I realized I was doing lots of threads on the same topic so I decided to consolidate my questions here.

I have read time series for years (I am not a statistician, I have graduate degrees that involve statistic, but effectively no knowledge of calculus or matrices). Certain questions continue to elude me in part because various writers on sites and books disagree with each other. Here is a very basic, but important question. When do time series concerns apply and when do they not? That is when must you do time series and when can you use cross sectional approaches?

I used to think that this was simple, if X occurs before Y than you do time series. But of course X always occurs before Y if it influences Y. We answer two types of questions. First, those where the results are measured at one or two distinct points in time (six or 12 months after a customer leaves). For us, spending influences (we assume) results, so types and quantity of spending is the predictor and say income gain is the result. But spending can take place at many, many points in time, it is rare for it to occur at just one point (and we are not always certain when it is actually provided just that it has occurred in the past). And the result takes places years after the spending with a customer getting many types of spending over time. Further, one case can have the spending at say a year before the result is measured another six months and so on.

For a second type of issue we deal with we need to know how spending each month is influenced by various factors. We have no theory, I have found none, to build predictions on. We are a government agency and the literature I have found contains no time series analysis.

Given the much greater complexity and judgement called for by all time series methods I know (for example VAR/VECM] I would always prefer cross sectional to time series methods - particularly since the later generally make it impossible to include the many control variables I feel are necessary.

  • $\begingroup$ People attempt to model the fact that the effect doesn't occur all at one time by using distributed lag models. The terminology varies depending on the field so you might try googling distributed lags, or arima model with exogenous regressor or koyck distributed lag etc. there are many different terms that are used but they're all similar and try to deal with the fact that the predictor effect is spread out over time. Whether it can be helpful or not is a different issue but that's usually the approach. it's often used in advertising versus sales. the literature is enormous. $\endgroup$
    – mlofton
    Commented Feb 24, 2019 at 22:16
  • $\begingroup$ I spent a fair amount of time studying ARDL models. One problem I have with them is what you do when you have predictors integrated of a different order from each other or the dependent variable. With my data I am not certain it even makes sense to use time series data? $\endgroup$
    – user54285
    Commented Feb 24, 2019 at 23:50
  • $\begingroup$ My concern with multivariate ARIMA is that there is a lot of judgement (and of course time associated) with such analysis. I am concerned with my ability to estimate correctly one series, several makes this a lot worse. $\endgroup$
    – user54285
    Commented Feb 24, 2019 at 23:59


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.