0
$\begingroup$

I have to make predictions about sales on a monthly base and I already have historical data from January 2011 until June 2015. What forecasting method should I use if my data is influenced by seasonality?

PS: I already used Double Exponential Smoothing for the previous months but the deviation is quite big!

$\endgroup$
  • $\begingroup$ post your data to dropbox.com $\endgroup$ – Tom Reilly Oct 9 '15 at 18:13
1
$\begingroup$

The simplest and often best method to forecast monthly data (that may be seasonal) is seasonal Exponential Smoothing - with or without a trend component. Here are details. (Incidentally, I'll never tire of recommending the entire free online forecasting textbook.)

You write:

I already used Double Exponential Smoothing for the previous months but the deviation is quite big!

The problem is that often series are simply not very forecastable. There may be a lot of residual variation. Or there may be unmodeled drivers, such as promotions. If you know about such external drivers, then include them in a causal model, e.g., a regression model with ARIMA errors.

Conversely, if you don't have any information about factors that may influence your time series, there may simply be a lot of variation that you simply cannot explain. A different model won't help you there. The only thing you can do is to understand the residual variation and cope with it - e.g., by setting appropriate safety stocks, or by working to reduce the variation, possibly by working with your clients and trying to get orders earlier in advance or some such.

$\endgroup$
0
$\begingroup$

There are a lot of different ways to forecast, depending on which application you are using. If you were to use R, two of the most common forecasting methods are an Arima model and a form of Exponential smoothing model. Additionally, if you use R, auto.arima() can find the best Arima model and ETS() can select the best exponential smoothing model. The creator of the forecast package, Rob Hyndman, recently tested these two methods and how they perform against other commercial software packages, and here are his results.

http://robjhyndman.com/hyndsight/show-me-the-evidence/#disqus_thread

But as the previous post said, Rob Hyndman's online book is free and a great resource for learning a solid foundation of forecasting (and has real examples with code).

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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