Forecasting Methods made simple - Moving averages

This is the first video in the series of 5 videos on the Forecasting methods made simple. Here the Moving average method and the Weighted moving average method for forecasting are described.

[@Kanele404]

Great video series

[@lesliebillhorn7794]

The two videos I've watched of yours so far are excellent! I actually understand the material as opposed to when my teacher tries to explain it. I'm so glad I found this :) Thank you!

[@piyushashah1]

Thanks for the compliment Leslie. Btw, I'm sure your teacher also has some perspective in teaching the way she does.

[@PanditJi]

excellent sir. love it

[@shakirullah5840]

Excellent !

[@PeterFortuna]

Thanks for sharing this!

[@rushadas1321]

Excellent!!!

[@chiki3850]

Hi Excellent video! I Have a question, What would be the best Forecasting method for longer duration? as you have mentioned that moving average is best for shorter periods only. Thanks

[@Alecohen1000]

Thanks!

[@djamalchaabane6706]

I did like this vedeo. Indeed it is very didactic, simple and easy english enables a medium student to follow

[@piyushashah1]

Great. Did you look at videos on other forecasting methods?

[@kalakar88]

A really nice video....kafi kam samay mein basic cocept clear hua......aise aur bhi mgmt topivs ke video upload kijiye sir.......thanks

[@piyushashah1]

Hey Kalakar, thanks for leaving a comment. I do have quite a few videos on my channel...have a look. Also do share the videos with your friends.

[@noelfernandes1221]

It would be really helpful to attach a link of the excel sheet so we could practice it as we watch the video.

[@Diego-ck9zl]

Hi, is there a criteria to choose the periods for the moving average? I mean, why 3MA or 4MA? Can´t it be 5MA o more? Do they have to minimize the error?

[@rohitbalabhadrapatruni1728]

This is good..Depending on Data pattern how much duration moving averages should be considered if data is highly fluctuating with more variation and also data with less variation ?

[@piyushashah1]

A simple rule is that the method that has the least error is good for that purpose. Forecast the past with different methods and check the duration that works best. Use that (and keep praying that the pattern of the past continues!). You can see this video on this: kzbin.info/www/bejne/jYqlfGekbJqelZo

[@anjanaharidasan]

Thanks ! I have a queation. Can we choose any value value as the weight?

[@piyushashah1]

+Anjana Haridasan Use weights so that you get the best possible predictions for your past data. There is of course an assumption that the past will repeat in the future.

[@1rasha]

I have one question? When we already have data for different months (here we have 20 months for example) then why we are forecasting at all?? What I am looking at is forecast beyond 20 month period. Can you suggest?

[@piyushashah1]

Good question Rayees. We do this to actually find out the best alpha for us. So, we predict past data with different values of alpha, check what gives least error and then use that alpha value for forecasting. For forecast, as and when new actual demand data becomes available, we forecast for the next period. Hope this helps.

[@hurifury961]

hi, thanks so much for your video, its the best explicit i've found so far, although it took me a while to find it on youtube could it be the title of video the cause...thanks again it helped me a lot! :)

[@piyushashah1]

Thanks for the feedback Kary. What do you think I should rename the video as?

[@hurifury961]

just something simpler like forecast methods, or basic forecasting just take the mechanics out ...because its really the best video but it took me a while to find it when searching youtube other videos appeared on first page, i waisted soo much time with the other videos...then found your and had my answers, thanks :)

[@piyushashah1]

I've changed the title. Now lets see if this improves the number of viewers. Thanks.

[@hurifury961]

ur so welcome :)

[@hurifury961]

Kary Deea also u can add more search engine words if you describe the video with lot of words, i think this is how the other do it...

[@MegaSoumo]

@ Piyush, At 2:09, forecast of 3rd month should be average of 1st, 2nd and 3rd month or it is what you said in video, forecast of 4th month is avg of first three. Can you please clarify and what different does it make.

[@piyushashah1]

inshu bhatnagar We will need the data of the past to forecast the future. So, the average of the actuals of the first three months will be the forecast for the 4th month in a 3 month moving average. Hope this clarifies.

[@MegaSoumo]

Piyush Shah Thanks for quick response

[@bearingsandaccessories_admin]

In calculating the weighted average for series 4, you included the actual value for series 4 in the calculation. Is this, in fact, correct or should you be only using the values from series 1 to 3?

[@piyushashah1]

Vincent Elliott Yes. The forecast for the 4th month in cell C5 is calculated as the mean of demand of 3 previous months - cells B2 to B4. I have not included the actual demand of month 4.

[@bearingsandaccessories_admin]

Thanks. I got the series and cell numbers mixed up. Great set of tutorials!

[@piyushashah1]

happy that you liked it!

[@jacksondsilva707]

is there a formula or some method to determine whether which of the two is more reliable (3 months simple moving average or 4 months simple moving average?)

[@piyushashah1]

One way would be to forecast past data based on both methods and choose the method that has less error. The assumption here of course is that the future pattern is similar to that of the past.

[@shruthil7913]

Thanks for sharing, I have one doubt - why m-1 = 0.5, m-2=0.3 & m-3 = 0.2 ? any reason behind it?

[@piyushashah1]

Shruthi L These are just random weights I have used. It could have been anything. Generally the sum of the weights is 1, else you have to divide the sumproduct with the sum of the weights. Also, the near period generally has a higher weight than the distant period.

[@vickkyfarooq]

How to forecast for more than one week or year or serial number?

[@piyushashah1]

For moving average and smoothing based methods, the assumption is that all variations are random, and we need to filter them out. Believing this, we claim that non-random component of demand is constant and the forecast for the next week / year / sr number is going to be the forecast for all future periods.

[@leratosemosa2416]

I have demand data for 6 months and I am asked to calculate the forecast for month 7 and month 8 using 6 months simple moving average. I can Calculate the forecast for month 7 but how do I calculate for month 8 as I don't have the actual demand for month 7. Please assist

[@piyushashah1]

The average of the 6 months of demand data is the forecast for month 7 and month 8 as well. When you get the demand data for month 7, you can use that to update the forecast of month 8. Hope this helps.

[@meyyappanlakshmanan5169]

Piyush - I have Incidents for month 1 to month 10. I need to estimate incidents for month 11, 12 and 13. Is it possible to moving average ....

[@piyushashah1]

I would find the reason for those incidents and try to model them using some regression technique. So, workplace near misses could be predicted by overtime.

[@asthatripathi9211]

Thanks but On what basis are you choosing weight for m-1 m-2 m-3?

[@piyushashah1]

One way is to look at past data and select a method that predicts the past with the least error.

[@asthatripathi9211]

okies..thanks a lot

[@AlejandroSandovalMurillo]

Can you develop more on deciding or mathematically calculating the beta value? Or where can I find information on it? Thanks... excellent videos.

[@nattaphornkhantirach8148]

I have a question regarding random situation. I understand that Covid-19 is an abnormal situation and it has affected our sales. In this case, in order to forecast our sales for the next six month, how can we handle this abnormal sales data?

[@piyushashah1]

The basic assumption of all these methods is that the past trends, cycles and seasons will repeat in the future. With Covid, this assumption is invalidated and hence I would not recommend using any of these time series based methods for forecasting in the current time.

[@nattaphornkhantirach8148]

@@piyushashah1 Thank you so much for your suggestion. This helps a lot. Your video is very helpful :)

[@DIGITAL_COOKING]

i missed how you puted 0.5 , 0.3, 0.2

[@piyushashah1]

These are random weights that I chose to use for this example. You can have different weights. Forecast the past data with different weights, and select the weights that give you the highest accuracy in past prediction. Have a look at the video on "Measures of Forecasting accuracy" for details on how to do this.

[@DIGITAL_COOKING]

ok, thanks

[@1rasha]

Videos are not clearly visible

[@syekaaa]

i don’t understand😤🤯😔