EViews.com

Dear Eviews veterans,
Using Eviews 6 I ran into the error "Unsufficient number of observations to estimate 9 coeficients per equation in VAR" while trying to estimate a VAR.

My data concerns multiple quotes per second for two hypothetical exchanges. I took the data from the NYSE TAQ database. While they TAQ explained me the time-stamp is only available per second, they did confirm the quotes are in chronological order. The data looks more or less like this:

(Example bids)
DATE TIME BID NYSE BID NON NYSE
03-22-2009 10:30:45 4.67 NA
03-22-2009 10:30:47 4.66 NA
03-22-2009 10:30:47 NA 4.65
03-22-2009 10:30:47 4.67 NA
03-22-2009 10:30:47 4.66 NA
03-22-2009 10:30:48 NA 4.65

I am researching the responses these markets give to eachother. So how fast do markets adjust, using how many lags. As the data is not available with a more detailed, frequent timestamp I thought this can only be done in Eviews by estimating a VAR model. I think the error I run into has to do with the NA´s turning up in the data as a VAR estimation requires a common, continuous sample for two series (say stock quotes per day). I do not know however how to fix this here. I suppose restructuring the two series to become one should work, but then you don´t have two series anymore.

Any suggestions? Please let me know if I need to be more more specific.

Thank you very much!

I do not think that the ultra high frequency data is the right choice to analyze the interrelation between the markets. You need to aggregate your data and use lower frequencies if you want to make a valid generalization of your empirical results. On the other hand, VAR may not be the most appropriate technique to model the relationship among markets. Analyzing the market microstructure data requires specific econometric techniques (e.g. duration models, decomposition analyses), since it has some unique characteristics that do not appear in lower frequencies. If you are willing to work at lower frequencies, I suggest you to try Multivariate GARCH methods (which also require continuous sample).