Estimation of Average Derivatives of Latent Regressors: With an Application to Inference on Buffer-Stock Saving

Dong Hao, Yuya SasakiarXiv 2022 · DOI · arXiv

Abstract

This paper proposes a density-weighted average derivative estimator based on two noisy measures of a latent regressor. Both measures have classical errors with possibly asymmetric distributions. We show that the proposed estimator achieves the root-n rate of convergence, and derive its asymptotic normal distribution for statistical inference. Simulation studies demonstrate excellent small-sample performance supporting the root-n asymptotic normality. Based on the proposed estimator, we construct a formal test on the sub-unity of the marginal propensity to consume out of permanent income (MPCP) under a nonparametric consumption model and a permanent-transitory model of income dynamics with nonparametric distribution. Applying the test to four recent waves of U.S. Panel Study of Income Dynamics (PSID), we reject the null hypothesis of the unit MPCP in favor of a sub-unit MPCP, supporting the buffer-stock model of saving.

Introduction

In econometric analysis, many fundamental variables are latent—unobservable to the researcher. A primary example is "permanent income," the long-term expected income of a household. While permanent income is a central driver of economic behavior, researchers only observe "measured income," which is the latent variable plus transitory shocks or measurement errors.

Traditional methods for handling these errors often rely on the assumption that the noise is symmetric (e.g., a bell curve). However, empirical evidence suggests that income shocks are frequently asymmetric and skewed. This paper introduces a new density-weighted average derivative estimator designed to handle latent regressors even when measurement errors are non-Gaussian and asymmetric. By applying this tool to U.S. household data, we provide new empirical evidence for the "buffer-stock" model of saving.

The Problem: Measuring the Unobservable

Economists are deeply interested in the Marginal Propensity to Consume out of Permanent income (MPCP). This metric defines how much a household's spending increases for every dollar increase in their long-term expected income.

Standard models often assume an MPCP of exactly 1, meaning households spend all of their permanent income gains. Conversely, the "buffer-stock" model suggests that households are precautionary; they maintain a cushion of savings for emergencies, leading to an MPCP strictly less than 1. Testing this theory requires calculating the derivative of the consumption function with respect to permanent income. Because permanent income is latent and its measurement error is often messy and skewed, standard regression techniques fail to provide consistent estimates.

Methodology

The Density-Weighted Average Derivative Estimator

The authors propose a nonparametric approach to estimate the average derivative of a relationship where the independent variable is hidden. The method requires at least two noisy measures of the same latent variable (for example, income data from two different years).

Key components of the methodology include:

  • Deconvolution: A statistical technique used to "peel away" the distribution of measurement error from the observed data to recover the distribution of the latent variable.
  • Kotlarski’s Identity: This serves as the mathematical foundation, allowing the identification of the latent variable's distribution using two noisy measures, even when the error distribution is unknown and asymmetric.
  • Root-n Convergence: The authors prove that their estimator is highly efficient, achieving a "root-n" rate of convergence. This ensures that the estimator is precise and follows a normal distribution in large samples, which is essential for conducting valid statistical hypothesis tests.
  • Nonparametric Flexibility: Unlike linear models that assume a constant relationship, this estimator allows the relationship between income and consumption to be flexible and nonlinear.

Empirical Application: Testing the Buffer-Stock Model

Data

The estimator was first validated through Monte Carlo simulations to ensure performance in small samples ($N=250$ and $N=500$). Following validation, the authors applied the method to the U.S. Panel Study of Income Dynamics (PSID). They utilized four waves of data (2013, 2015, 2017, and 2019), resulting in a balanced panel of 5,976 household units.

Results

The core of the empirical analysis was a formal hypothesis test: is the MPCP equal to 1, or is it less than 1?

The authors calculated a point estimate of $\hat{\theta}_1 = -0.0607$. In their framework, this negative value represents the deviation from a unit MPCP. The statistical results allowed the authors to formally reject the null hypothesis of a unit MPCP.

This finding provides strong support for the buffer-stock model. It suggests that U.S. households do not spend 100% of their permanent income increases; instead, they withhold a portion, likely driven by a precautionary saving motive to protect against future uncertainty.

Conclusion

This research provides a robust new tool for economists and statisticians dealing with latent variables and asymmetric measurement errors. By moving beyond the restrictive assumption of symmetric noise, the density-weighted average derivative estimator offers a more realistic way to analyze household behavior. The empirical rejection of the unit MPCP reinforces the importance of precautionary savings in modern macroeconomics. Future applications of this estimator could extend to other latent variables, such as measuring the impact of innate "ability" on earnings or the effect of underlying health status on economic productivity.

Summary updated

Abstract

This paper proposes a density-weighted average derivative estimator based on two noisy measures of a latent regressor. Both measures have classical errors with possibly asymmetric distributions. We show that the proposed estimator achieves the root-n rate of convergence, and derive its asymptotic normal distribution for statistical inference. Simulation studies demonstrate excellent small-sample performance supporting the root-n asymptotic normality. Based on the proposed estimator, we construct a formal test on the sub-unity of the marginal propensity to consume out of permanent income (MPCP) under a nonparametric consumption model and a permanent-transitory model of income dynamics with nonparametric distribution. Applying the test to four recent waves of U.S. Panel Study of Income Dynamics (PSID), we reject the null hypothesis of the unit MPCP in favor of a sub-unit MPCP, supporting the buffer-stock model of saving.
BibTeX
@misc{hao2022estimation,
  author = {Dong Hao and Yuya Sasaki},
  title = {{Estimation of Average Derivatives of Latent Regressors: With an Application to Inference on Buffer-Stock Saving}},
  year = {2022},
  eprint = {2209.05914},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2209.05914},
}
Figure