很郁闷啊，就是不能replicate很famous的results - 未名空间MITBBS历史存档

国际科技财经博客移民网络热点娱乐民生时事公众号

Redian新闻

>未名空间

>Economics - 经济

很郁闷啊，就是不能replicate很famous的results

很郁闷啊，就是不能replicate很famous的results# Economics - 经济

B*e2003-06-16 07:06

1 楼

data都是compustat和CRSP里的公共data,我也是严格照着paper写的去做的．program
一步步的不知检查，rerun过多少遍了，就是出不来应该是很neat的结果．
很郁闷啊．这里面会有什么trick嘛？我想重要的步骤paper应该全说了，可能有
一些小的data filter,去掉outliers什么的没说？可是他们不应该对results有
大的影响啊:(
有没有人能告诉我，我应该从哪下手check? 还是干脆把sas codes推翻重写:(

S*Y2003-06-16 07:06

2 楼

those little tricks can dramatically change the result.

【在 B****e 的大作中提到】

: data都是compustat和CRSP里的公共data,我也是严格照着paper写的去做的．program
: 一步步的不知检查，rerun过多少遍了，就是出不来应该是很neat的结果．
: 很郁闷啊．这里面会有什么trick嘛？我想重要的步骤paper应该全说了，可能有
: 一些小的data filter,去掉outliers什么的没说？可是他们不应该对results有
: 大的影响啊:(
: 有没有人能告诉我，我应该从哪下手check? 还是干脆把sas codes推翻重写:(

B*e2003-06-16 07:06

3 楼

thanks for reply:)
but my boss told me: they are very robust results. many people can replicate:(
郁闷ing~

【在 S****Y 的大作中提到】

: those little tricks can dramatically change the result.

S*Y2003-06-16 07:06

4 楼

i was told that once too, then found out it's caused by "little" tricks.

【在 B****e 的大作中提到】

: thanks for reply:)
: but my boss told me: they are very robust results. many people can replicate:(
: 郁闷ing~

w*i2003-06-16 07:06

5 楼

To calculate t-stat, you should estimate variance first. If you use NeweyWest
method to estimate variance, you should choose time lag. Different time lags
give different results. Nobody knows how many lags you should choose.

considering
this
replicate:(