For a time series object (in terms of theory/concepts), do we have to have the data for each date? In some cases the customers do not make purchases/sales amounts during the same month. For example, I want to predict the sales for each customer based on their dates. I am missing some dates for sales, would this be a problem for prediction? The data below:

> amount <- c(10,11, 50, 12,30)
> dates <- as.Date(c("2012-01-01", "2012-02-01", "2012-05-15", "2012-08-15", "2012-12-10"))
> data <- data.frame(amount , dates)
> data
  amount      dates
1     10 2012-01-01
2     11 2012-02-01
3     50 2012-05-15
4     12 2012-08-15
5     30 2012-12-10

> ts(data)

We can see that row 1 and 2 are consecutive, however 3rd, 4th, 6th, 7th, 9th, 10th, and 11th month are missing. Would this be a problem?


  • 1
    $\begingroup$ It all depends on your model. Inserting zero $amount$ for the missing dates could be of help for models that need consecutive data points. $\endgroup$ Feb 3 '15 at 20:48
  • $\begingroup$ Thanks! I was thinking that too, I was not confident. I am bit new to Time Series, is required that time series need consecutive dates? $\endgroup$
    – sharp
    Feb 3 '15 at 20:56
  • 1
    $\begingroup$ Most of the models (like ARIMA, VAR, VECM) normally do need consecutive dates (although there are models like Autoregressive Conditional Duration model where the dates themselves are being modelled and need not be consecutive). Do you have an idea of what model you are going to use? $\endgroup$ Feb 3 '15 at 21:00
  • $\begingroup$ @RichardHardy. I was thinking to use ARIMA. I have not yet worked with Autoregressive Conditional Duration. I'll have read couple article about this model. Any good article should I refer to about Autoregressive Conditional Duration or ARIMA? Thanks a bunch! $\endgroup$
    – sharp
    Feb 3 '15 at 21:09