Yimin Yang, Fei Jia, Haoran Li

Read Full Paper

This paper investigates panel data models with mixed sampling frequencies (MIDAS), which arise when data is collected at different frequencies. Traditional panel models assume the same frequency for all variables, but in practice, researchers often face data sampled at different frequencies. A common approach is to aggregate variables to the same frequency using an equal weighting scheme, but such an aggregation can lead to biased estimates, especially when using fixed-effects estimators. To address this, the paper proposes a data-driven method to determine aggregation weights, ensuring consistent estimators.

The paper first identifies that the fixed-effects estimator with equal weights aggregation is generally inconsistent unless certain special conditions hold. It then introduces a new weighted aggregation method that provides consistent estimators for mixed-frequency panel data. Additionally, the paper examines the Mundlak and Chamberlain approaches, showing that while the Mundlak method no longer applies in the MIDAS context, Chamberlain's approach remains valid.

In the empirical section, the paper applies the new method to estimate the effects of temperature fluctuations on economic growth. By more accurately processing high-frequency temperature data, the method avoids the biases caused by simple averaging. The findings suggest that temperature fluctuations affect economic growth through a level effect rather than a growth effect for poor countries. The paper also presents simulation results to demonstrate the finite-sample performance of the proposed estimators and discusses their practical implications.

Overall, this paper provides a new solution for handling mixed-frequency panel data, extending traditional panel data estimation techniques, and offering theoretical and methodological support for empirical analyses based on mixed-frequency data.