حسين شايسته

The persistent increase in markups over recent decades has raised serious concerns regarding declining competition, deepening inequality, an‎d the fragility of economic resilience. However, most existing approaches rely on ex-post measurement an‎d depend on micro-level data that are not available in many countries. This study, adopting a forward-looking, prediction-oriented approach based on macroeconomic an‎d structural data, proposes a novel framework for forecasting markups. To this end, harmonized data from 126 economies in the food sector over a 20-year period were employed, an‎d the markup forecasting task was redefined as a supervised learning problem. The proposed architecture consists of a heterogeneous ensemble of base models, including boosted trees, additive ensemble models, an‎d neural networks, which are combined through an ElasticNet meta-learner weighted by an attention mechanism. This mechanism allows the contribution of each model to be dynamically adjusted according to structural conditions an‎d data context, thereby better capturing country heterogeneity an‎d nonlinear interactions. Empirical results demonstrate that the proposed framework achieves higher accuracy an‎d stability in forecasting compared to individual models an‎d static ensembles. Furthermore, interpretability is provided at two levels: variable importance analysis using SHAP values, an‎d examination of the model weighting logic through the attention mechanism. The findings revealed that production scale, labor, capital, an‎d price indices are the most important drivers of markup variations, with significant differences observed across income groups an‎d regions. These results indicate that the use of interpretable machine learning can serve as a credible complement to classical econometric methods, enabling scalable an‎d transparent forecasting of market power in data-constrained environments. Ultimately, the developed framework, in addition to its methodological innovation, carries important policy implications for monitoring competition, regulating markets, an‎d guiding development planning at the global level.

مارك‌آپ , پيش‌بيني اقتصادي , يادگيري ماشين , تحليل تفسيرپذير , اقتصاد كلان

Markup , economic forecasting , machine learning , interpretable analysis , macroeconomics

Hossein Shayesteh

Dr. Mahdi Abdolhamid