1

I am trying to predict demand of Automobiles at dealer and Variant level. I have 2 year, 2 months daily sales data. I need to predict demand at monthly level based on daily data for Enquiry.Dummy data for forecasting

Just to make it more clear, if I am on 5th of March 2017 so based on enquiries so far, earlier sales data etc. I need to predict total sales of March. Again, if I am on 15th of March so based on total enquiries till 15th, sales so far in month (in last 30 days etc.), I need to predict total sales for the month of March.

I have tried regression using lagged variables like total sales in last 30 days, total enquiries in last 30 days etc, but results are very poor. I have tried auto.arima with exogenous variable but results are not very good.

Sales are high on weekends and on March and October (Festival season in India).
I have few questions:

  1. What should be my approach?
  2. What are the other techniques I can use?
  3. I had built ARIMA at day of month level but it gives me only 26 data points. (12*2+) 31 models in total and it does not take care of weekend effect. Can this be improved?
  4. Can some state space model be used here?
Richard Hardy
  • 54,375
  • 10
  • 95
  • 219
Arpit Sisodia
  • 1,029
  • 2
  • 7
  • 23
  • Please let me know if i am not cogent enough or some explanation is required. – Arpit Sisodia Mar 27 '17 at 14:37
  • maybe this site help:http://stats.stackexchange.com/questions/269503/predictions-remain-same-for-arima-model/269560#269560 and http://datascience.stackexchange.com/questions/2368/machine-learning-features-engineering-from-date-time-data?rq=1# lagged variables always overfiting, so try some aggregate on it. – wolfe Mar 27 '17 at 14:41
  • 1
    Above site, I post an answer about how to construct the aggregate feature on the time-dependent features. – wolfe Mar 27 '17 at 14:58

0 Answers0