Hello
I am trying to fit a regression model for attached data.The problem I'm having is that I get back coefficient value of 1 for all independent variables. Maybe my process is wrong. Here's what I'm doing and I wondered if anyone can help me please.
online_visits and offline_visits are my independent variables.
total_visits is my dependent variable.
When I run the regression
ls total_visits c online_visits offline_visits
I get a coefficient value of 1 for both online_visits and offline_visits. Both are significant. I'm expecting offline to be lower than online as the values are much lower.
After reading various things online, I looked at Unit Root Test. I found dependent variables are not stationary without the intercept and trend. So I differenced the series to make it stationary.
ls d(total_visits) c d(online_visits) d(offline_visits)
And getting 1 for coefficients.
Correlogram shows significant autocorrelation at lag 1 and partial correlation at lags 1, 2 and 7. So I tried
ls d(total_visits) c d(online_visits) d(offline_visits) ma(1) ar(1) ar(2) ar(7)
but that gives my an error:
Log of non positive number
Can anyone help me on how do I fit the model properly using EViews or why I get 1s for coefficients please?
Many thanks
Emily
Time series regression with all coefficient result of 1
Moderators: EViews Gareth, EViews Moderator
-
- Posts: 2
- Joined: Thu Jan 31, 2019 12:25 pm
Time series regression with all coefficient result of 1
- Attachments
-
- visitors.csv
- (17.67 KiB) Downloaded 238 times
Last edited by emilyrosecharm on Thu Jan 31, 2019 2:02 pm, edited 1 time in total.
-
- Fe ddaethom, fe welon, fe amcangyfrifon
- Posts: 13319
- Joined: Tue Sep 16, 2008 5:38 pm
Re: Time series regression with all coefficient result of 1
You're estimating an identity.
A regression model would be:
Total_Visits = Constant + Beta0*Online_Visits + Beta1*Ofline_visits + error_term
Where the error term has random noise in it.
Your data actually has this:
Total_Visits = Online_Visits + Ofline_visits
Which implies a few things:
1) Constant = 0
2) Beta0 = 1
3) Beta1 = 1
4) error_term = 0
Since there is no error_term, this isn't something you should be estimating. Least squares is not appropriate.
A regression model would be:
Total_Visits = Constant + Beta0*Online_Visits + Beta1*Ofline_visits + error_term
Where the error term has random noise in it.
Your data actually has this:
Total_Visits = Online_Visits + Ofline_visits
Which implies a few things:
1) Constant = 0
2) Beta0 = 1
3) Beta1 = 1
4) error_term = 0
Since there is no error_term, this isn't something you should be estimating. Least squares is not appropriate.
Follow us on Twitter @IHSEViews
-
- Posts: 2
- Joined: Thu Jan 31, 2019 12:25 pm
Re: Time series regression with all coefficient result of 1
Ahaaa.. thank you soo much Gareth . That makes sense.
Return to “Econometric Discussions”
Who is online
Users browsing this forum: No registered users and 29 guests