing on 13 economic indicators such as GDP per capita, unemployment rate, expenditure on social protection, and government debt. A total of 4,992 data points were compiled, covering the six selected countries.

Data preprocessing involved handling missing values through divisional, mean, and linear interpolations. For instance, asylum application data was linearly interpolated, while economic sentiment indices were averaged to smooth inconsistencies.

Methodology

The methodological approach involved two primary analysis techniques:

1. Machine Learning Models

Regression-based models were employed to predict asylum acceptance rates based on economic indicators. The following algorithms were utilized:

  • ElasticNet Regression: Combined Lasso and Ridge regression to handle multicollinearity.
  • Support Vector Regression: Focused on minimizing prediction errors using hyperplane separation.
  • K-Nearest Neighbors: Compared data points based on proximity in feature space.
  • Random Forest Regression: Aggregated multiple decision trees for robust predictions.
  • Multi-Layer Perceptron (MLP): Applied a neural network with hidden layers for complex pattern recognition.

2. Granger Causality Analysis

To assess potential causality between economic factors and asylum acceptance rates, a Granger causality test was conducted. The objective was to identify leading economic indicators that could potentially forecast future asylum rates based on lagged data.

Results and Analysis

Machine Learning Models:

  • The Random Forest model outperformed other models, achieving the highest R2 score of 0.688 and a root mean squared error (RMSE) of 0.554. The most significant predictors identified were GDP per capita, population structure, and tertiary education rates.
  • The ElasticNet model showed suboptimal performance with an R2 score of 0.295 and an RMSE of 0.833, indicating that simple linear relationships were insufficient for accurately predicting asylum acceptance rates.

Granger Causality Analysis:

  • The analysis revealed limited causal relationships between economic factors and asylum acceptance rates. While some indicators, such as government debt and unemployment rates, exhibited lagged predictive capabilities, the overall Granger causality test indicated that economic factors alone could not robustly forecast asylum decisions.

Limitations and Future Work

  • Data Limitations: Data availability was inconsistent, particularly for asylum application statistics and certain economic indicators.
  • Model Constraints: While Random Forest exhibited promising predictive power, further optimization and integration of additional factors, such as political climate and refugee origin, could enhance forecasting accuracy.
  • Granger Causality: The linear nature of Granger causality may not fully capture complex interactions, suggesting the potential for multivariate causality models in future work.

Conclusion

This project demonstrated that while economic factors can influence asylum acceptance rates to a certain extent, their predictive power remains limited. The Random Forest model provided the most robust predictions but was constrained by data quality and feature selection. Future work could explore incorporating socio-political factors and advanced causal inference techniques to refine predictive capabilities.

For the full presentation, please access the slides here.