I used time series data for 15 years , total 6 variables. 5 are independent and 1 is dependent.
For α=0,05 4-dL=1,79 and 4-du=3.44.
You are in the area you cannot decide.
Use Breusch-Godfrey to test for autocorrelation.
The Durbin-Watson test has a lot of drawbacks when you are applying it to the time series. In addition to Breusch-Godfrey test suggested by George Halkos, Portmanteau tests (Box-Pierce or Ljung-Box) are very common in testing serial correlation in the literature.
Shinwari Riazullah with the use of above suggested methods you can only cross checked you results. for removal of auto correlation, you can follow the followings (But conditional to you data and model specification)
1. change the the formation of you regression (non-linear to linear). 2. estimated through dummy variable technique and GLS. 3 you can also estimated it through taking 1st difference variables (but only gives you short run relationships )
Before you solve the problem you have to test if it exists. At the moment and with DW you can not decide.
Use Breush-Godfrey and if auto exists then Durbin 2 steps or CO iterative or CO or Δs
Do you have a linear regression model, with or without a lagged dependent variable.
What is the time frequency of your data?
If annual data for 15 years, and 5 explanatory variables, you will have 9 degrees of freedom after including an intercept.
No test will have any power with 9 degrees of freedom, including any test of serial correlation (DW, B-G, BP, LB, or anything else).
Furthermore, your empirical estimates are unlikely to be robust to any changes in the assumptions underlying the model, including correct specification.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression