Econometric and Statistical Methods

Econometric and statistical methods furnish a structured toolkit for quantifying relationships, testing causal mechanisms and uncovering latent structure in data. Core approaches include regression frameworks—ranging from ordinary least squares and maximum-likelihood estimation to generalised method of moments and Bayesian inference—which form the backbone of empirical modelling. Time-series techniques address persistence and equilibrium via cointegration tests, error-correction models and vector autoregressions. Panel-data methods exploit both cross-sectional and temporal variation through difference and system GMM, quasi-ML estimators and factor-structure specifications to control unobserved heterogeneity and cross-sectional dependence. Causal inference has advanced with regression discontinuity designs, instrumental variables and synthetic controls, while nonparametric and semiparametric tools—kernel density estimates, spline smoothers and mixture-of-experts architectures—offer flexible alternatives when parametric assumptions fail. Specialized families of models handle discrete outcomes (negative binomial, Conway–Maxwell–Poisson, generalised Poisson) or accommodate underdispersion and excess zeros via hurdle and zero-inflated formulations. Mixture models, estimated by expectation–maximisation or variational inference, elucidate hidden subpopulations in clustering and density estimation. Sequential and multistage sampling procedures deliver fixed-width confidence intervals and risk-bounded estimators in contexts where data arrive over time. Together, these developments underpin rigorous analysis across economics, finance, epidemiology and beyond.

Research from Nature Portfolio

A new clustering algorithm addresses the risk of “inferior” solutions when Gaussian mixture models are misspecified. By introducing a penalty term targeting component asymmetry and poor interpretability, researchers developed a selection criterion and clustering procedure (SIA) that avoids both classical “spurious” and newly identified “inferior” high-likelihood fits. Empirical studies demonstrate SIA’s improved performance under contamination and model misspecification.

Comprehensive updates to the social cost of carbon integrate probabilistic socioeconomic projections, advanced climate models, revised damage functions and consistent risk-valued discounting. The resulting open-source framework produces mean estimates of the social cost of carbon that markedly exceed current policy benchmarks, underscoring the sensitivity of climate–economy assessments to statistical calibration and transparent uncertainty treatment.

Research from all publishers

A three-stage sequential design for Rayleigh-scale estimation minimises sampling operations by blending bulk and incremental draws. This method simultaneously delivers point estimates and fixed-width confidence intervals with asymptotic second-order efficiency, as confirmed by Monte Carlo simulations across small to large sample regimes.

For firm-level panel data with short time dimensions, a transformed quasi-maximum likelihood estimator robustly handles individual, time and interactive effects. This approach delivers consistent, asymptotically normal parameter estimates under heterogeneous initial conditions and unobserved common factors, with fast convergence even when both dimensions grow, and extends to nonlinear settings with bias correction.

Under-dispersed count data are effectively modelled by the generalised Poisson distribution via two novel MM algorithms. One algorithm targets maximum-likelihood estimation of distribution parameters without covariates; the other extends to covariate-driven mean regression. The methods also yield likelihood-ratio, Wald and score tests, overcoming longstanding challenges in underdispersion modelling.

Econometric and Statistical Methods publication trend

The graph below shows the total number of articles in econometric and statistical methods across all publications each year (not limited to Nature Index journals).

Technical terms

Expectation–maximisation (EM) algorithm: An iterative procedure to compute maximum-likelihood estimates in models with latent variables by alternating expectation and maximisation steps.

Conway–Maxwell–Poisson distribution: A two-parameter generalisation of the Poisson distribution that flexibly captures overdispersion and underdispersion in count data.

Generalised method of moments (GMM): An estimation technique that exploits moment conditions from model assumptions to obtain consistent parameter estimates under endogeneity.

Regression discontinuity (RD) design: A quasi-experimental causal inference method comparing observations just above and below a predetermined threshold in a forcing variable.

Fixed-width confidence interval: A sequential estimation goal that terminates sampling once the interval around an estimate reaches a prespecified half-width.

References

1. Avoiding inferior clusterings with misspecified Gaussian mixture models. Scientific Reports (2023).
2. Comprehensive evidence implies a higher social cost of CO2. Nature (2022).
3. Multistage Estimation of the Scale Parameter of Rayleigh Distribution with Simulation. Symmetry (2020).
4. Short T dynamic panel data models with individual, time and interactive effects. Journal of Applied Econometrics (2023).
5. Modeling Under-Dispersed Count Data by the Generalized Poisson Distribution via Two New MM Algorithms. Mathematics (2023).