4
$\begingroup$

Each financial quarter I collect data on the number of potential clients, contacted potential clients, and potential clients that become actual clients. I have this quarterly data going back only 6 years (6*4=24 total time steps). If I want to generate forecasts for this vector (three metrics) for the upcoming quarter what are my best bets in terms of statistical/ml models? The business is very dependent on the economy so I think it would be interesting to include a few external quarterly metrics such as unemployment rate to improve the forecast if possible. Can anyone help me reason through the best way to approach this?

I don't often deal with time series data and when I do it typically has had tons of data available so often go to more complex models (many-to-one LSTMs, GRUs etc) but I don't think those will work well in this case. I would like to do better than just a moving average, does anyone have any insight into what models would be most appropriate here?

Improve this question
$\endgroup$
2
  • $\begingroup$ Scarce data -> Bayesian statistics $\endgroup$
    – John
    Commented Jul 9, 2018 at 21:02
  • $\begingroup$ Or a good model -> Can we have a graph ? $\endgroup$
    – AlainD
    Commented Jul 10, 2018 at 13:16

1 Answer 1

Reset to default
1
$\begingroup$

I would initially just focus on the response variable and consider an ensemble forecast where you weight the predictions from a few models (can just simply average them for lack of anything better). Using a few models will help prevent any single model from causing a large miss in your forecast. Some of the models worth considering might be:

  • seasonal arima
  • seasonal holt winters exponential smoothing
  • triple exponential smoothing
  • some form of naive forecast (perhaps last year's quarterly value)

There's some nice discussion on ensemble approaches at the following link:

Ensemble Time Series Models

Improve this answer
$\endgroup$
1
  • $\begingroup$ This is super helpful thank you. Do you have any thoughts how I might go about predicting each piece of this problem (total potential clients, contacted, became actual clients). Should I make a separate model to forecast each piece (this could have the odd result where the number of contacted clients is more than the number of potential clients which doesn't make sense, but maybe it doesn't matter)? Should I try to forecast the conversion % rate instead (% of potential customer contacted etc)? Any thoughts? $\endgroup$
    – Adrix
    Commented Jul 10, 2018 at 16:40

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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