This repository contains the complete data, code, trained models, intermediate outputs, robustness checks, and figures supporting the manuscript “Regime-Aware Reinforcement Learning for Inflation Forecast Combination in Sub-Saharan Africa.” The empirical dataset covers eight Sub-Saharan African economies—Côte d’Ivoire, Ghana, Kenya, Nigeria, Senegal, Tanzania, Uganda, and South Africa—from 1972 to 2024. The dataset includes inflation measures and key macroeconomic variables together with their lagged values.The repository provides the full Python analysis pipeline used to reproduce the empirical results, including pooled OLS, fixed-effects regression, Elastic Net, Random Forest, XGBoost, country-specific Markov-switching regime estimation, and the regime-aware Q-learning forecast-combination framework. The final reinforcement-learning configuration uses country-specific regime probabilities, an absolute-error reward function, temperature parameter τ = 2.0, learning rate α = 0.5, discount factor γ = 0.3, exploration rate ε = 0.15, and 400 training episodes.Also included are saved model objects, the final Q-table and metadata, country-level regime probabilities, out-of-sample predictions, model comparison results, multi-seed stability tests, regime/reward ablation analyses, and hyperparameter-sensitivity results. The repository further contains all figures used to document model performance, feature importance, regime dynamics, reinforcement-learning choices, Q-table structure, and robustness analyses.The files are intended to support transparency, verification, replication, and extension of the study. The main executable file is 01_rl_regime_pipeline.py, while ssa_panel_macro_dataset.csv contains the principal analysis dataset and FINAL_test_results.csv reports the final out-of-sample performance of all benchmark and ensemble forecasting models
This repository contains the complete data, code, trained models, intermediate outputs, robustness checks, and figures supporting the manuscript “Regime-Aware Reinforcement Learning for Inflation Forecast Combination in Sub-Saharan Africa.” The empirical dataset covers eight Sub-Saharan African economies—Côte d’Ivoire, Ghana, Kenya, Nigeria, Senegal, Tanzania, Uganda, and South Africa—from 1972 to 2024. The dataset includes inflation measures and key macroeconomic variables together with their lagged values.The repository provides the full Python analysis pipeline used to reproduce the empirical results, including pooled OLS, fixed-effects regression, Elastic Net, Random Forest, XGBoost, country-specific Markov-switching regime estimation, and the regime-aware Q-learning forecast-combination framework. The final reinforcement-learning configuration uses country-specific regime probabilities, an absolute-error reward function, temperature parameter τ = 2.0, learning rate α = 0.5, discount factor γ = 0.3, exploration rate ε = 0.15, and 400 training episodes.Also included are saved model objects, the final Q-table and metadata, country-level regime probabilities, out-of-sample predictions, model comparison results, multi-seed stability tests, regime/reward ablation analyses, and hyperparameter-sensitivity results. The repository further contains all figures used to document model performance, feature importance, regime dynamics, reinforcement-learning choices, Q-table structure, and robustness analyses.The files are intended to support transparency, verification, replication, and extension of the study. The main executable file is 01_rl_regime_pipeline.py, while ssa_panel_macro_dataset.csv contains the principal analysis dataset and FINAL_test_results.csv reports the final out-of-sample performance of all benchmark and ensemble forecasting models
REINDOLPH OSEI-ANIM· Figshare· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
