# Variables with zero variation in time series random forest estimation

I am trying to build a forecasting model for three products. Because I have only 20 -25 daily readings per product sales and these series have gaps during the time span of analysis, I switched to the cumulated yearly sales series.

By using them, I am trying to forecast next year's sales figure. In that sense, my whole estimation process changes to cross-sectional exercise.

However, I have some macro variables such as "inflation", "consumer confidence" etc... for three years. Because my all variables are in a yearly aggregated state, and I have limited data to convert my project to a panel time series model, I only have zero variance economic variables. Such as a variable for 2015's inflation, a variable for 2016's inflation or 2017's inflation. Do you have any suggestion to add these economic variables to my data's current structure?

• there are several concurrent themes in your question. Suggest to narrow down the focus. The question title suggests zero deviation whereas the question body mentions zero variance. Please see this post and learn how to ask a good question. – mnm Jul 25 '18 at 10:26

Simply use a predictor inflation and insert the 2015 value for your 2015 observation, the 2016 value for your 2016 observation and so forth.