Scrap Market Review — September 2026
Market analysis from TDC Ventures. Not trading advice.
The month in one line
Turkey’s scrap rally, which our August review expected to hold inside a narrow band, broke out instead — handing us our second-worst forecast miss on record before a mid-month fix to how we score a silent publisher helped us land four straight hits into the close.
Our track record this month
We publish a scored record of every forecast we make. The full, unedited version — ranges, actuals, verdicts, errors — is at Forecast Track Record.
September was a genuinely hard month for us, and it should read that way.
- Resolved this month: eleven price-basis sub-components across five forecast calls, plus one spread call. Five of the eleven landed in range — a 45.5% hit rate, well below our all-time published-benchmark rate.
- September mean absolute error: $6.89/t, against $5.67/t across the whole book to date.
- One publisher leg closed the month’s first-ever “unscoreable” verdict — not a hit, not a miss, excluded on rules we wrote down before we knew the outcome. More on that below.
- A new forecast, written under a rule change adopted mid-month, resolved cleanly on its first outing.
The book as a whole now stands at 14 of 23 in range on our headline basis (60.9%), mean absolute error $5.67/t, median $4.61/t, direction accuracy 47.8%. Twenty forecasts written: seventeen have resolved, three remain open.
Calls that landed
Four straight hits into month-end. Our 25 September calls resolved cleanly on every leg: Turkey shredded printed $421.00 on Kallanish (+$5.00 against our $416 central) and $417.19 on Fastmarkets (+$1.19), both inside range; Pakistan shred 211 printed $417.72 on the Fastmarkets index (−$2.28), also inside range. It’s the first Turkey resolution all month with no leg missing high.
We flagged the trap in August, and it caught the market exactly as described. Our August review’s forward call on Pakistan said our then-open $408–418 range already looked behind the tape, with the 28 August print already sitting above it at $419.96 — and told readers to expect the call scored as written, not moved after the fact. It resolved on 4 September at $420.16, a $7.66 miss, precisely the outcome the August caveat anticipated.
Where we missed — and what it taught us
Our second-worst forecast ever printed this month, and it repeats a mistake we’d already named once.
Our 6 September Turkey shredded forecast, written into a market that had already jumped twice the week before, read the move as a level to consolidate around. A Turkish mill source we cited capped “the realistic further move” at $2–3/t, and we built our central close to that ceiling. The market did not consolidate: by the 11 September target, Kallanish printed $413.50 — a $17.50 miss and Fastmarkets $404.84, $8.84 above range. The actual move was three to six times the ceiling we’d been given.
This was not new. The exact same failure — naming the right upside driver and sizing it as the smaller alternative to a downside case that never fired — had already cost us a $7.66 miss on Pakistan one week earlier. Naming a risk is not the same as weighting it. Standing correction: when a forecast’s own sourcing names a specific, near-term upside catalyst, size it toward the top of the range we’re given, not the bottom — especially once that catalyst has already fired once in the current cycle.
The month’s other structural miss was in our own scoring mechanics, and we fixed it before it could repeat. A Kallanish price line declared as a secondary leg on our 11 September Pakistan forecast went quiet after a print five weeks earlier and never printed again. Scoring it as a miss against a five-week-old number would have manufactured a result out of silence, so it closed unscoreable instead — no error, excluded from every ratio, on a rule written down before we knew the leg would go dark. We then adopted a standing rule: a publisher leg with no print for a month or more at the time we write a new forecast is excluded from scoring in advance. Our 25 September Pakistan forecast, the first written under that rule, hit cleanly.
The month’s market arc
Turkey’s rally was real, sustained, and confirmed by more than one publisher. Shredded scrap into Turkey climbed from around $392 heading into the month to $421 by the last week — not a single-print spike but a four-week grind, with Kallanish and Fastmarkets converging to within pennies of each other by month-end. A measurement wrinkle mid-month — one publisher split its blended Turkish price into separate US-origin and EU-origin lines — looked to some readers like a fresh $6–8 regional premium opening overnight. It wasn’t: the split coincided with a genuine price move a second publisher’s own index confirmed independently, and the two new lines have since sat frozen relative to each other for multiple sessions — the signature of a reporting change, not a live spread.
Pakistan’s official index finally reprinted after a stretch of silence, and the “boring” number won. Reported supplier asking prices spiked as high as $440–445/MT mid-month and were never repeated by an actual sale. When the delayed index finally landed, it came in essentially flat, close to the physical anchor the desk had been trusting all along — not the exciting offer chatter.
Non-ferrous had its own, separate story. Zinc posted its sharpest single-day move of the quarter late in the month, then gave almost all of it back over the following three sessions. Nickel moved more quietly but more persistently, sliding through the month’s final week with no single dramatic print but a steady multi-session decline — neither move had a clean, stated driver in the public record, a reminder that non-ferrous references can move faster than a quote sheet gets updated.
Freight risk stayed elevated all month without ever resolving. Vessel traffic through the Strait of Hormuz remained well below normal throughout September, Iran and the US continued talking without a reopening agreement, and no carrier published a new war-risk surcharge tied to the tension. Late in the month Saudi Arabia resumed oil loadings from Yanbu after a pipeline restart and crude eased — a genuine improvement in regional supply capacity not yet confirmed as lower container freight.
Structural signals we’re tracking
The EU’s non-OECD waste-export regime is the largest medium-term sourcing risk on our board, and it has a hard calendar now. The public consultation on the draft list of countries authorised to receive EU-origin non-hazardous waste closes 16 October 2026; the first approved-country list is due 21 November 2026; the restriction itself takes effect 21 May 2027. Pakistan is not on the current draft list. Not law yet — but every EU-origin supplier relationship priced today should account for a possible gate in under two years.
China’s refined-copper supply is tightening, and it hasn’t fully reached scrap pricing yet. Reporting late in the month pointed to refined-copper output growth slowing sharply in 2026 on tight concentrate and scrap feedstock, plus scheduled smelter maintenance. Copper-bearing scrap, cable, and brass sellers have a real argument for holding formula pricing rather than accepting fixed-price offers while this plays out.
Dubai’s temporary scrap-export restriction is a near-term forcing function, not a footnote. It runs through 8 October 2026. Buyers still leaning on UAE-origin material without a written exemption in hand are the ones most exposed if it extends.
What we’re watching next month
Three specific calls for October. These are what next month’s review gets scored against.
- Does Turkey’s rally extend into a fifth consecutive week, or does it top out? Our open forecast targets 2 October at $412–428 (central $420) — publishing it as written, not adjusting it after the fact.
- Does Pakistan’s shredded index hold near its physical anchor, or finally confirm the higher offer chatter? Our open forecast targets 2 October at $413–425 (central $418). A print at or above $425 would validate the offer side for the first time all quarter.
- Does the Kallanish Turkey US-origin/EU-origin split ever move independently, or does it stay locked in place? If the two lines start diverging on their own, it stops being a reporting artefact and becomes a real regional premium worth pricing.
How we forecast
We anchor to published benchmarks and observed transaction levels, cross-check every call against leading indicators — freight, iron ore, finished-steel margin — and declare the resolution basis in writing before the target print exists, so a call cannot be quietly re-scored against whichever number happens to flatter it. Every forecast is then scored against the first print on or after its target date, hit or miss, and published. The misses stay up.