Estimation And Extrapolation Of Policy Effect

Randomized controlled trails(RCT)is one of the most effective and credible tools for policy evaluation and have been widely used in health,labor,education,and behavioral economics.However,RCT is limited by the cost of time and economic and constrained by moral and ethical elements.Moreover,not all policies can be implemented on a large scale through RCT.In practice,data are often not obtained through rigorous experiments.In the actual policy implementation process,the assignment of individuals’ treatment state may not be random,so the existence of "selection bias" has to be considered.Therefore,the problem of endogeneity also needs to be addressed.Accordingly,to overcome the limitations of RCT and the endogeneity problem,researchers have turned their attention to quasi-natural experimental methods to evaluate policies,such as instrumental variable,difference-in-difference,and regression discontinuity design.The method of instrumental variable is an important tool to address endogeneity.However,instrumental variables need to satisfy both the assumption of relevance and exclusivity(exclusion restriction).Hence,there is a difficulty in finding the ideal instrumental variable.In the absence of instrumental variables,the regression discontinuity design(RDD)can both identify and estimate the effect of policies under weaker assumptions,which are usually testable.In addition,RDD is also closer to randomized experiments than other quasi-natural experimental methods.Nevertheless,the focus of analysis in the RD framework has mostly been on the estimation of average treatment effect(ATE),but ATE does not provide sufficient information on the distribution of policy heterogeneity on the outcome variable.While quantile treatment effect(QTE)is an effective tool to characterize heterogeneity to cover the shortage.When the treatment variable is 0-1 binary variable,this paper extends and extrapolates RDD by proposing a new set of moment condition restrictions,which makes it possible to estimate the QTE of the whole group at the at the threshold using a new non-parametric method.Thus,it is possible to capture the potential heterogeneity of the policy effects.Also,on this basis,the method is applied to a specific empirical study to analyze the heterogeneous effect of retirement on consumption.In addition,when the treatment variables are continuous variables,this paper applies the method under continuous variables in the context of relevant environmental policies to obtain error-corrected estimates of QTE and ATE.Chapter 2 of this paper proposes a new method for the identification and estimation of quantile treatment effects based on the assumption of local rank similarity in the framework of regression discontinuity design,which is able to identify and estimate the QTE of the whole population at the threshold,accounting for the presence of covariates in a fully nonparametric manner.In contrast to the monotonicity assumption commonly used in the analysis of regression discontinuity design,this paper imposes the assumption of local rank similarity on the outcome equation,which restricts the evolution of individual ranks across treatment status in a neighborhood around the threshold,and further gives the corresponding tests.In this paper,closed-form expressions for the cumulative distribution functions(CDFs)of the potential outcomes based on the moment restrictions is obtained.It consists of identifiable conditional CDF and conditional probabilities,so that it can be estimated by plug-in method.In the estimation,compared with local linear regression,this paper uses local quadratic regression to estimate the distribution function and conditional probabilities at the threshold,thus effectively correcting the second-order bias due to the large bandwidth.In addition,a multiplier bootstrap-based inference method with robustness against large bandwidths that applies to uniform inference is developed din this paper.Using this new approach,this paper assesses the heterogeneous impacts of India’s 40 billion national rural road construction program on the reallocation of labor out of agriculture.The results find that new road construction has a significant negative effect in the central parts of the agricultural employment share distribution.However,there is no evidence of a significant effect of new road construction in the lower and upper parts of the agricultural employment share distribution.Chapter 3 of this paper estimates the average treatment effect of retirement on consumption using RDD in the context of China’s mandatory retirement policy.It finds that retirement significantly reduces households’ consumption of nondurable goods,work-related consumption,and consumption of food and drink at home.Further,the newly proposed estimation method in Chapter 2 is used to estimate the difference in pre-retirement and post-retirement consumption of nondurable goods for households at different consumption quartiles.The results show that retirement has a significant negative effect on the consumption of nondurable goods only for households at the lower and higher quartiles.Specifically,the negative effect of retirement on work-related consumption and leisure consumption gradually strengthens as the quantile rises,while the negative effect on consumption of food and drink at home gradually decreases.Also,this paper conducts a rank similarity hypothesis test.The results show that households’ consumption preferences for nondurable goods and other non-durable goods do not change at the retirement.Based on Chapter 3,Chapter 4 of this paper analyzes the relationship of digital finance and the impact of retirement on consumption by taking into account the development of the Internet economy.First,as in the previous chapter,the impact of the degree of development of digital finance on the consumption level of residents before and after retirement is examined from both a theoretical framework and an empirical analysis using the discontinuity constructed in China’s mandatory retirement policy(i.e.,whether one is over the legal age)as an instrumental variable for retirement status,and also using geographic spherical distance as an instrumental variable for the digital financial inclusion index.The study finds that the development of digital finance raises the relative consumption level of households after retirement,especially basic consumption.At the same time,the enhancement effect on consumption differs significantly at the level of individual characteristics of household heads.Specifically,household consumption will be more significantly affected by digital finance for those with high levels of education,access to the Internet,and high levels of awareness.Furthermore,mechanism analysis shows that the development of digital finance improves the convenience of payments and alleviates the liquidity constraints of households,thus contributing to the improvement of the consumption after retirement relative to that before retirement.While study in the previous chapters deal with discrete treatment variable,Chapter 5 of this paper considers RDD when t treatment variable is continuous.In this chapter,the discontinuity is constructed using the binding targets for PM2.5concentrations in the 13 th Five-Year Plan.First,the impact of air pollution on net population inflow is investigated by using general RD method.It finds that the implementation of the binding indicators leads to a significant decrease of 3.424μg/m3 in PM2.5 concentration in non-compliant prefecture-level cities compared with compliant areas.With the improvement of air quality,the net population inflow will increase by 199,020 people for every 1 μg/m3 decrease in PM2.5 concentration.Second,using the RDD method under continuous variables to estimate the heterogeneous effect of air quality on population movement,the estimation results of QTE indicate that the more serious the air pollution is,the more difficult it is to control the air pollution,but the more attractive the air improvement is to the net population inflow.The error-corrected ATE is further obtained based on the QTE estimates,and the results show that for every 1 μg/m3 decrease in PM2.5concentration,the net population inflow increases by only 109650 people.Then,Chapter 5 also uses microdata analysis to find that mobile populations that are married,older,less educated,and have high air quality in their domicile are more sensitive to air pollution and their mobility decisions are more susceptible to air quality.Finally,the results of the mechanism analysis suggest that the attraction effect of improved air quality on the population may work through improvements in health and relative increases in employment opportunities.The formulation and implementation of policies are important tools for national governance,and the modernization of national governance system and capacity.They are inseparable from the introduction and implementation of reasonable and effective policies.The policies and reform enacted in process of modernization provide fertile "soil" for the evaluation of policy effects.At the same time,the scientific evaluation of policy effects provides important support for the optimization and improvement of policies,the enhancement of policy quality and the improvement of implementation effects.Micro-econometric methods such as quasi-natural experiments are effective tools for policy evaluation.Through the practical application and extension of RD,this paper enriches the identification and estimation of policy effects in the absence of instrumental variables.The capture of policy heterogeneity will facilitate precise policy implementation.