Turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

- Home
- /
- Analytics
- /
- Forecasting
- /
- Big difference between actual and predicted values in GMM Estimation

Options

- RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Mute
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content

Posted 12-26-2017 01:16 PM
(1063 views)

Hi Expert Fellas;

I have a panel of some firms (GVKEY) in a period of time (FYEAR) with some variables related to them. here is a link to a sample of my data (real database has more than 60000 rows):

https://www.dropbox.com/s/hpphgpluknk5414/Data%20Sample.csv?dl=0

I tried to run a GMM Estimation using the following code:

```
proc panel data=Lev_lag outest=outst noprint;
output out=outx PREDICTED=predict;
inst depvar(LEVEL);
MM: model Lev = Lev_1 MTBV Size Tangibility Profitability Cashes Tax TB3Ms
/ gmm nolevels twostep maxband=3 noint fixtwo ;
id gvkey fyear;
run;
```

"outx" is an output that shows both actual and predicted values of my model. now, the problems are:

1. as it is depicted in the following picture, there is a huge difference between actual and predicted values of my dependent variable. this is, while all the variables have significant coefficients in the model.

2. I wanted to have a GMM model with Fixed effect for both Year and Firm. So, I added "fixtwo" to my model options. However, the output file shows me another result only for Fixtwo model which is seperated from the GMM results!

as you can see, the results for fixed effect model are logical, and actual model is pretty similar to the predicted model (good fit). However, It's not GMM anymore!

Now, would you please guide me through this problem? I'm not sure what step of my work is wrong which gives me this model.

thanks everybody.

2 REPLIES 2

- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content

Hello,

I run your model using the sample data you provided. Here are few suggestions that you might want to try.

1. FIXTWO option has no impact on estimating the GMM model. Basically it just estimates FIXTWO model in addition to GMM one.

2. DEPVAR(LEVEL) and NOLEVELS options are conflicting. DEPVAR(LEVAL) says to use levels as instruments while NOLEVELS is saying to ignore it. In the end NOLEVELS is ignored and it estimates the model with levels as instruments only.

3. Is this the correct model you want to estimate with levels as instruments only? You might want to use differenced equations too. See the DEPVAR option in SAS/ETS documentation.

4. The first step is estimable (see this by removing TWOSTEP option) but using your sample data there is a problem with residuals. Is this model parametrized correctly?

5. Using the sample data the model fit is pretty bad (t-values close to zero). Likely that this parametrization is causing problems in 4.

Jan

- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content

Thanks for your suggestions dear Jan.

I am working on the GMM model and my data sample, for over a month. I'm almost sure about parameterization of the model since it's a well-known model which I just try to replicate. there is only one thing in GMM model which was interesting for me. this model is highly sensitive to outliers. after I winsorized the data, the model fit surprisingly became much better. This is, while winsorizing didn't have a significant effect on FIXTWO results.

referring to #4, I got this error too and it seems that there is a multicollinearity in my model. However, I couldn't find it!! I could solve this problem by increasing the MAXBAND in GMM or omitting some specific variables from the model.

Overall, I found GMM a very sensitive model which its output can change easily and drastically by model's options like type of instruments, number of lags or maxband.

thanks again for your comments.

Hamed

Registration is open! SAS is returning to Vegas for an AI and analytics experience like no other! Whether you're an executive, manager, end user or SAS partner, SAS Innovate is designed for everyone on your team. Register for just $495 by 12/31/2023.

**If you are interested in speaking, there is still time to submit a session idea. More details are posted on the website. **

Multiple Linear Regression in SAS

Learn how to run multiple linear regression models with and without interactions, presented by SAS user Alex Chaplin.

Find more tutorials on the SAS Users YouTube channel.