经史通义
首发于经史通义
是智者的忏悔,还是强者的咆哮 --《无用的计量经济学》背后的那点事

是智者的忏悔,还是强者的咆哮 --《无用的计量经济学》背后的那点事

乔治敦大学的教授John Rust近来写了一篇文章《Mostly Useless Econometrics? Assessing the Causal Effect of Econometric Theory》引起了很大的反响。毕竟,一位以计量经济学起家的教授自己说计量经济学目前走向了弯路,需要纠正所带来的影响是不可小觑的。于是,我和@焚琴煮鹤的Yoyo 找来了这篇文章读了一遍,读完之后,不禁感慨万千,一方面确实是觉得目前计量经济学界非常的理论化,Rust也指出了当前的一些弊端和倾向,而另一方面,也感叹老爷子妙笔生花^_^。为什么说妙笔生花呢,接着看我们这篇合写的专栏文章吧。

如果只看这一篇文章,可能还会怀疑Rust是个反对结构类实证模型(structural model)的。其实不然,老爷子是structrual model 的开山鼻祖之一。83年毕业于MIT,执教于威斯康星。一时间风头正劲,econometrica发的手软。

直到了2002年,他命中的“克星” -- Thierry Magnac 发表了里程碑的式的论文:Identifying Dynamic Discrete Decision Processes,这篇文章如此的重要,以至于说打开了后来Magnac当选为Econometric Society 的Fellow的道路都不为过。

----

补充一下:为什么这里说Magnac的文章如此重要呢?因为从那以后,所有人做dynamic discrete choice model 的时候,都知道discount factor 是不可以被identify的。而且error的distribution也是不可能被identify的。可以这么说,目前还在做micro structual model estimation 的人,都很难绕过Magnac的这篇文章另起炉灶。毕竟,Econometric Society的Fellow是整个经济学界的核心,贡献不够是很难选进去的。

而之前的文章,对这一点的理解并不清楚。这个是历史的局限性,本来也轮不到具体某一个人背锅,但是问题在于,Rust老爷子从87年到02年的这几篇econometrica都estimate了discount factor...

----

Magnac这篇文章开头两段是这样的:

OVER THE LAST DECADE, ... Little is known however about the identification of these models except that they are generically not identified(Rust (1994)).

In this paper, we analyze the nonparametric identification of dynamic discrete choice models. ... Not only do we show that these models are not identified (Rust(1994)), we are also able to determine their exact degree of underidentification...

直接指出了从Rust(1987)的一系列文章中存在严重的 non-parametric identification 的问题。而偏偏Rust老爷子的这篇94文章是在 Handbook of Econometrics,是个参考书,然后他87年的文章也是一个影响力很大的文章,都经常被引用,而一旦有人引用,往往就成对的也引用Magnac的文章,这十几年来一直如此。

Rust 的内心估计是这样的:

批判一次也就罢了,这每次看到别人引用自己的文章,同时都会顺便引一下批判自己的文章,这是什么道理啊!于是在这篇Mostly Useless Econometrics?...》他终于狠狠的把憋在自己内心很久的话说出来了:

It should not be necessary to establish a causal mechanism or a non-parametrically identified structural model that provides an unambiguous explanation of empirical phenomena as a litmus test for publication.

我们继续看老爷子又在吐槽什么:

Though I will avoid naming names, many econometric theorists are more like pure statisticians or mathematicians who do little empirical work themselves.

They call themselves “econometricians” mainly because salaries in economics are much higher than in statistics departments which have long been in decline and in some cases eliminated (such as at Princeton), or nearly eliminated (such as at Yale).

这几段,有点尖刻,不指明道姓的再说一位耶鲁的计量经济学家。这位经济学家很多人应该都知道,她是唯一的华裔econometric society做计量的fellow。为什么说是她呢?线索不仅仅在Yale上,还在这一段:

Does anyone really use k-class estimators, or 3 stage least squares, or maximum score?

这位fellow正好就是k-class estimators方面的权威。


Rust老爷子还抱怨econometrica被一群理论大V掌控了:

The leading journals such as Econometrica have great influence on the type of work done in economics due to the hierarchical way the profession is self-organized.

Unfortunately too few leading theorists have a real interest in the real world application of their theories: this is a task left for the worker bees.

看老爷子的publication,似乎确实2002年之后,再也没有econometrica了... 这十来年的委屈,老爷子容易么?

老爷子呼吁:

Economics journals should lower the burden of proof for empirical work and raise the burden of proof for econometric theory. Specifically, there should be more room for descriptive empirical work in our journals.

这是在说,你们这些人讷,不要总想着搞个大新闻,说我identification有什么问题,实证文章,理论部分差不多就行了,我们收集数据也不容易啊,多发点我们的structual model吧,别纠结我们的identification了!至于那些做纯理论的,我们就是要提高标准,难死他们,难死他们^_^

纵观下来,老爷子这篇文章看上去是作为计量大牛的忏悔;而在忏悔的背后,还包藏着落寞强者的咆哮,其实他的观点是有利于做传统计量经济学的学者呢,他也是在为自己这一派做structural的张目,大家都知道structural model难写难发,identification更是一个技术活,如果真的如他所说,提高theory的发表标准,降低empirical的发表标准,老爷子说不定还能够老骥伏枥,时隔十来年再来几篇trica也未可知呢~

编辑于 2016-12-13

文章被以下专栏收录