I am going to estimate ARDL model based on monthly data using Eviews. However, I don't know how to determine the max lags for dependent variables and repressors. And is there any difference between max lags and optimal lags？ Could I find optimal lags based on the VAR model firstly, and then use optimal lags as max lags to estimate ARDL model?
Dear Elena Wang , if you agree to chose optimal lags according to information criterion (AIC, SC, HQ) then you can find it in the "options" submenu (top-right corner of the ARDL model specification window; it's the same submenu where you specify coefficient covariance matrix).
You can search on the RG and/or Eviews Forum (Q&A section) for more information; there are a couple of similar questions/discussions regarding the ARDL lag selection.
This is an autoregressive distributed lag model for which the dependent variable follows an autoregressive model with lagged parts of the independent variable. Therefore the partial autocorrelation function of the dependent variable shall cut-off at the autoregressive order.
You can use AIC, BIC, SIC, HQ criteria. The optimal lag will be which has the minimum value as per AIC, BIC, SIC and HQ.
There is no theory behind the max lag. But according to the rule of thumb, the max lag number can be set to log of the number of observations. Once you set max lag number based on the number of observations then info criteria like AIC or BIC help to identify optimal lag order.
I think the standard procedure is to either use the log of the total number of observations or to rely on the results of the AIC and HQ. What i know for sure is that there is no a universal route every researcher has to follow
If you use STATA, you can run the following code to determine the optimal lag for your model:
varsoc Y X1 X2 X3, maxlag(2)
- Badges
- Science topic
- Similar topics
- Statistics
- Time Series