Do we use the Before or After Redefinitions tables?

[WesIngwersen]

The BEA provides the benchmark IOT Make and Use as "Before redefinitions" (BR) and "After redefinitions" (AR). CEDA uses the AT tables. USEEIO uses the BR tables.

What are the conceptual differences? What are the quantitative differences? What are the advantages and disadvantages of using AR vs. BR?

RESOLVED: Use After redefinitions

[WesIngwersen]

Background on Redefinitions

The BEA manual refers to this as the principle of homogeneity. BEA describes 3 types of co-production "redefinitions, reclassifications, and “other secondary products."
from BEA manual ch 9
"Redefinitions are secondary products that are produced using input patterns that differ substantially from that of the primary product of the industry in which they are produced."

The BEA develops the After redef tables for the purpose of reassigning the co-production from redefinition back to the original sector of interest. This shifts around industry output but commodity output stays the same.

According to this short report on the math behind the domestic requirements derivation, this is there way of applying the commodity technology assumption when the secondary products clear use different input requirements. For the remaining secondary production BEA is comfortable using the industry technology assumption.

BEA ch9 described some cases where they always do redefinitions:

• Construction activities performed by other industries are redefined to construction.
• Manufacturing activities in nonmanufacturing industries are redefined to manufacturing.
• Trade activities in nontrade industries are redefined to trade. Redefinitions
  are not made between wholesale and retail trade.
• Rental activities in nonrental industries are redefined to real estate and rental industries.
• Service activities in nonservice industries are redefined to services."

[WesIngwersen]

Over 35 industries have >10% of their output moved to other industries with the redefinitions

[WesIngwersen]

BEA authored a nice study evaluating the effects of the industry technology assumption and commodity technology assumption both before and after redefinitions. The study is from 2002 and using 1992 data so its a bit dated, but it has very useful analysis of the changed in the total requirements across these approaches.

[MoLi7]

@WesIngwersen could you link the study? Do you think we can / it's worth it to replicate it using more recent data?

[WesIngwersen]

Added link. Interesting it describes that BEA has been doing the redefinitions to move secondary commodity output where production is very much separate from the primary commodity, and then perform the domestic requirements derivation using an industry tech assumption for the rest .. since the 1970s.

We can hold off from replicating for now due to LOE but keep it as a reference if we want to do more rigorous comparisons.

[jvendries]

I added a script to quickly try to recreate the reallocations used to create the after redefinitions tables, using the method described in chapter 9 of the BEA IO Manual and the before and after tables as inputs, as well as some additional ad-hoc tests besides. A few quick conclusions from this exercise:

1. Using only the method as described in the BEA IO manual is not enough to find the exact reallocations that BEA perfromed and that result in the transformation of the Use table from before to after redefinitions (see example output). This is likely due to a "Review and adjustments of reallocations" process carried out by the BEA after the initial reallocations (see below). Without knowledge of these specific reallocations, there are additional movements between the rows and columns of the tables that are not captured in the initital reallocations procedure that we cannot exactly recreate.

BEA Manual Review and Adjustments of reallocations

2. We likely don't need to know the exact reallocations, and can probably use an estimation method like the one mentioned here (or similar) to transform any "before redefinitions-aligned dataset" to an "after redefinitions-aligned dataset". Accordingly, this shouldn't prevent us from using after redefinitions as our baseline, though it will be necessary to decide on the exact transformation method for the "before redefinitions-aligned datasets".

[WesIngwersen]

Why use After Redefinitions?

Both USEEIO and CEDA have used an industry-technology assumption to derive their A matrices. The assumption is that all industry production, regardless of the commodity, fits the same profile. This assumption leads to cleaner profiles of commodities when there is not much co-production and the industry and commodity performance is more similar. Redefinitions remove the co-production, so the more you can remove co-production, then the better.

[WesIngwersen]

Why NOT use After Redefinitions

1. Detailed gross output annual time series and chain-price indices are not published in an "after redefinitions" equivalent
   Redefinitions change industry output as described in the background. The regular industry gross output series at the detail level that USEEIO models rely on for non-benchmark years for being able to make price adjustments and related other year environmental data is only provided in what is equivalent to Before redefinitions. This implies that in order to use that gross output time series that it would need to be adjusted to be equivalent to the After redefinitions based on the reference reallocation of that output, probably from a benchmark year.

Also the Supply Use tables appear to only be offered in a Before Redefinitions equivalent if we plan to use those for a data source.

2. Following the reasoning behind USEEIO v1's justification, there might be cases where environment reports from industry include all activities occurring at a facility over a given year, including co-production. However there is no good quantitative support for this.

[bl-young]

Perhaps we are already making a similar assumption as to (1) above when we use the benchmark year commodity mix to adjust the commodity output each year.

This implies that in order to use that gross output time series that it would need to be adjusted to be equivalent to the After redefinitions based on the reference reallocation of that output

Ultimately this could take the form of a year x industry matrix of industry output before / after.

[WesIngwersen]

Recommendation

The After redefinitions will better meet our model requirements. Commodities overall have more homogenously input structure in After Redefinitions, and we will continue to have a commodity-based model. We will need to be aware of this assumption when modeling in these ways:

• All economic data for or prepared based on before redefinitions industry output will need to be adjusted to be consistent with the same after redefinitions adjustments
• Environmental data may need to be reallocated if it would be better match this structure. The Industry or Commodity distinction is possible using SectorType in flowsa and could be used in to make it clear that such data have already been reallocated from industries to commodities

[bl-young]

Still some question as to which of our environmental datasets or assumptions might be most impacted by this (e.g. consider facility level datasets assigned to NAICS or other datasets that are sector aligned)

[yyandgin]

Before near-complete clarity on the methodological differences between BR and AR, I wonder if it makes sense to do some calculations based on both methods and then compare the differences. If the differences are negligible, either method is fine.