Yassin Alaya

Toulouse School of Economics

Working papers

The Divide-and-Conquer Estimator for High-Dimensional Time Series Regression under Weak Exogeneity Job market paper

Last revised 30 October 2026.

Abstract

Mikusheva and Sølvsten (2025) show that ordinary least squares is generally inconsistent for weakly exogenous time-series regressions when the number of regressors is proportional to the sample size, and propose a new estimator for high-dimensional time-series regressions. Consistency of this estimator, however, relies on homoskedasticity, and inference results are limited. We propose the Divide-and-Conquer (DaC) estimator as an alternative. It is consistent under heteroskedasticity when feedback from the error term to future regressors has a known, finite horizon. The idea is simple: partition the sample into subsamples, spaced so that strict exogeneity holds within each, estimate OLS on every subsample, and average. The resulting estimator is unbiased and consistent for linear contrasts of the slope parameter when the number of regressors is proportional to the sample size. Asymptotic normality in the proportional regime is established and a conservative inference procedure is proposed. Simulations compare the DaC estimator, OLS, and the Mikusheva–Sølvsten estimator.

Quadratic Voting with Vote Credits

Last revised September 2024.

Abstract

A society faces K binary decisions to be decided through K simultaneous polls using Quadratic Voting (QV) with vote credits: each voter is endowed with a budget of vote credits, which she uses to buy a number \(v_k\) of votes (positive or negative) on each poll k at a cost of \(v_k^2\) vote credits per poll. I characterise voting equilibria as the number of voters tends to infinity and compare total welfare under QV with vote credits and majority voting under different preference distributions. Welfare under QV relies on overall preference intensity being uncorrelated with preferences on individual polls. Independent preferences and a large number of polls create equality in preference intensities and favours QV with credits. I present a pathological example where QV decides in favour of arbitrarily small preferences.

Work in progress

Consistent Estimation and Inference for High-Dimensional Time Series Regressions using Principal Components Regression

Abstract

When the number of regressors of a time series regression is large relative to the sample size, OLS may have a non-negligible bias. However, many high-dimensional data sets exhibit low effective dimensionality in the form of spectral decay, meaning that the eigenvalues of the covariance matrix of the regressors decay to zero. In this paper, I show that a principal component approach to time series regression can mitigate weak exogeneity bias while introducing only negligible omitted variable bias when eigenvalues decay fast enough.