擂台赛：计量分析路线图

是 否 +2 论坛币

k人 参与回答

经管之家送您一份

应届毕业生专属福利!

求职就业群

赵安豆老师微信：zhaoandou666

经管之家联合CDA

送您一个全额奖学金名额~ !

立即领取

感谢您参与论坛问题回答

经管之家送您两个论坛币！

+2 论坛币

计量分析越来越复杂。经常会碰到一个问题不知道怎么解决。

计量经济学就是在解决一个个问题的基础上逐渐形成发展形成现在的理论体系的。

请牛人们一起来描画出经济计量分析路线图，从数据分析、模型设计、到计量检验等等各个阶段，都会遇到的各种问题及其解决办法，都列出来。这相当于是搞实证研究的一个入门导引。

请斑竹给有功之人给以金钱奖励（可惜我自己不多，要不我就给了的）。

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

分享0 收藏0 回帖

关键词：计量分析 路线图 擂台赛 经济计量分析 计量经济学 计量 路线图 擂台赛

好主意,福泽后人的壮举, 可惜我的水平差些.

很不错

希望大家支持楼主

其实计量的核心就是寻找具有一致性、无偏性和有效性的统计量~~

你好，抛砖引玉。难免有错，请大家修正。我呢，就当自己复习一下

A,ECONOMIC DATA
there are three kinds of economic data: quantitative (cardinal), ordinal(ordered), qualitative.

the ordinary can not use in the econometrics model. coz they do not express any quatily measures. for example: how much you love your job? 5: very much 4: like 3: do not like or hate 2: do not like 1: hate.

the qualitative sometimes called indicator wariable or binary or Boolean variable. more often, dummy variable.

INSTRUMENTAL VARIABLE ESTIMATORS
three kinds things to violate the zero-conditional-mean assumption: endogeneity, ommitted-variable bias, and errors in cariables. the solution is the same. IV

B,BASIC Model ASSUMPTIONS

一，classic model

For finite model, assumptions are

1,linearity ,to make sure that the model is correct, if the model is not linearly , the estimation methods can not used. Sometimes the model are not linear, we can change it to the linearly one.

2,strict exgoneity ,this is can be violated and cause the series correlation problem.

3,no multicorrelation. This is only for calculation, nothing special

4,homo, this is also can be violated and become the heter problem

If we need test, we need normality assumption. Of course, it is just a popular assumption, somebody pointed out normality is not a perfect assumption

For large sample.

The assumptions are

1, linearity （同FINITE SAMPLE）

2,predetermined, this is weaker than the a2 in finite, but it is more practical.

3ergodic and stationary, this is like the iid which is useful to get the CLT and combind the data with normality

4,g is mds whith 2^nd finite moment, this is to make sure that the s is consistent.

5, rank condition(identification for iv), this is only for calculation, too.

We do not need the normality assumption thanks to the CLT.

And the home assumption makes the estimation much easier.

the clues to study econometrics

1,assumption:understand why we need this assuption

2,estimator and the methods. here, the exams often have this kind of questions: why we do not use ols estimation. or what is we use ols estimation here. we can consider 2 things. biasdness and consistency.

3,test, normally to test estimator, we have 3 methods: simple. multiple, nonlinear

4,how to test the model assumptions? eg, heter, using white, serrial corrlation, using box, dubin, overidentification, using j stataistic,unit root, using df, adf.

5, special method of estimation, eg, gls, wls, 2sls, five, 3sls, sur,understand the conditions of these metods

接着来 ,

c,study notes-criteria for estimators

computational cost

least squares

highest R-square

unbiasedness - smaple distribution

efficiency -smaple distribution

mean square error -smaple distribution

asymptotic property - small sample

maximum likelihood

monte carlo studies -small sample

baby姐写得很快啊，不知道能不能有奖励呢？都是记忆里面的东西，没有仔细翻书，错的话请指出。baby姐在这里谢谢了。

C, Violate the assumptions

Violate assumption of serial correlation. Use I V ,how to test: vif, box-pierce , ljung box

Homo use gls how to test: white

D,Other models:

1,Gmm-------the eror term and the regressors are not orthogonality, including single gmm and multiple gmm.

2,Panel -------have multiplae observations for a sectional variable. Including fixed effect model and random effect model. And the random effect estimation is a special case of multiple gmm.

3,Time series(ARIMA) ---the time series data go with the serial correlation the ARIMA is a common method.

4,Extremum ---when the observations are not all observed, e.g. truncates model , censored model ,the method is ML.

算不上是什么路线图，这个图是很难做出来的，但是如果了解整个计量的框架，对于计量的分析就会比较有思路，而且计量往往就是这样，同样的问题，不同的人，会有不同的解决方法。计量经济学家不过时将很多有特色的归类了，但并不是所有的数据都有明显的特色。

暂时写到这里，没有看书，有的地方写得也不够详细。如果baby有兴趣还会来补充。baby很多东西发到中国经济学教育科研网的计量经济学板块了。

原创。如需转载需得到baby许可，谢谢。

这里写一下time series基础部分.

如何处理time series model

1, 对所有的时间序列作图，找出其中的规律，大致看一下，是否平稳，是否有时间趋时或者是否是一阶平稳。不同的变量之间的时间序列图是否存在联系，比如是否有协整地可能性。

2，判断平稳或协整。单位根检验是否平稳，如果不平稳，用df adf，ressidial based ADF test检验是否协整。平稳或协整则进入下一步。

3，choose the model ，用sequence p , AIC BIC都作一遍选择最简单的模型。

剩下的就不用多讲了。