Search code examples
mathstatisticscoefficientstrend

How to split single didgit on periods with apply two or more modifying coefficients?


For example I know that we sold 5000 pieces of something in previous month. I know that month split into weeks and every month we sell on 20% more than in previous month. This is some kind of global trend that goes for years.

So I can know exact amount of sales every week by using this scheme: Month split into 4 weeks and we sold 5000 things in that month. First week we selling 100%, second is 105, third is 110 and last one is 115.

We can use this as coefficients:

1 week    2 week    3 week    4 week /*Total 5000 things*/
100       105       110       115    /*we can summ coeffs and get 430*/
0.2325    0.2441    0.2558    0.2674 /*we divide 100/430, 105/430 and etc to normalize it*/
1162.5    1220.5    1279      1337   /*we multiply 0.2325 on 5000 to get sales amount in first week and etc*/

And that's really working, each of these values 1162.5, 1220.5, 1279, 1337 greater than previous on 5%.

But that's where things gets unclear. We also know that there is seasonal changes. We know that fist week we sell a little bit less, than in second, in second a little more, then third, and third is way bigger, than last. We can describe that seasonal changes as another set of coefficients:

1 week    2 week    3 week    4 week /*Total 5000 things*/
0.9       1         0.9       0.5    /*summ is 3.3*/
0.2727    0.3030    0.2727    0.1515 /*normalized*/
1363.5    1515      1363.5    757.5  /*real sales per week*/

So we calculated separately each week sales like if we had global trend only and seasonal changes only, but how we can mix those?

I don't know what to do with 1337 and 757.5 for last week and etc.


Solution

  • If your second table is really direct week/week ratio and it is the same in different months as it looks from your second example, then I don't understand where your trouble is. What's wrong with simple

    amount(month_index, week_index) = 5000 * ((1.2) ^ month_index) * normalized_week_coefffcient[week_index]
    

    In other words first use your 20% per month to calculate total amount in the month and then split that total amount between weeks according to normalized week coefficients from your Table #2