# Building time series models on data with high % of missing values

I am trying to build time series models for prediction of monthly sales volume of a product. I am given data for monthly sales volume of different products over the last 5 years, and I have to build individual models for individual products.

Let us assume, I have only 1 product in the dataset.

The problem I am facing is the sales volume for most of the months in the data is missing. Something similar to the following.

year  month sales volume (in M)
2011  10       3
2012  7        4
2013  7        4
8        4
2014  4        5
7        5
10       5
12       7
2015  1        5
4        5
2016  4        6
6        7
12       8


I am confused as to how do I deal with this kind of irregular data and build a forecasting model?

• First, why is it missing? Did you loose the data, ... or maybe the missing really means there was no sale that month, that is, zero? – kjetil b halvorsen Apr 6 '18 at 18:37
• I lost the data, to be more precise, this dataset was given to me by someone else, and they seem to have a data logging issue. – Nitish Apr 6 '18 at 19:12
• Well, could possibly the "data logging issue" be caused by high activity, maybe at times with high sales? then the data is definitely not missing at random ... – kjetil b halvorsen Apr 6 '18 at 19:23
• Do you know that the months you have are correct? For example, in July 2012 are there really 4 sales, -- or could there be 6 sales with 2 of them failing to be logged? If that's the case, you actually have no verifiable data and personally I'd give up. Otherwise, I might try a simple "main effects" model with annual trend and seasonality, to impute the missing months. But I hope your job doesn't depend on a highly accurate forecast :) – zbicyclist Apr 7 '18 at 13:50
• It does look like there might some issue with the data source itself, I am re-checking it again with the source. Thanks for your suggestions. – Nitish Apr 9 '18 at 5:03