Nonlinear Sales Forecasting

[Dokat]

Hi,

I have a 52 week historical sales data for various price points. The relation between price and demand is non linear. When i ran a regression it returns low RSquare and high P value which indicate regressors are not predictive. Did anyone come across the same issue and work around to build predictive pricing model in Excel.

Thanks

 

[Hui]

Is there a Monthly or Quarterly or other cycle that is more regular?

 

[Dokat]

no it is not based on historical data but unit sales at various price points. Thanks

 

[r2c2]

Are you sure Price is the only variable in this scenario? Here is how I would approach the problem:

• Plot price vs. demand (unit sales) and inspect for any visible pattern.
• Exclude outliers
• If there is no specific pattern, repeat the exercise with
  • price bands vs. sales
  • price vs. sales split by time of sale (by weekday / month / weeknum / quarter etc. depending on your industry norm)
  • Is there a mixture of patterns? (ie liner relationship up to certain price and then exponential / polynomial after that)

If you see no obvious pattern, validate your hypothesis that price determines demand. May be the relationship is weak or non-existent. Or it is masked by other variables not present in the data. So either include more columns or conclude that such a prediction is useless.

If you end up building such a model, share it with us.

all the best.