In a assignment we are given macro economic indicators like GDP, Consumer price index, Producer Price index and Industrial production index. Also we are given Crude oil, Sugar prices and FM-CG Sales. We are required to forecast future quarter sales and give a model. As I'm new to this subject, I don't know where to start with it, or what to read. Can anyone provide me with some examples of what to do, or any PDFs which might be helpful.

  • $\begingroup$ I realize you're not sure where to start, but this is quite broad. Maybe you can at least clarify what you want: a paper, a tool, a use case? what is your level of background? $\endgroup$ – Sean Owen Mar 14 '15 at 22:10

If you are new to forecasting, I would recommend starting with a very simple model for your sales, e.g., Exponential Smoothing, which will use only the sales data and no external information, but will capture trend and seasonality. Here is an extremely good free online textbook.

Once you are comfortable with your simple model, you can look at more complex ones that do include external causal variables. For instance, you could run ARIMAX models (the X stands for eXternal or eXplanatory variables). For this, look at chapter 8 in the online textbook and check the auto.arima() function in the forecast package for R, where you can specify external variables using the xreg parameter.

Alternatively, you could look at other methods like Neural Networks, or possibly even Vector Autoregression. To learn more, you could look into standard econometrics textbooks.

Whatever you do, keep two things in mind:

  1. If you include explanatory variables in your forecast, you will need to forecast those variables themselves. In an assignment-type situation, you may be able to work with actual future macroeconomic variables, but of course in a "real" forecasting situation, you don't know next month's sugar prices yet. Forecasting these to feed the sugar price forecast into your sales forecast model adds an additional bit of uncertainty.
  2. You may just find out that your fancy model with lots of additional explanatory variables does not do a much better job at forecasting than your original simple model. This would not at all be surprising. Simple methods very often outperform complex ones in forecasting.

Not the answer you're looking for? Browse other questions tagged or ask your own question.