In this project, we develop and evaluate forecasting models to predict the probability of staff leaving the Royal Berkshire NHS Foundation Trust (RBFT) over horizons ranging from one to twelve months. We begin with conventional predictive models to identify key determinants of staff turnover and progressively implement advanced machine-learning approaches to enhance predictive performance. The final Balanced Random Forest models perform strongly across all forecasting horizons. At the 12-month horizon, the model achieves approximately 99.7% overall accuracy, recall close to 100%, and precision of around 96%, correctly identifying nearly all future leavers while maintaining a low false-positive rate. Performance remains similarly robust at the 6-month horizon, with precision around 90% and recall close to 100%. In line with management priorities, we also implement forecasting under a two-month notice-period setting, reflecting typical RBFT practice. Under this configuration, the model attains approximately 99.6% accuracy and around 94% precision, correctly flagging more than 99% of staff who subsequently leave. Cross-validation and expanding-window testing confirm that model performance remains stable and robust over time. The resulting framework functions as a practical early-warning system for staff retention. By identifying at-risk employees in advance, managers can intervene earlier, target support effectively, and reduce recruitment and training costs.

We examine whether corporate green bond issuance causally reduces firms’ greenhouse gas (GHG) emissions intensity and whether linguistic structure shapes this environmental impact. Using an unbalanced panel of 3,867 publicly listed non-financial firms across 65 countries (2010–2024), we implement a staggered Difference-in-Differences design, complemented by entropy balancing to strengthen comparability between treated and control firms. We find that green bond issuance leads to economically meaningful emission reductions of approximately 19–21% on average. The effects are stronger in carbon-intensive industries, among high-investment firms, and for long-maturity bonds, suggesting real operational adjustments rather than symbolic commitments. Crucially, we document a novel moderation channel: firms headquartered in countries with weak Future Time Reference (FTR) languages experience significantly larger emission reductions following issuance. This linguistic effect remains robust when using a CEO-level Weak-FTR index constructed via machine-learning-based phonetic classification. Our findings demonstrate that financial instruments and cultural-linguistic structures jointly shape corporate environmental outcomes.

Ethnic diversity is frequently debated as either a structural source of instability or a foundation for long-term social resilience. However, existing research largely relies on static correlations and provides limited insight into how sudden increases in diversity affect conflict dynamics over time. Using panel data for 152 countries from 1990–2013, we conceptualize entry into the top quintile of the Historical Index of Ethnic Fractionalization (HIEF) as an exogenous diversity “shock” and estimate its dynamic effects on intra-state armed conflict. We find clear causal evidence that diversity shocks increase conflict intensity. Following a shock, conflict dyads rise significantly and remain elevated for approximately six years before gradually attenuating. This dynamic escalation–adaptation pattern constitutes a novel contribution to the literature, which has not previously clarified the temporal adjustment process following large shifts in ethnic composition. Mechanism analyses reveal substantial heterogeneity across regions and development levels. The effect is most pronounced in Asia and the Middle East, moderate in the Americas and Europe, and weaker elsewhere. Countries with lower Human Development Index (HDI) levels experience stronger post-shock conflict escalation, while higher-HDI societies demonstrate faster stabilization. Diversity shocks are also associated with higher conflict-related mortality rates, indicating effects not only on incidence but also on intensity. Our findings remain robust across two-way fixed-effects models, staggered difference-in-differences event studies, regression discontinuity designs around the diversity threshold, and alternative panel specifications. By isolating the dynamic shock component of ethnic diversity, our study clarifies when and where diversity destabilizes societies and how institutional capacity shapes subsequent adaptation.