Threshold Estimation in Regression Models

Threshold estimation in regression models concerns identifying points at which the relationship between independent and dependent variables shifts distinctively. These models extend linear frameworks to accommodate abrupt or smooth changes in slope or level beyond chosen threshold values, enabling nuanced characterisation of nonlinear dynamics. Applications span economics, environmental science, biology and engineering, where phenomena such as policy impact, ecosystem tipping points, biomarker response and market regimes exhibit regime-specific behaviour. Estimation strategies range from parametric and semiparametric methods to fully nonparametric approaches, often involving likelihood or least-squares criteria modified to accommodate non-differentiability at change-points. Recent advances address issues of endogeneity, high-dimensional covariates, stochastic threshold variables and uncertainty quantification via bootstrapping or Bayesian schemes. Such developments enhance robustness against model misspecification, improve computational tractability and broaden applicability to panel and time series data. The resulting threshold estimates inform decision-making by pinpointing critical transition values and quantifying heterogeneous effects across regimes.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Recent advances in control function techniques have extended endogenous threshold regression by embedding threshold effect information in both conditional mean and variance. New estimators correct bias in structural threshold estimators and employ generalised method of moments to improve inference, demonstrating superior finite-sample performance in simulations and practical applications in international trade data.

A nonparametric panel threshold regression framework has been proposed to examine firm productivity, allowing threshold variables to be endogenous and functional forms to be unspecified. Estimation combines extreme kernel threshold detection with local linear regression, and standard errors are obtained via bootstrap. The approach reveals ownership-specific turning points in leverage-productivity relationships, illustrating regional productivity disparities.

Innovations in segmented regression include a geometric optimisation method for breakpoint detection that minimises residual dispersion by fitting adjacent polynomial surfaces. This multidimensional paraboloid approach yields more accurate threshold abscissas in simulations and real-world datasets, enhancing predictive accuracy and interpretability in growth modelling.

Threshold Estimation in Regression Models publication trend

The graph below shows the total number of articles in threshold estimation in regression models across all publications each year (not limited to Nature Index journals).

Technical terms

Threshold regression model: A model allowing different functional relationships between variables on either side of a threshold point.

Change-point: The value of a predictor at which the regression relationship shifts between regimes.

Segmented regression: A piecewise technique fitting separate lines or curves to data segments joined at breakpoints.

Control function: A method for correcting endogeneity by modelling the determinant of an endogenous variable within the regression.

Nonparametric estimation: An approach that imposes minimal assumptions on functional form, often using kernel smoothing to estimate relationships.

References

1. NEW CONTROL FUNCTION APPROACHES IN THRESHOLD REGRESSION WITH ENDOGENEITY. Econometric Theory (2023).