There are many papers use bootstrap standard error when they do econometrics.

My question are:

  1. why do we need bootstrap standard error?

  2. when to use it?


Normally, you use bootstrapping when you cannot derive a formula for the variance-covariance matrix of your estimator. Since you do not have this formula, you cannot obtain the standard errors of parameters, and so your econometric analysis is very limited. In these cases, using bootstrapping is the only option available. This is common in complex, non-linear models.

Similarly, you do not need them for those cases where you can derive robust, close-form formulas for the standard errors (as in OLS, or linear panel data models, or most of models found in most of textbooks). In this cases, you already have a formula which is asymptotically efficient, and bootstrapping would be redundant, and probably counterproductive if done incorrectly.

Notice that bootstrapping is not only useful for obtaining the standard errors of parameters. You can also bootstrap full tests. For example, you can bootstrap the Hausman test that compares RE versus FE. To go this way would imply a longer answer.

For a quick and hands-on introduction, check Cameron and Trivedi (2010), Chapter 13.

  • 1
    $\begingroup$ I would add that maybe bootstrap se's are also useful when you to check whether some outliers are driving your statistical significance. $\endgroup$ Aug 23 '16 at 20:01

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.