AstralTraveller

02-24-2008, 02:28 PM

Dear all,

Today is the last day of my summer holidays (they only lasted one week this year...darn). So, this will be one of my last posts until I receive admission somewhere :luck2:, and thought I might use a little help from you guys.

Tomorrow I go back to work, and next monday I will teach for the second time a course on Advanced Econometrics for senior Engineering Students (6th year).

This is the last time this class will be taught at this university (it's a small university), at least until they find a new teacher to replace me :p, and I want the kids to get the most out of it. I have --once again-- all degrees of freedom available regarding the program of study. This will be a semester long course on grad level Econometrics for students with very good math qualifications but nil statistical/probability foundations.

After I found out that all the statistics undergrad sequence was not enough for these kids to begin properly the last incarnation of this class, I decided to give them a small "refresh" on the subject, before I even begin with OLS regression properties. Their previous experience with regression analysis has more to do with the pure mechanics of OLS/GMM/MLE, not with the underlying foundations. So they have no clue about biases, consistency or efficiency :hmm:

I had been thinking about three distinct approaches:

* First, a geometry/spectral projections approach, a la Davidson and MacKinnon (quite useful to properly learn GMM afterwards, and everything else as a special case of GMM).

* Second, a more probability/measure theory approach, with Borel and sigma algebras, to get them to master all CLTs, and then onto Likelihoods, the 3 asymptotically equivalent tests (Efficient Scores, Lik-Ratio, and Wald), and then regressions.

* Thirdly, a "recipe like" approach, useful to learn everything by memory. This approach I dislike the most.

Restrictions? I only have one month to teach this. The rest of the semester (up 'til June) I have to teach OLS properties, and then go for Time Series Analysis (Hamilton), Cross Section (Maddala) and Panel Data (Baltagi). I teach 6 clock hours a week. I have a lot of ground to cover.

My audience? Senior students who have to write their professional degree dissertations the following semester. Although several will go for private jobs after getting their degree, I know that several of them (including my assistant!) want to go to graduate school either for Econ or Business academia after getting their degree.

What would be the best approach, in your opinion? I don't want to "orphan" the kids who want to pursue academia, but also I can't forget that this is a course for people wanting to write a dissertation for a professional degree, so it has to be reasonably applied as well.

Anybody? Comments will be really welcome.

Thanks,

Astral

Today is the last day of my summer holidays (they only lasted one week this year...darn). So, this will be one of my last posts until I receive admission somewhere :luck2:, and thought I might use a little help from you guys.

Tomorrow I go back to work, and next monday I will teach for the second time a course on Advanced Econometrics for senior Engineering Students (6th year).

This is the last time this class will be taught at this university (it's a small university), at least until they find a new teacher to replace me :p, and I want the kids to get the most out of it. I have --once again-- all degrees of freedom available regarding the program of study. This will be a semester long course on grad level Econometrics for students with very good math qualifications but nil statistical/probability foundations.

After I found out that all the statistics undergrad sequence was not enough for these kids to begin properly the last incarnation of this class, I decided to give them a small "refresh" on the subject, before I even begin with OLS regression properties. Their previous experience with regression analysis has more to do with the pure mechanics of OLS/GMM/MLE, not with the underlying foundations. So they have no clue about biases, consistency or efficiency :hmm:

I had been thinking about three distinct approaches:

* First, a geometry/spectral projections approach, a la Davidson and MacKinnon (quite useful to properly learn GMM afterwards, and everything else as a special case of GMM).

* Second, a more probability/measure theory approach, with Borel and sigma algebras, to get them to master all CLTs, and then onto Likelihoods, the 3 asymptotically equivalent tests (Efficient Scores, Lik-Ratio, and Wald), and then regressions.

* Thirdly, a "recipe like" approach, useful to learn everything by memory. This approach I dislike the most.

Restrictions? I only have one month to teach this. The rest of the semester (up 'til June) I have to teach OLS properties, and then go for Time Series Analysis (Hamilton), Cross Section (Maddala) and Panel Data (Baltagi). I teach 6 clock hours a week. I have a lot of ground to cover.

My audience? Senior students who have to write their professional degree dissertations the following semester. Although several will go for private jobs after getting their degree, I know that several of them (including my assistant!) want to go to graduate school either for Econ or Business academia after getting their degree.

What would be the best approach, in your opinion? I don't want to "orphan" the kids who want to pursue academia, but also I can't forget that this is a course for people wanting to write a dissertation for a professional degree, so it has to be reasonably applied as well.

Anybody? Comments will be really welcome.

Thanks,

Astral