How to Holts Winters Method in Excel & optimize Alpha, Beta & Gamma. Want to know more and see the slides & data files shown in this video? Sign up for my free course on Seasonal Forecasting! www.bcit.ca/fr...
Legit this was the most helpful tutorial on applying this forecasting that I have been able to find. Well done! Intelligence is when you can take a complicated topic and simplify it for anyone to understand, and this definitely did that.
You are actually a saviour. You are legitimately the ONLY person who actually knows what to do to forecast beyond 1 season out of the sample. You have earned a sub. Thank you
Thank you so much you are saving me right now. They don't teach anything about how to use the method in class and expect you to do it on the assignments.
The way you initialize the seasonal, level, and trend parameters was helpful to see. Solver is a bit sketchy if the parameter space is too large and that keeps the list of parameters down to just alpha, beta, and gamma. Thank you.
fantastic, now i can run an educational seminar whenever someone asks "how are you generating these numbers" during any S&OP, usually this is often asked sarcastically or in a condescending way.. lets see how clever they are now
This is brilliant. I am studying forecasting and was hoping to see someone doing it in an Elegant way. You have delivered. One thing I am still struggling with is Normalising Seasonality Indices. Could you please explain how to incorporate normalizing into that workflow in an clean and elegant way? ( My biggest doubt is that whenever normalizing I would retrospectively modify historical data and I don't think I am supposed to do that)
Thank you, Stay Focused for your kind words! Feel free to message me via LinkedIn (www.linkedin.com/in/leslie-major-943a7610/) with a data set where we could normalize the seasonal indices for you.
Hi Leslie, I think your video is very valuable. Just your dataset here is "superpredictable" being actually the rule of thumb for beginning learning time series ("airpassangers" data??) which is worth mentioning and thus clarifiyng that's the reason why the errors here are so low and that's probably not going to be that easy in the practive. Also you do not provide the holdout set error which minimazing is the overall goal of the forecasting. Great work anyway!
Hello Leslie and thank you for your contribute; I'm interested to better understand the "time zero" phase, in other words the initial parameters set up. KInd regards and thank you. Bruno
Very clear!Thank you. For the calculation of initial values you've said (according to me...)that it's not very important because you're more interested by the future (forecasting)than by the past. So why calculating RMSE from the beginning to the end to appreciate the pertinence of the model for forecasting? Could'nt we only calculate RMSE for the last quater( or third part) of the last data and results to get a model closer to the present ( and forecast future)?Thanks
Hey Leslie.. the video was very helpful! Massive thanks for that!.. Is it possible to calculate the AIC value of the model in Excel, and minimise it based on that?
QUESTION: I forecasted a line into the future; now i want to add confidenz corridor. Which data should i use? the forecast data (for every period or just the future one??) or should i take the level for that??? GREAT video thank you very much!
How can I build forecast.ets Excel function essentially by hand so I have more control over inputs? This is for seasonality as well. I ran forecast.ets.stat so I have the smoothing parameters. Do you have videos on the equations so I can reproduce the results of the exponential triple smoothing by hand? Is this video the same thing? Or is that more complicated?
Dear Mrs.Leslie Major, I am Anna from the Netherlands, I study applied mathematics and at the moment I am busy with making a website regarding a course. My question is if I could get your permission to use this RU-vid video fot my website? I am looking forward to hearing from you. Beste wishes, Anna
Hi Leslie, this is well explained in comparison to some of the other videos we have been watching here. I have a question, does it matter how much data we have? For instance, I have 33 months of call volume data for our model.
Hello hope you're doing good, I was just wondering how to calculate the alpha beta and gamma as I'm just starting to understand how this method works, thanks in advance.
Very interesting excercise, just a quick question, I didnt see id there was a normalization step for the sesonal factors, i believe is a necessary step to make sure they sum up to 4, perhaphs is the one step that makes it difficult to keep track of this forecast in spreadsheets,
seems like an error in the forecasting formula put in excel at @1.54. The trend of previous period is to be multiplied with "k", let me if there is something else i am missing.
Hi, Just a question, why do you use the last values and not an average of all values? I understand that the initial values were where you had the most errors and last values are the best information we have towards the present of the company, but past values can correct for outliers no? Thank you for the video, really helpful.
Hi leslie, thank you so much for the tutorial!! As ur tutorial is for quarter, would u mind to share with me how if the season need 12 (monthly)? How to calculate them all? Thanks in advance!! 🙏🙏
I am using your tutorial on M=12. I am curious if there is an fault in your Initial Values formula or not. It says "Use formula for seasonal value S12. It should be S13 right? Also is this additive or multiplicative? Thank you
Hello Madam, Thanks for the video, one BIG Doubt i have is How you can do Linear regression forecasting for multiple SKU without using the Data Analysis>> Solver method, like the similar method in the video where whole Model is Dynamic. I am really finding difficult to find the solutions. Please Help!!
Hi Asha. That is such a great question! I need to find the answer to that i have used quite a few different textbooks over the years in teaching and i grabbed these initial values from one of them. Note, the initial values arent very significant if you have a large data set - the forecasts get refined over the timesteps so you can almost start with any initial values you choose. If you also want to get "fancy", you can optimize your initial values!
Those are only methods to optimize the parameters. They will likely lead to very similar results. Evolutionary is more a "brute force", taking more time to converge, but solves functions with even less linearity, where GRG gets locked in local optimal (not our case here).
Is anyone else encountering the same problem I am: When I go to optimize Alpha, Beta, and Gamma in solver, It keeps setting Alpha to 0 and Gamma to 0. Beta is something between 0 and 1 as expected. I get that mathematically assigning 0 to alpha and gamma may be making my error come out minimized, but it defeats the concept of doing winter's method forecasting if you're not consistently updating your level and seasonality factor (if alpha is 0, it's not modifying your level. and if gamma is 0 it is not modifying your seasonality).
Hmmm. Are you able to send me your file so i can have a look? It is possible gamma optimizes to 0. Usually that means there actually isn't any seasonality
And further to that comment, if beta optimizes to 0 then there isn't any significant upward or downward trend. Kinda cool - the method itself will help you see what is happening with your data.
Finally, if alpha optimizes to 0, it means you are never using your current data value in your current level calculation, instead you are taking 100% of your previous value