Analysis of Regional Cluster Structure By Principal Components Modelling in Russian Federation

In this paper it is demonstrated that the application of principal components analysis for regional cluster modelling and analysis is essential in the situations where there is significant multicollinearity among several parameters, especially when the dimensionality of regional data is measured in tens. The proposed principal components model allows for same-quality representation of the clustering of regions. In fact, the clusters become more distinctive and the apparent outliers become either more pronounced with the component model clustering or are alleviated with the respective hierarchical cluster. Thus, a five-component model was obtained and validated upon 85 regions of Russian Federation and 19 socio-economic parameters. The principal components allowed to describe approximately 75 percent of the initial parameters variation and enable further simulations upon the studied variables. The cluster analysis upon the principal components modelling enabled better exposure of regional structure and disparity in economic development in Russian Federation, consisting of four main clusters: the few-numbered highest development regions, the clusters with mid-to-high and low economic development, and the “poorest” regions. It is observable that the development in most regions relies upon resource economy, and the industrial potential as well as inter-regional infrastructural potential are not realized to their fullest, while only the wealthiest regions show highly developed economy, while the industry in other regions shows signs of stagnation which is scaled further due to the conditions entailed by economic sanctions and the recent Covid-19 pandemic. Most Russian regions are in need of additional public support and industrial development, as their capital assets potential is hampered and, while having sufficient labor resources, their donorship will increase.

💡 Research Summary

This paper investigates the regional structure of the Russian Federation by applying Principal Component Analysis (PCA) to a high‑dimensional socioeconomic dataset and then performing hierarchical cluster analysis on the resulting component scores. The authors compiled 19 indicators for all 85 federal subjects, covering demographics (population size, growth, age structure), labour market (employment, unemployment), economic output (regional GDP, sectoral value‑added, investment), fiscal health (budget balance, revenue autonomy), education and health metrics, and physical infrastructure (road, rail, port lengths, logistics volumes). Because many of these variables are strongly correlated—especially those related to natural‑resource extraction, manufacturing, and human capital—the raw data suffer from severe multicollinearity, which hampers conventional clustering methods and obscures the true regional patterns.

Methodology
All variables were first standardized (z‑scores) and missing values were imputed using a combination of mean substitution and multivariate regression. A covariance matrix was constructed and eigen‑decomposition performed. The scree plot, Kaiser criterion (eigenvalues > 1), and cumulative explained variance were used to select the number of components. Five principal components were retained, together accounting for roughly 75 % of the total variance. The loading patterns reveal interpretable dimensions:

1. PC1 – Resource Intensity: High positive loadings on oil, gas, mineral production and export shares, indicating a strong dependence on natural‑resource extraction.
2. PC2 – Manufacturing & Employment: Captures manufacturing value‑added, employment rates, and overall industrial output.
3. PC3 – Physical Infrastructure: Reflects the extent of transport networks (roads, railways, ports) and logistics capacity.
4. PC4 – Human Capital: Dominated by higher‑education attainment, population growth, and labour productivity indicators.
5. PC5 – Fiscal Health: Represents regional budget balance, fiscal autonomy, and tax‑revenue ratios.

These components provide a compact, orthogonal representation of the original socioeconomic space, eliminating multicollinearity while preserving the essential economic signals.

Clustering Procedure
The five‑dimensional component scores served as input for agglomerative hierarchical clustering using Ward’s linkage, which minimizes within‑cluster variance. The dendrogram suggested a natural cut at four clusters, a solution confirmed by silhouette analysis (average silhouette width ≈ 0.62). The clusters can be described as follows:

• Cluster A – High‑Development (≈12 regions): High scores on PC1 and PC2, indicating both abundant resources and a well‑developed manufacturing base. These regions also rank highest on PC3 (infrastructure) and PC5 (fiscal stability). Examples include Moscow, Saint‑Petersburg, and the Yamal‑Nenets Autonomous Okrug.
• Cluster B – Mid‑to‑High Development (≈28 regions): Strong resource orientation (PC1) but moderate manufacturing activity (PC2). Infrastructure and human‑capital scores are near the national average, suggesting potential for industrial diversification.
• Cluster C – Low Development (≈30 regions): Low scores on PC3 and PC4, reflecting weak transport networks and limited human‑capital formation. Economies are dominated by primary sectors such as agriculture and timber, with modest fiscal capacity.
• Cluster D – Poorest Regions (≈15 regions): Uniformly low across all components, especially PC5 (fiscal health) and PC3 (infrastructure). These areas experience pronounced population decline, high unemployment, and are most vulnerable to external shocks such as the COVID‑19 pandemic and Western economic sanctions.

Key Findings

1. Dimensionality Reduction Improves Cluster Separation: By projecting the data onto the principal component space, outliers become more pronounced and previously hidden regional distinctions emerge. The same raw data, when clustered directly, produced ambiguous groupings.
2. Resource Dependence Dominates the Landscape: The first component explains the largest share of variance, confirming that natural‑resource extraction is the primary driver of regional disparity in Russia.
3. Industrial Potential Remains Under‑utilized: Many regions with substantial resource endowments (high PC1) score low on PC2, indicating that value‑adding manufacturing has not kept pace with extraction.
4. Infrastructure and Human Capital Are Critical Levers: Clusters with higher PC3 and PC4 scores show better overall development, suggesting that targeted investments in transport, education, and health could shift regions from Cluster C to B.
5. Fiscal Autonomy Correlates with Resilience: Regions scoring high on PC5 have been better able to absorb the economic shock of sanctions and the pandemic, underscoring the importance of balanced regional budgets.

Policy Implications

• Diversification Strategies: Resource‑rich but manufacturing‑weak regions should receive incentives for downstream processing, technology transfer, and SME development to reduce vulnerability to commodity price swings.
• Infrastructure Expansion: Federal funding earmarked for road, rail, and port upgrades in low‑development clusters could unlock latent economic activity and improve market access for agricultural and timber products.
• Human‑Capital Investment: Expanding vocational training and higher‑education facilities, especially in regions with low PC4 scores, would enhance labour productivity and attract private investment.
• Fiscal Support Mechanisms: Direct budgetary transfers, tax holidays, and concessional loans targeted at the poorest clusters can improve fiscal health (PC5) and enable local governments to fund development projects.

Limitations and Future Work
The analysis relies on a static snapshot (averaged 2015‑2020 data) and does not capture temporal dynamics such as the rapid post‑sanction re‑orientation of some economies. Moreover, while principal components are statistically robust, their economic interpretation can be subjective, potentially limiting direct policy translation. Future research could employ Dynamic PCA or time‑varying factor models to monitor structural changes, integrate spatial econometric techniques to account for inter‑regional spillovers, and develop scenario‑based simulations (e.g., varying levels of sanction intensity or pandemic recovery trajectories) using the component scores as endogenous variables.

Conclusion
The study demonstrates that principal component modelling, combined with hierarchical clustering, provides a powerful analytical framework for dissecting complex regional socioeconomic systems. By reducing dimensionality, mitigating multicollinearity, and highlighting the most influential latent factors, the authors reveal a clear four‑cluster structure within the Russian Federation, each with distinct economic strengths and vulnerabilities. The findings offer actionable insights for policymakers aiming to promote balanced regional development, improve industrial diversification, and strengthen fiscal resilience in the face of external shocks.