区域协调发展政策效果的再评估--来自因果森林算法的异质性处理效应分析

Using Machine Learning to Revalue the Regional Coordinated Development Policy—Analysis of Heterogeneous Treatment Effects from Causal Forest

在“有为政府”与“有效市场”双重机制下,尽管我国东西部地区发展在差距不断缩小的同时,经济仍能保持较高速度增长,但由于抑制空间自由分工可能会引发资源错配,其结果可能是牺牲经济效率换取区域平衡发展;因此,各界对旨在加快欠发达地区发展的区域协调政策的效果争议颇大。有鉴于此,本文首次利用因果森林为主的机器学习方法评估了2003年以来区域政策对各城市生产效率和相对公共福利的影响,并通过岭回归、lasso估计和正则化分析了因果森林平均处理效应(ATE)和条件平均处理效应(CATE)的稳健性,结果发现:(1)2003—2017年,区域政策对缩小地区之间相对公共福利差距的作用不明显,但是显著提升了欠发达地区人均GDP发展速度,每年贡献度达2.2%;(2)区域政策对提升欠发达地区生产效率的作用在不断下降,目前很多城市“阶段性”下降至“零”效果的附近;(3)倾斜性政策对促进南方城市生产效率提升的效果大于北方;(4)区域政策干预并不能弱化港口因素对欠发达地区经济发展的制约,南方地区尤其明显。

Although under the dual mechanism of“promising government”and“effective market”,the development gap between the East China and the West China has been narrowing and the economy is still growing at a high speed.For a long time in the past,the two phenomena coexist at the same time.However,restraining the spatial division of labor will lead to resource mismatch,and this way of narrowing the regional development gap will usually sacrifice economic efficiency,so the effect of regional coordination policy aimed at promoting the rapid development of underdeveloped areas is controversial.In view of this,this paper uses causal forest based machine learning for the first time to evaluate the impact of regional policies on urban production efficiency and relative public welfare since 2003,the results show that:(1)from 2003 to 2017,the role of regional policy in narrowing the relative public welfare gap between regions is not obvious,but it significantly improves the per capita GDP growth of less developed regions,with an annual contribution of 2.2%;(2)the role of regional policy in improving the production efficiency of less developed regions is declining,and at present,many cities have gradually dropped to near the“zero”effect;(3)the effect of preferential policies on promoting the production efficiency in southern cities is greater than that of northern regions;(4)regional policy intervention can not weaken the restriction of port factors on the economic development of underdeveloped regions,especially in southern regions.Ridge regression,Lasso and regularization test were used to test the validity and robustness of the causal forest results.