Abstract
To assess the dynamic distributional impacts of macroeconomic policy, we propose quantile policy effects to quantify disparities between the quantiles of potential outcomes under different policies. We first identify quantile policy effects under the unconfoundedness assumption and propose an inverse probability weighting estimator. We then examine the asymptotic behavior of the proposed estimator in a time series framework and suggest a blockwise bootstrap method for inference. Applying this method, we investigate the effectiveness of U.S. macroprudential actions on bank credit growth from 1948 to 2019. Empirically, we find that the effects of macroprudential policy on credit growth are asymmetric and depend on the quantiles of credit growth. The tightening of macroprudential actions fails to rein in high credit growth, whereas easing policies do not effectively stimulate bank credit growth during low-growth periods. These findings suggest that U.S. macroprudential policies might not sufficiently address the challenges of soaring bank credit or ensure overarching financial stability.
Supplementary Material
The online supplements contain three files:qpe_supplement_appendix.pdf: This file outlines all the proofs of Lemmas and Theorems in the main paper, the step-by-step implementation procedures for constructing pointwise confidence intervals and confidence bands for quantile policy effects, and the sensitivity analysis of the unconfoundedness assumption. qpe_data_20240225.csv: The CSV file contains the data we used in our empirical studies. qpe_JBES_20240225.r: The R code file replicates our empirical results.
Acknowledgments
We would like to thank Xu Cheng, Atsushi Inoue (joint editor), Andreas Lehnert, Frank Schorfheide, the associate editor, two anonymous referees, and seminar participants at Penn for their helpful suggestions on the article. The views expressed in this article are solely those of the authors and do not necessarily reflect the position of the Central Bank of the Republic of China (Taiwan).
Disclosure Statement
The authors report that there are no competing interests to declare.
Notes
1 Several studies apply this approach to assess the impact of policies. See Forbes and Klein (Citation2015), Forbes, Fratzscher, and Straub (Citation2015), Jordà, Schularick, and Taylor (2016), and Acemoglu et al. (Citation2019).
2 Frölich and Melly (Citation2013) and Hsu, Lai, and Lieli (Citation2022)use the instrumental variable method to develop estimators for unconditional quantile treatment effects without the conditional independence assumption.
3 Even in cases with unbounded support, it is possible to estimate the quantile function for
for any
, where the density function
remains bounded away from 0.
4 We must exclude the last NL–T observations from the last sampled block to ensure that the bootstrap sample size remains equal to T.
5 Our definition of the policy variable Dt aligns with methodologies used in earlier works by Forbes, Fratzscher, and Straub (Citation2015) and Richter, Schularick, and Shim (Citation2019).
6 Variables such as bank credit, the industrial production index, the consumer price index, reserve money, the 3-month Treasury bills interest rate, and the 10-year government bond yields are from FRED. S&P 500 stock prices, CAPE, and the real home price index are from Shiller (Citation2016). All these variables, except interest variables and the yield curve spread, are taken from their official, seasonally adjusted values and converted into annual growth rates. To account for inflation, we transform bank credit, equity price index, and monetary aggregates into real values by dividing them by CPI.