Introduction

Human economic behavior is complex in design, which makes it difficult to conform to the simplifying assumptions of classical linear regression models (CLRM), particularly that their behaviors are assumed to be normal in nature, and the relationships depict a constant and linear pattern across the distribution of the dependent variable. For example, in income-consumption relationships, the marginal propensity is likely to change when an income cluster of households changes, creating heterogenous behavioral patterns with ordinary least squares (OLS), which may, in turn, oversimplify using conditional mean as a point estimate of the relationship (Koenker and Bassett 1978). This behavior is denoted as distributional heterogeneity, which can be observed in other economic relationships like monetary policy transmission, inflation dynamics, and economic growth, whereby the point estimate may also depend on the economic state and time period.

The methodological limitations of OLS include the presumption that the slopes are constant for any distributional position of the dependent variable, which led economists to explore complex models. While OLS was the best linear unbiased estimator (BLUE) under Gauss–Markov`s assumptions, the use of the mean as a point estimate makes the model over restrictive and ignores the information at the tails and within subclusters of data (Buchinsky 1998). Non-normality of residuals coming from skewed or extreme value data may lead to issues for inference, especially when the data may have subdistributional groups. Traditionally, data transformations, interaction terms, and polynomial specifications are tested to absorb this distributional heterogeneity, but these approaches can only capture the distributional heterogeneity from the independent variable side and may not fully comprehend the complexity of the relationship, which is distributional position–dependent.

Quantile regression (QR) was introduced by Koenker and Bassett (1978) as an alternative approach that provides conditional quantile functions. This methodology can examine how explanatory variables are affected differently at different points in the distribution of outcome variables, revealing heterogenous treatment effects which were hidden in OLS-based models. QR had thus emerged as an alternative statistical tool for an in-depth complete assessment of stochastic relationships. This model is also flexible, in cases, where the data are normally distributed, QR estimates equate to OLS estimates. There were several phases in the evolution of QR models. It started in the 1980s for cross-sectional contexts, then extended into time-series and panel data. The recent variants include quantile autoregressive distributive lagged (ARDL) models, quantile vector autoregressive (VAR) systems, and quantile on quantile (QonQ) models. Recent development also introduced quantile estimation within the generalized method of moments (GMM) to address endogeneity.

Contemporary economic analysis involves confronting datasets that are heavy on tails, distributional asymmetries, and heterogenous subgroups, especially in the case of microdata with repeated annual surveys like Multi-Indicator Cluster Surveys (MICS), World Values Surveys (WVS), Global Entrepreneurship Monitor Surveys (GEM), and so forth. In such cases, QR is a useful approach by providing flexibility, robustness, and insights into conditional relationships. The flexibility can be understood in the scenario of policy intervention for income inequality, where changes in income distribution/cluster may have differential effects. In economics, there can be scenarios where the generic behavior of the dependent variable is expected to change (like the tax-backets effect and cases that use piecewise regression where the breaks could be multiple or unknown) with the size of the dependent variable. While all the functional form transformations can be adapted in QR models, it can further extend the model in terms of the transformation of the dependent variable, where OLS is silent. In such cases, QR can perform better than OLS. Other examples include the case of financial contagions, a tail-dependent that requires robust risk assessment.

The development of QR models is not free of computational and performance challenges. Computational complexity is substantially increased when implementing QR models in panel data or time series data, when compared to cross-sectional data. The inclusion of dynamic effects in the presence of fixed effects induces biasness. The interpretation of the QR model must also be carefully handled, especially when the data are skewed and when the quantile slopes are connected to their economic meanings across quantiles. This step may only represent a selection of effects rather than handling distribution heterogeneity. Lastly, there is a debate about how to decide the number of quantiles to be estimated. The use of statistical and economic reasons has its own trade-offs in handling distributional analysis versus parsimony in interpretation and implementation.

Practically, a QR model in policy analysis raises additional consideration regarding connecting quantile effects with actionable policy recommendations. While QR points towards heterogenous policy, it does not help how policy makers can retrofit this into their intervention design. Recent studies also integrated machine learning models with QR, which improved the performance of these robust models (Arshed, Bakkar, et al. 2025). However, this setup has its own interpretability challenges. Still, there are a growing number of studies that recognize QR as a complementary model rather than a replacement for OLS. Research objectives, data characteristics, and theoretical flexibility are primary supports that motivate the selection of OLS or QR.

This study reviews the evolution, application, and methodological development of QR in economics. By adapting a critical perspective in acknowledging the contribution and limitations of QR, this study presents the comparative advantage of QR and OLS. We conduct systematic reviews on empirical applications of QR to enhance understanding within economic research. The study starts with a genealogical approach that discusses the methodological evolution of Koenker and Bassett`s (1978) original formulation. A further examination includes how the economic research community has adapted or modified QR models. This study synthesized the broader trends in econometrics that gained from the flexibility and robust nature of QR that can handle theoretical rigor and data complexity. The outcomes are categorized in terms of how the QR methods contribute to the literature.

While we discuss the growing reputation of QR in terms of flexibility in data handling and inference beyond the estimation of slopes, this model is still considered underdeveloped in economics. This study used the concept development frameworks of Howie and Bagnall (2020) and Podsakoff et al. (2016) to synthesize the QR model as a distribution-sensitive econometric assessment and presents its methodological evolution, empirical scope, and future development. Thus, this study moves beyond a descriptive inventory to a theory-oriented synthesis of quantile-based econometrics.

This study conducts the SLR using guidelines provided by studies like Okoli (2015), Sauer and Seuring (2023), and Tingelhoff et al. (2025). The Scopus database is used to find relevant studies. Using a specified search query, this study extracted 47 studies between 2011 and 2025. The next section details some prominent recommendations extracted from the SLR. This review positioned QR not only as a methodological contribution but also as a theoretical bridge that can broaden the econometric modeling frameworks with dynamic capabilities and increased stakeholder involvement by subcluster policy evaluation.

Full Article: https://onlinelibrary.wiley.com/doi/full/10.1111/joes.70057