Scrap and used goods handling improvements: Consider adding specificity

[WesIngwersen]

Building off a discussion on the scrap adjustment.
The BEA does not provide any detail on what scrap or used goods consist of. Providing more resolution could be useful for material tracking and also adding intermediate processes for salvaging or preparing scrap before selling to a user.

[WesIngwersen]

There is a source for the specificity this issue asks for, and it fell out of the nowcast Step 3 materials work (cornerstone-data/bedrock#497). Posting the measurements here because they bear directly on the question, and because they let the scrap concentration in this comment be checked against a source that is not BEA.

The Economic Census names scrap streams, at NAICS-6, for 2017 and 2022

ecnmatfuel — materials consumed by kind — carries five named scrap streams on the purchase side, all flagged "excluding home scrap", which is BEA's own concept for S00401: scrap bought in, not scrap generated and reused on site.

MATFUEL code	stream	2017 $B	2022 $B
33000045	Iron and steel scrap (excluding home scrap)	18.73	36.75
33000042	Aluminum and aluminum-base alloy scrap (excluding home scrap)	5.37	9.00
33000046	Copper and copper-base alloy scrap (excluding home scrap)	2.56	3.36
33000051	Precious metal and precious metal alloy scrap	1.50	3.11
33000053	Nonferrous scrap other than aluminium, copper, precious metal	1.46	2.07
	total	29.63	54.30

For comparison, BEA's S00401 row into manufacturing is $49.1B in the 2017 detail Use table, so Census's metal-only purchases are about 60% of it.

It independently reproduces the concentration, and sharpens it

Scrap as a share of each industry's own materials bill, straight from Census:

NAICS	2017	2022	(BEA S00401 share, from discussion #5)
331314 Secondary smelting/alloying of aluminum	0.628	0.569	0.30
331492 Secondary smelting of other nonferrous	0.529	0.522	—
331511 Ferrous metal foundries	0.401	0.440	0.26 (331510)
331110 Iron and steel mills	0.328	0.440	0.30
331420 Copper rolling/drawing/extruding	0.141	0.143	0.04

Same industries, and a finer split — Census separates secondary aluminium smelting from secondary nonferrous, which BEA's detail does not. ⚠️ The shares run higher than the BEA ones because the denominator differs: this is the materials bill, BEA's is total intermediate, which includes services.

⚠️ 331110's scrap intensity rose from 0.328 to 0.440 between the two censuses. That is a large structural move in five years, and it is the kind of thing a frozen 2017 benchmark cannot see.

The asymmetry that matters for this issue

On the purchase side, Census names metal and nothing else. There is no wastepaper code, no cullet, no plastic regrind, no textile rag code anywhere in ecnmatfuel. Paper appears only as 32200000 "Paper and paperboard products" ($83.5B) and glass only as 32721000 "Glass and glass products" ($19.2B) — general material groups, not recovered-material streams.

That is consistent with the explanation already given in discussion #5: the market value of used paper and glass is low enough that it does not earn its own line in a cost survey. It does mean Census can add specificity for metal scrap and cannot, on the purchase side, add it for paper, glass, plastics or textiles.

On the output side it is the reverse — ecnpxi carries six recyclable streams including the ones MATFUEL lacks:

product code	stream	2017 $B	2022 $B
4003650000	recyclable ferrous metal scrap	34.31	52.35
4003275000	recyclable paper and paperboard	10.01	8.81
4003675000	recyclable nonferrous metal scrap	0.84	44.00
4003520000	recyclable plastics and rubber	0.18	5.04
4003100000	recyclable textiles, including rags	0.62	1.07
4005490000	other recyclable materials	0.73	2.31

⚠️ Two warnings before anyone uses that table. First, these are wholesale sales, not producer-price commodity output — 96% of the 2017 total sits in 423930 recyclable material merchant wholesalers, so in BEA terms most of it is trade margin rather than the scrap commodity itself. Second, the 2022 nonferrous figure is not credible as economics: 0.84 → 44.00 is a fifty-fold jump, almost certainly a disclosure or product-code change, and it should be diagnosed before the 2022 column is used at all.

S00402 used and secondhand goods has no source here

Nothing in ecnmatfuel, ecnbasic, ASM or AIES names a secondhand or used good. The nearest concept any of them carries is CSTRSL cost of resales, which is new goods bought to sell on untransformed and is a different thing. S00402 is $0.14B into manufacturing, so there is little at stake, but the specificity this issue asks for is not available for used goods from these sources.

What this supports

• Metal scrap can be split into five streams, at NAICS-6, on two vintages, from a source independent of BEA's own construction — enough for material tracking of the metal streams, which are also where essentially all the value is.
• The intermediate processing this issue mentions ("salvaging or preparing scrap before selling to a user") is visible on the output side as the collection and consolidation service codes (6000200000, 6000225000, 6000325000) and as the wholesaler margin, so there is something to build an intermediate process from — but at wholesale rather than producer prices.
• Paper, glass and plastics can be given specificity only on the output/wholesale side, not as purchased materials.

Extraction lives in Census_EC_MatFuel and Census_EC_PxI; the scrap codes are named in MATFUEL_SCRAP_CODES in bedrock/extract/census/Census_EC.py, where they are mapped to S00401. Reproduce the purchase-side tables with inputs_structure.py --coverage.

Numbers measured on the 2017 and 2022 Economic Census; the BEA column in the second table is quoted from discussion #5 rather than recomputed.