Length of time series to estimate VAR model

I created a VAR(p) model with 12 time series (Y) highly correlated. p is detected iteratively in the interval [1,4] by comparing the aic index.

dim(Y) = (more than 350000,12).

Each time series has more than 350000 rows, there aren't nan elements.

I use the function "estimate" to find the parameters of the VAR(p) model: estimate(Mdl,Y).

This function uses all the elements of Y (except the p-rows for lag) or it uses a lower number of rows? Which is the optimum number of row to estimate the VAR(p) parameters? This is necessary to speed-up the solution.

Answer 1

[Ayush]

Hey Gaia Ceresa,

I understand that you have created a VAR(p) model with 12 highly correlated time series. The lag order (p) was determined iteratively using the AIC index within the range of 1 to 4. The time series data consists of more than 350,000 rows for each series without any missing values. The "estimate" function is used to find the parameters of the VAR(p) model. You want to know whether the function uses all the rows or a lower number of rows, and what is the optimum number of rows for estimating the parameters to speed up the solution.

To speed up the solution and reduce the computational burden, you can consider using a subset of the available rows to estimate the VAR(p) parameters. However, it is important to note that reducing the number of rows used for estimation may result in less accurate parameter estimates. The optimum number of rows to estimate the VAR(p) parameters depends on various factors, such as the time series characteristics, the level of noise, and the specific requirements of your analysis. It is generally recommended to use as many data points as possible to obtain reliable parameter estimates. However, if computational efficiency is a concern, you can experiment with different subsets of the data and evaluate the trade-off between speed and accuracy. You can try using a smaller subset of the data and assess the impact on the parameter estimates and the subsequent analysis but sometimes reducing the number of rows may introduce additional uncertainty and potentially affect the model's performance.

Please refer to the documentation of “Econometrics Toolbox” for more help on time series and VAR estimation: Econometrics Toolbox Documentation - MathWorks India

Hope this helps!

[Gaia Ceresa]

Hi @Ayush

You perfectly understood the problem I posed.

So, there is no limit to the number of rows used by the funcion VAR(p), it uses all the available data.

I will try to do some comparison between the subsets of rows as you suggest.

Thank you very much!