This paper derives the limiting distributions of least squares averaging estimators for linear regression models in a local asymptotic framework. We show that the averaging estimators with fixed weights are asymptotically normal and then develop a plug-in averaging estimator that minimizes the sample analog of the asymptotic mean squared error. We investigate the focused information criterion (Claeskens and Hjort, 2003), the plug-in averaging estimator, the Mallows model averaging estimator (Hansen, 2007), and the jackknife model averaging estimator (Hansen and Racine, 2012). We find that the asymptotic distributions of averaging estimators with data-dependent weights are nonstandard and cannot be approximated by simulation. To address this issue, we propose a simple procedure to construct valid confidence intervals with improved coverage probability. Monte Carlo simulations show that the plug-in averaging estimator generally has smaller expected squared error than other existing model averaging methods, and the coverage probability of proposed confidence intervals achieves the nominal level. As an empirical illustration, the proposed methodology is applied to cross-country growth regressions.

This article considers the problem of inference for nested least squares averaging estimators. We study the asymptotic behavior of the Mallows model averaging estimator (MMA; Hansen, 2007) and the jackknife model averaging estimator (JMA; Hansen and Racine, 2012) under the standard asymptotics with fixed parameters setup. We find that both MMA and JMA estimators asymptotically assign zero weight to the under-fitted models, and MMA and JMA weights of just-fitted and over-fitted models are asymptotically random. Building on the asymptotic behavior of model weights, we derive the asymptotic distributions of MMA and JMA estimators and propose a simulation-based confidence interval for the least squares averaging estimator. Monte Carlo simulations show that the coverage probabilities of proposed confidence intervals achieve the nominal level.

We propose a test for a generalized regression monotonicity (GRM) hypothesis. The GRM hypothesis is the sharp testable implication of the monotonicity of certain latent structures, as we show in this article. Examples include the monotonicity of the conditional mean function when only interval data are available for the dependent variable and the monotone instrumental variable assumption of Manski and Pepper (2000). These instances of latent monotonicity can be tested using our test. Moreover, the GRM hypothesis includes regression monotonicity and stochastic monotonicity as special cases. Thus, our test also serves as an alternative to existing tests for those hypotheses. We show that our test controls the size uniformly over a broad set of data generating processes asymptotically, is consistent against fixed alternatives, and has nontrivial power against some ${n^{ - 1/2}}$ local alternatives.

This paper considers the problem of inference for nested least squares averaging estimators. We study the asymptotic behavior of the Mallows model averaging estimator (MMA; Hansen, 2007) and the jackknife model averaging estimator (JMA; Hansen and Racine, 2012) under the standard asymptotics with fixed parameters setup. We find that both MMA and JMA estimators asymptotically assign zero weight to the under-fitted models, and MMA and JMA weights of just-fitted and over-fitted models are asymptotically random. Building on the asymptotic behavior of model weights, we derive the asymptotic distributions of MMA and JMA estimators and propose a simulation-based confidence interval for the least squares averaging estimator. Monte Carlo simulations show that the coverage probabilities of proposed confidence intervals achieve the nominal level.

This paper proposes a novel model averaging estimator for the linear regression model with heteroskedastic errors. Unlike model selection which picks the single model among the candidate models, model averaging, on the other hand, incorporates all the information by averaging over all potential models. The two main questions of concern are: (1) How do we assign the weights for candidate models? (2) What is the asymptotic distribution of the averaging estimator and how do we make inference? This paper seeks to tackle these two problems from a frequentist view. First, we derive the asymptotic distribution of the averaging estimator with fixed weights in a local asymptotic framework. The optimal weights are obtained by minimizing the asymptotic mean-squared error (AMSE) of the averaging estimator. Second, we propose a plug-in averaging estimator which selects the weights by minimizing the sample analog of the AMSE. The asymptotic distribution of the proposed estimator is derived. Third, we show that the confidence intervals based on normal approximations suffer from size distortions. We suggest a plug-in method to construct the confidence interval which has good finite-sample coverage probability. The simulation results show that the plug-in averaging estimator performs favorably compared with other existing model selection and model averaging methods. As an empirical illustration, the proposed methodology is applied to estimate the effect of the student-teacher ratio on student achievement. We find that the insignificance of the student-teacher ratio variable from previous literature could be potentially explained by the fact of ignoring the model uncertainty.