I've been working on a weekly web traffic forecast based on time series. The primary business driver is marketing investment (spending)
I have historical spending data but not future spending data. In this case, how can future web traffic be predicted?
I'd like to have an input request switch where I could change the cost input to see future traffic. (If you spend this much, you will receive these many visits; if you do more, you will get more, and vice versa.)
You can either use your explanatory variable(s) or not.
If you do not want to use your predictors, you can use simple time series models, like exponential smoothing, ARIMA (both not recommended if you suspect intra-yearly seasonality - better to use an STL method for such "long" seasonalities), or many others.
If you do want to use your predictor, you can use simple methods like regression with ARIMA errors, or alternatively run a regression on the predictors, then fit an exponential smoothing model to residuals. (See above on potential seasonality.) Or do a fancy ML method, like Deep Learning, transformers, or Boosting... but quite honestly, I would go for low hanging fruit like a regression first.
However, in the latter case you will need to feed some future predictor values into the model to get a forecast. You can have the user specify future marketing spend, or use any "reasonable" number, like the historical mean spend, or the last spend, or indeed the result from forecasting the spend itself.
