📈 economics

Contextual Factors Behind Income Inequality: A Predictive Analysis

Using machine learning on data from 88 countries, this study identifies regional classification as the dominant predictor of income inequality across various measures, followed by political and technological factors, while demonstrating that a parsimonious set of indicators can effectively capture the predictive power of a broader model.

Original authors: Arda Aktas, Miguel Poblete Cazenave, Narasimha Rao

Published 2026-09-21

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Arda Aktas, Miguel Poblete Cazenave, Narasimha Rao

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

For decades, economists have watched a persistent puzzle unfold: why do some nations grow rich while their citizens remain far apart in wealth, while others manage to share prosperity more evenly? The question is not just about numbers on a page; it is about who gets to eat, who gets a home, and who has a voice in how society is run. Early theories suggested that as a country gets richer, inequality would naturally rise and then fall, like a hill that everyone eventually climbs. But in recent years, that pattern has broken down. In many places, wealth has piled up at the very top while the rest of the population has been left behind, even as the overall economy grows. This shift has forced researchers to look beyond simple economic growth and ask what other forces are at play. Is it the way technology changes jobs? Is it how governments distribute power? Or is it something deeper, rooted in the history and location of a country itself?

A team of researchers from the Netherlands and the United States set out to answer these questions by treating income inequality not as a single mystery to be solved, but as a complex picture that looks different depending on how you view it. They gathered data from 88 countries spanning nearly fifty years, from 1970 to 2017. Instead of relying on a single number to describe inequality, they looked at six different ways to measure it. Some measures focused on the gap between the richest and the poorest, while others zeroed in on just the top one percent of earners or the top ten percent. By using a powerful computer learning tool that can spot patterns in massive amounts of data without forcing them into a rigid box, the team tested thirty-five different factors. These factors ranged from the level of education and the size of the financial sector to the strength of political institutions and the sophistication of a country's technology.

The most striking discovery was that where a country is located on the map matters more than almost any other single factor. The researchers found that grouping countries by their political and geographic regions was the strongest predictor of inequality. A country's neighbors and its shared history seemed to create a pattern of wealth distribution that was hard to break. This does not mean that geography is destiny in a simple sense, but rather that countries within the same region often share deep-seated political, legal, and institutional legacies that shape how money is distributed. When the researchers removed these regional labels from their analysis, their ability to predict inequality dropped significantly, suggesting that these broad regional characteristics hold a key to understanding the problem that other factors alone cannot explain.

Beyond location, the study highlighted two other powerful drivers: the nature of political institutions and the state of technology. Countries with political systems that ensure equal rights, fair access to power, and a more even distribution of resources tended to have lower levels of income inequality. This held true whether the researchers were looking at the gap between the rich and poor or the concentration of wealth at the very top. Technology also played a major role, but its effect depended on which part of the income distribution was being examined. Advances in productivity and the sophistication of a country's exports were particularly strong predictors of how much wealth the top one percent held. This suggests that as technology makes the economy more complex and efficient, it tends to concentrate the rewards at the very top, even if the overall system is working well.

The researchers also tested whether the factors that drive inequality change depending on how you measure it. They found that while some drivers, like regional location and political fairness, were important across the board, others were more specific. For instance, the gap between the returns on money invested in capital and the rate of economic growth was more relevant for measuring the distance between the top and bottom of the income ladder. In contrast, the level of education in a country mattered, but the researchers found that how that education was shared among the population was a better predictor than the average number of years spent in school. Surprisingly, factors often blamed for inequality, such as the size of the financial sector or the degree of global trade, were less consistent in their ability to predict the outcome.

One of the most practical takeaways from this work is that you do not need a massive, complicated list of variables to understand the big picture of inequality. The team showed that by selecting just one representative factor from each major category—such as one measure of politics, one of technology, and one of education—they could capture almost all the predictive power of their full, complex model. This means that policymakers and observers can get a clear, reliable view of inequality trends without getting lost in a sea of data. The study confirms that while the drivers of inequality are varied, a few key themes stand out: the deep influence of regional history, the critical role of fair political systems, and the double-edged sword of technological progress. These findings offer a clearer map for understanding why some societies struggle to share their wealth, pointing the way toward a more focused conversation on how to fix it.

Technical Summary: Contextual Factors Behind Income Inequality: A Predictive Analysis

Problem Statement
Income inequality remains a central concern in economics and public policy, yet empirical conclusions regarding its drivers often vary depending on the specific inequality measure employed and the set of indicators considered. Traditional literature, ranging from the Kuznets curve to Piketty's capital accumulation arguments, suggests that factors such as human capital, technology, globalization, financial development, and political institutions influence inequality. However, these factors are broad conceptual categories often represented by diverse indicators, and their predictive relevance may differ across various parts of the income distribution (e.g., top-income concentration vs. gaps between the upper and lower tails). This study addresses the need to identify which specific indicators within these broad categories are most informative for predicting future income inequality across different measures, moving beyond single-metric analyses to a comparative predictive framework.

Methodology
The authors employ a machine learning approach using the XGBoost (Extreme Gradient Boosting) framework to assess the relative predictive importance of 35 candidate indicators across 88 countries from 1970 to 2017. The dependent variable is defined as next-year inequality, measured using six distinct metrics from the World Income Inequality Database (WIID): the Gini coefficient, the income shares of the top 10% and top 1%, the Palma Ratio, the S80/S20 ratio, and the P90/P10 ratio.

The analysis proceeds through three complementary model specifications:

Full Specification: An unrestricted model including all 35 candidate indicators to rank their relative predictive importance for each inequality measure.
Category-Representative Specification: A reduced model selecting one common indicator from each of six conceptual categories (financial sector, human capital, globalization, political institutions, regional aggregation, and technology) based on their average predictive importance across all six inequality measures.
Inequality-Specific Specification: A constrained model search allowing the optimal predictor within each conceptual category to vary across the six different inequality measures.

The models are estimated using a random 70% training sample over 50 bootstrap iterations to ensure robustness. Feature importance is derived from XGBoost scores, normalized within each inequality measure to allow for comparison.

Key Contributions
The paper makes three primary contributions:

Simultaneous Evaluation: It evaluates a broad range of factors emphasized in the literature (political institutions, human capital, globalization, financial development, and technology) within a unified predictive framework.
Multi-Metric Analysis: By examining six alternative inequality measures, the study assesses whether the drivers of overall inequality differ from those driving top-income concentration or relative income gaps, providing a more nuanced view of distributional dynamics.
Predictive Perspective: Adopting a predictive rather than causal perspective, the study identifies which contextual factors consistently contain information about future inequality outcomes and determines whether a parsimonious set of indicators can retain the predictive power of a full model.

Results

Dominance of Regional Classification: Across all specifications and inequality measures, regional classification (specifically politico-geographic and World Bank regional groupings) emerges as the dominant predictor. This suggests that broad historical, political, and geographic contexts are strongly linked to cross-country inequality differences. However, the broad nature of this variable limits the identification of specific underlying regional characteristics.
Consistency of Political and Technology Indicators: Political institutions and technology provide the most consistent additional predictive information. The egalitarian component index of political institutions ( $v2x\_egal$ ) is particularly informative for the Gini coefficient, Palma ratio, and S80/S20 ratio. Technology indicators, specifically alternative Total Factor Productivity measures ( $tfp\_alt$ ), are highly informative for top-income concentration (top 1% share and P90/P10 ratio).
Variable Importance of Other Factors: The importance of human capital, financial development, and globalization varies across inequality measures. Educational inequality ( $e\_peedgini$ ) is selected more frequently than average years of education, suggesting distribution matters more than the mean. The ratio of returns to capital to economic growth ( $r\_g$ ) is more predictive for measures capturing gaps between the upper and lower distribution (S80/S20, P90/P10). Financial sector and globalization indicators show moderate to low predictive importance compared to other categories.
Model Fit and Parsimony: All specifications demonstrate high predictive fit ( $R^2 > 0.98$ ). The category-representative specification retains most of the predictive information of the full model, indicating that cross-country inequality can be analyzed using a more parsimonious set of indicators without significant loss of predictive power.

Significance and Claims
The paper claims that the predictors of income inequality are neither entirely common nor completely measure-specific. While regional aggregation, political institutions, and technology provide the most consistent predictive information, the relevance of other factors depends on the specific dimension of inequality being measured. The study concludes that informative predictions of income inequality can be achieved without relying on a large, complex set of indicators. By identifying which indicators are most informative for different dimensions of the income distribution, the approach supports more focused inequality monitoring.

The authors explicitly note limitations: the analysis is predictive, not causal; regional classification, while highly predictive, is a broad proxy that does not reveal specific underlying mechanisms; and the use of country-level data cannot fully capture within-country household or individual mechanisms. The findings are intended to complement, not replace, causal identification efforts.

Technical Summary: Contextual Factors Behind Income Inequality: A Predictive Analysis

More like this