Building Monthly Cost Indices for Changing Material Groups

Summary

The document describes a problem in constructing monthly price or cost histories for groups of materials whose membership changes over time. Across roughly 200 groups, each containing from a handful to several thousand materials, purchases are intermittent and quantities vary substantially. The desired history spans three years, with each purchased material carrying an observed price and quantity in the months it is bought.

Two attempted approaches are reported. Restricting the index to materials purchased every month reduced a much larger universe to a smaller subset, sometimes leaving a group with only one material or none and producing unrepresentative trends. Filling missing months instead risks giving an artificially carried-forward material too much influence relative to other items. The document asks for suggestions but supplies no index formula, test, or final recommendation. It therefore frames the challenges of changing composition, sparse observations, and quantity weighting without establishing which index methodology would best address them.

Key ideas

• A monthly material cost index must handle changing group membership and intermittent purchases.
• Purchase quantities differ widely, so material weights matter to the aggregate index.
• Keeping only materials purchased every month can leave groups with too few constituents to represent them.
• Filling unobserved months can distort a material’s influence on the index.
• The document identifies the problem but does not present or validate a solution.

Tags

# How to make a historical index of a group of materials in which the set of materials changes every month?


# How to make a historical index of a group of materials in which the set of materials changes every month?












The question may sound simple however for the moment it is a brainteaser to get it right, let me explain:

the exercise is to be done on +/- 200 groups of materials (matgroups)

one matgroup can contain 5 to 5000 materials. We want to see a price or cost evolution by month going back 3 years.

Every month the materials we buy have a price and a purchased quantity. However not all materials are bought every month.

so matgroup X can consist of materials A, B, C, D with each a price and quantity in january, but in february the same matgroup X can consist of materials A, D, E, F, G, H. It can take months until material B or C is purchased again. The weights (=quantities) of the materials can be very different within the matgroup.

our 1st solution is to only keep materials in the index that have been bought in every month. Out of a set of 20.000 materials, we in the end had 3000 materials left that were valid... some matgroups only had 1 material left, some 0. No surprise that this would give very strange trends, and not very representative.

2nd solution was to fill the blanks if a material isn't bought, however if this is a material with quite some volume next to it, it gets too important to be an artificially added material...

We're a bit stuck. Any experiences with something similar? Do you have enough info to think about an answer? Any suggestions/comment welcome.

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.