How many observations do I need?

[samewing7]

Hello,

I am completing a paper on the effects of the oil price on the brazilian economy. I am using a level VAR model and want to include the following variables: GDP, INflation, interest rates, terms of trade, family consumption and the oil price. However, due to the lack of data in Brazil I only have 75 observations (quarterly data from 1996-2015) and I feel this may not be enough as I want to include 4 lags.

To add to the problem, I have completed a chow break point test in 2008 and found the relationship between GDP and the oil price breaks down following this. I therefore wanted to split my regression up to pre 2008 and post which leaves me with only 45 observations and 30 observations for those respective periods.

I am unsure what to do now. Would people suggest to do two separate VARs with fewer variables? or is there any way I can model the break point in 2008 into a VAR model using the data set from 1996-2015?

Any responses would be appreciated.

Cheers,

Sam

[dakila]

Hi Sam,

You need to convert the quarterly data into the monthly.

[samewing7]

Hi Sam,

You need to convert the quarterly data into the monthly.

Hi dakila,

Thank your for your reply.

I can not convert my data into monthly data as this is not available for all variables.

Can you think of another solution?

Thanks

[trubador]

Converting to monthly frequency would create other problems (i.e. increasing the noise). These things are very difficult to know in advance, so you need to examine the results post-estimation. Start with traditional VAR and see if that works for shorter lags. You can reduce the number of endogenous variables in your VAR if need be. You can make your decision looking at the usual model diagnostics. Structural breaks depend on the model formulation, so it may turn out that there is no breakpoint in your final model. If you really need to use all of the endogenous variables along with the longest possible lag structure, then I definitely suggest the use of Bayesian VAR.