4

I would like to stay away from finance and economic examples since they are too abstract for me to understand. Are there any "real world" examples for example healthcare, exam marks, environmental science and so on? From a managerial perspective I would like a basic understanding of how VAR could potentially be used to model real world phenomenon.

deltanovember
  • 277
  • 3
  • 5
  • 2
    I highly doubt simply switching the topical material will help you understand the technique any better (even if you have a greater understanding of the theoretical material). Perhaps simply asking for an accessible intro to VAR would be more enlightening? – Andy W Jul 13 '11 at 13:11
  • @deltanovember, and economics is not a real world example why? VAR is used for time-series and its main application area is economics. As for real life example in one project I used VAR to estimate how macroeconomic factors influence the sales of different types of confectionary products. – mpiktas Jul 13 '11 at 14:06

2 Answers2

5

Do not get hung up by the Finance or Economics tags.

VAR is a time-series technique. This bears repeating: time-series, and nothing but. Its key innovation, back in the day, was that it removed all underlying domain-specific information encoded in a model and reduced it to pure time series interaction: in the simplest case, does the past of series A and B affect either one or both of A and B? If one series is shocked, how long does a change persist etc. Economics, particularly macroeconomics, simply became a big application area as there are often time series that are interrelated, and a model-free way (in the VAR sense) can not only be useful for model fitting but also for forecasts from these models.

Dirk Eddelbuettel
  • 8,872
  • 1
    What do you mean by "model free"? As far as I can see, VAR is definitely a model. It is a flexible and complex model, but it is still a model. Perhaps you mean a more realistic description of the "real world" compared to simpler methods? – probabilityislogic Jul 13 '11 at 14:27
  • 1
    In the economics domain, something postulating that consumption reacts in form (specified by some model) to income or changes therein because some micro-eonomic theory postulates it as such. A VAR model would just relate the time series--and is (in this sense) 'model-free'. – Dirk Eddelbuettel Jul 13 '11 at 14:31
  • 2
    So you really mean "theory free", in the sense that no "causal mechanism" is postulated between the two series? The phrasing "model free" gives the impression of "assumption free", which VAR certainly is not. – probabilityislogic Jul 13 '11 at 14:59
  • 1
    Yes, I think that is another good way to put it. – Dirk Eddelbuettel Jul 13 '11 at 15:02
  • As fas as i know, VAR is a "Linear" model, so I dont think it is model free in a sense. You can argue that it is model-free because you dont need to know the causal mechanism as long as the relations are linear, but i don't think anyone would agree that it can be called 'model-free'... – KH Kim Sep 16 '18 at 08:18
1

I'm using VAR to model retail employment in census blocks near transit stations. My vector is the time series of the 300+ census blocks within 1/2 mile of a transit station. I am attempting to find out if the total retail employment is significantly different after the advent of transit than before, and if the total employment is significantly different than it would have been without the addition of transit. So I'm using VAR to model my time series, using my VAR model to predict new data, and then comparing that new data to actual data during the years after the transit station opened.

Mox
  • 305
  • How do you justify using VAR since it seems to me quite a limited model(after all it's linear)! I do most econometric model are linear or just make outcome variables on a log scale but it also looks quite limited model! – KH Kim Sep 16 '18 at 08:24