Getting confused in selecting Optimal Lag

[Radoyan Hossain]

i am struggling in selecting optimal lag for Ardl model. my model is based on 33 observations . for this which command should i run for selecting optimal lag am getting confused Alot , i tried to take help from chatgpt but the provided ans are ambiguos.
therefore i would be glad if the precise command for lag selection is given to me keeping in mind of 33 observations of my model with 7 variables.best regards.

[Radoyan Hossain]

is there anyone ? help me out

[Manh Hoang Ba]

The user-written ardl command has a maxlag() option that allows you to specify the maximum lag order for the variables in the model (starting with the dependent variable). Combined with the AIC or BIC, it can be used to select the optimal lag structure by estimating all possible ARDL models and choosing the model with the minimum AIC (or BIC).

 
 ardl y x1-x7, maxlag(2) bic

By default, ardl uses maxlag(4), and I suspect this is causing the problem in your case.

Your dataset contains only 33 time-series observations, while the model includes as many as seven variables. If four lags are allowed for each variable, with a constant and no trend, up to 35 regression coefficients would need to be estimated (including both the level terms and lagged terms). This is not feasible with only 33 observations, so the lag-order search procedure stops.

Although you can reduce the lag order in maxlag(), with only 33 observations, seven variables are still quite a lot even if you use only one lag. You would be left with fewer than 20 degrees of freedom after estimating the model, which makes the postestimation diagnostics potentially unreliable.

[Radoyan Hossain]

The user-written ardl command has a maxlag() option that allows you to specify the maximum lag order for the variables in the model (starting with the dependent variable). Combined with the AIC or BIC, it can be used to select the optimal lag structure by estimating all possible ARDL models and choosing the model with the minimum AIC (or BIC).

ardl y x1-x7, maxlag(2) bic

By default, ardl uses maxlag(4), and I suspect this is causing the problem in your case.

Your dataset contains only 33 time-series observations, while the model includes as many as seven variables. If four lags are allowed for each variable, with a constant and no trend, up to 35 regression coefficients would need to be estimated (including both the level terms and lagged terms). This is not feasible with only 33 observations, so the lag-order search procedure stops.

Although you can reduce the lag order in maxlag(), with only 33 observations, seven variables are still quite a lot even if you use only one lag. You would be left with fewer than 20 degrees of freedom after estimating the model, which makes the postestimation diagnostics potentially unreliable.

thank you so much .
i need to ask one more thing , would it be fine if i take maxlags 3 ?
i am in absolute constraint i cant drop any variable and also cant increase the size of the observations due to not availability.
thanks for your valuable time ,peace

[Manh Hoang Ba]

Since post-estimation inference requires a certain level of reliability, I think maxlag(1) would be the best choice.