I have been working on my paper where the only dependent variable that I can use, and is fitted based on my research objectives, is a data from a survey that is taken every 3 years. My other variables, the independent variables, have a complete time series data. My goal is to find the relationship between my dependent and independent variables. Is this possible to test? Is there an econometric method that I can use? The data for the dependent variable also have an available data per region that I might use. Responses are very much appreciated, thank you!
Since it is a time series data, pre-tests will dictate the best model for your study. As for the dependent variable which is measured every 3 years it is just a multi-year. To achieve your goals of fitting a suitable time series model to your series, you must ensure that all the variables/series are measured at the same low frequency and that will depend on your research questions or objectives.
Best wishes!
My pleasure @ Elle Rae,
There are lots of models (methods) you can use like I said earlier but it all depends on the pre-tests of your variables. For accurate specification of the best model, below are the step-wise procedures that will guarantee your model selection:
Step 1: Visualize the time series plots of all the variables to determine if they are level stationary process {I(0)s} or difference stationary process of order one {I(1)}.
Step 2: Conduct stationarity tests based on the results you get from Step 1 above. Since you are dealing with multivariate time series variables your variables must either be all i(1)s or mixtures of I(0)s and I(1)s for you to be able to fit suitable multivariate time series models to them.
Step 3: If the variables satisfy Step 2, then conduct cointegration testing on the variables. The essence of this test is to determine if there is long-run relationship among the variables. If there is a long-run relationship, then the possible models here are either ARDL (p, q) variants or the Vector Error Correction Model (VECM). But if there is no long-run relationship, the possible models are restricted VAR (p), augmented VAR (p+q), or structural VAR model depending on your research questions/objectives.
Note:
If the stated stationarity assumptions of the multivariate time series mentioned in Step 2 are relaxed, then you can fit the simultaneous equation models to the variables since it is not perturbed by the order of integration (stationarity assumptions) of multivariate time series.
Best wishes!
@ Elle Rae, if the missing values are not many, you can replace them with previous values. I have a reference for this but will send it to you later this week.
- Badges
- Science topic