I have 100 Investment funds (Flexible allocation Morningstar category, same investment area, currency and distribution status: the sample is homogeneous) over a 10 yrs period. I want to estimate a model (AR/MA/ARMA) for the mean and for the variance (ARCH/GARCH) in order to study time dependency of this specific fund category. My doubt is: if i consider for the time series analysis, each fund alone, I probably will get 100 similar models (cause the sample is intentionally homogeneous); But I'm not sure of that!

In order to do so, is it better to consider a multivariate time series analysis or N univariate time series analyses?

NB I am sorry if my question isn't clear, english is not my first language, even less time series terminology.


closed as unclear what you're asking by mkt, whuber Aug 10 at 15:14

Please clarify your specific problem or add additional details to highlight exactly what you need. As it's currently written, it’s hard to tell exactly what you're asking. See the How to Ask page for help clarifying this question. If this question can be reworded to fit the rules in the help center, please edit the question.

  • 2
    $\begingroup$ Hi Welcome to Cross-Validated: Can you be a little bit more clear about what you mean by "analyze" ? $\endgroup$ – Skander H. Aug 10 at 15:16
  • $\begingroup$ I edited my question, now should be better. Thank you. $\endgroup$ – CB18 Aug 10 at 16:51

You have to be more precise. First of all your question may be re-expressed as “should I use a multivariate analysis or N univariate analyses?” Well, if you want to maximize the amount of info used by your model (which would be optimal), then you have to use multivariate analysis for the obvious reason that the model will capture the interaction between the time series.. however, this must be weighted against the additional computational complexity, which may be high if you have 100 time series.. However in this case you do not have too many data points as you are working with very low frequencies, so:

  • on one hand, this is a con because you have a few data
  • but on the other hand, the good news is that your computational complexity will not explode (however, remember that you will pay for this in the form of increased standard errors in light of the tight dataset of 12 months for 10 years datapoints).

Clearly the multivariate analysis has a lot more parameters than the univariate one on the same number of funds..

  • $\begingroup$ Thank for your answer! You think would be better if I'd use daily returns instead of monthly returns? $\endgroup$ – CB18 Aug 11 at 9:41
  • $\begingroup$ If all your funds have daily Nav (price) absolutely yes, so that you will have more data which is better all other conditions being equal. Clearly this suggestion holds as long as that the objective of your analysis is still met with a higher frequency (otherwise stay with the monthly ones). If some of the funds have weekly nav then use weekly.. in other words try to be as granular as possible provided that the gradual frequency still meets the objectives of your data $\endgroup$ – Fr1 Aug 11 at 11:24

Not the answer you're looking for? Browse other questions tagged or ask your own question.