Suggestions for data structure to create extrapolation?

Question

I'm trying to write good code to make extrapolation on economic data and I have a hard time finding an efficient/ingenious way to do it. (to have a nice code that can be used in a different way later with ease)

Example of economic data split on an Country => year => value structure

['FR'] => [ 2008 => 50, 2009 => 100],['US'] => [ 2008 => 70, 2009 => 20]

However sometime additionnal classifications goes in : So for example ['export'] => precedent_array ['import'] => another one

I'm asked to calculate average progression coefficient per year per classification (squishing country key)

(['export'] =>[2008] => 1.1)

, then to apply it to the simpler country => year => value form, then calculate relative error on existing data.

In the end, I will have to calculate another 'per-year' coefficient from another classification, again calculate the relative error on an application, and to finally take the best one of the twos.

To resume: I have a various Key level that could go away due to an average, and still, I will need to apply/use this new data where this key_level could be present (or another one, but I think it's not the main problem)

So I thought I could try a Decorator pattern but instead of having only one decorator, each object would have an array of Decorator who would have the same key-level minus one.

But changing fundamental of design pattern sounds not good to me (considering my level) So here I am asking for suggestions: x I've done this before but even with 2 iterations the code, while working, was really painful to write to read&change.

Thanks for help

• Robert Harvey: Writing software is about writing code, not stitching together software patterns. Data structures are chosen based on their performance characteristics. Your code doesn't have to be "ingenious;" it just has to be maintainable.
• plalx: Couldn't you use SQL for that? Seems like you are looking at finding a solution to aggregating flat metrics in various ways, based on different groupings. That's quite simple and efficient to perform with a SQL database.
• Kenny_mk: Depends on what you mean by aggregating, but maybe yes. Thing is, while i could see how SQL row data format could ease the things, i fear it is totally nullified by the syntax, the lack of tools to debug etc. I'm currenctly using PHP while it's clearly not the best choice, only because yes, i want to update/insert those results in SQL in the end and those numbers come from a database. But i was in pain to write simple things in Procedural MySQL. So for me it's kinda a no-no. Though i can be wrong