Forecast method

Ron04

New Member
#1
I have developed a Timeseries Multiplicative Triple Forecasting model. (Holt Winters model as baseline) . With my Alpha,Beta and Gamma parameters I am able to forecast future demand.That is not the issue. It provides 26 .8 % reliablity.That obviously has to be at least 60 % or higher.

Now for automatically function and calculating the advised and optimal Alpha,Beta and Gamma ( I currently do this manually without Excell formula and use Wessa .net ) , I would need help to further develop this model. I have this set up in Excel.
Is there anyone on this forum who can help me with this?
Maybe advise to use another Online Model or Statistical program ? or advise how to improve within Excel and can advise me on formula's to use?
How do i achieve a 60 % or higher forecast reliablilty ?

Cheers.Ron
 

Attachments

#2
Hi Ron,

Thank you for posting on this forum. I struggle with understanding how you are measuring the accuracy of your model as it stands currently.
"Reliability" isn't a measure I am familiar with, but you should look into mean absolute percent error, mean absolute error or mean percent error. Those are common metrics to evaluate time series models.

In terms of knowing if it's a "good" model. I suggest that you use a simple naive approach by seeing how well it predicts compared to just predicting yesterdays value (or whatever time interval you're using).

Without seeing your time series, I can't tell you what models to try. Time series modeling is a rigorous art that can't always be brute forced like other prediction problems. I would look into the ARIMA family of models and Exponential Smoothing.

Take care
 

Ron04

New Member
#3
Hello J,
I found a free course to learn R. (at page 17 of 180...so I have some to go still :)) . My first learning shows that I need to create a working directory . I have just learned to add 2 numbers.
So I am far away from creating a forecast. I believe its called a Predict model in R.
I don't even know if someone already made a predict model with a relation to an imported file., so I can learn from that and down the road I could understand what it does. I would need a predict model as where it calculates already what the gamma, beta and alpha values are for every singles lines based on previous values and have a predict accuracy of 70 % or higher.(confidence interval of 95 % at least) .As a Six Sigma Black Belt I am familiar with statistics , but then again Statistics is much more than only Six sigma.
Cheers
Ron