F2 (supplier DSO plus OCF-lags-revenue) has no discriminating power at AMAT, LRCX and KLAC: it fires less often before the 2001, 2009 and 2019 breaks than in ordinary quarters
The question
"Does a DSO / receivables-versus-operating-cash-flow divergence at semiconductor equipment makers fire ahead of the 2001, 2009 and 2019 capex troughs, and does it ALSO fire in non-trough quarters?"
Context: this tests candidate feature F2 from [[2026-09-26-financing-led-capital-cycle-phase3-precedent]]. F2's trigger is "DSO rising 2+ consecutive quarters while operating cash flow growth lags revenue growth." The parent brief proposed F2 for the financing-led capex-leg block of [[2026-05-18-memory-cycle-v1.1]]. The false-positive half is the load-bearing half.
Answer up front: No on both counts, and the false-positive count settles it. Across 3 firms and about 27 years (341 firm-quarters reconstructed from SEC filings), F2 as specified fired in 1 of 32 evaluable pre-break firm-quarters (3%). It fired in 12 of 141 "clean" expansion quarters (9%). At the episode level, it caught 1 of 3 breaks (2019, at AMAT, in the peak quarter itself). By comparison, 64% of random clean 4-quarter windows contain at least one firing across the three names. Its lift is below 1. The two variants I tested do no better in any way that survives n=3 episodes.
What we already know (from the vault)
- The parent brief built F2 on Lucent's 1998-2000 print: a $5.4B receivables build against $304M of FY2000 operating cash flow. It called F2 "the cheapest genuinely-new feature" and explicitly asked whether it also fires in non-trough quarters ([[2026-09-26-financing-led-capital-cycle-phase3-precedent]]).
- That brief also ranked "filing-derived supplier ratio first" over market-wide credit series, because only supplier ratios had a demonstrated real-time trace. This brief tests that claim for semicap specifically.
- [[2026-10-25-nvidia-vendor-financing-exposure-capex-indicator]] found a 45-to-60-day NVIDIA DSO jump that management attributed in an 8-K to "extended payment terms on large, multi-quarter agreements." That brief's verdict was that the leading-indicator reading "fails on sign before it fails on frequency."
- [[2026-07-15-capex-financing-layer-as-memory-cycle-phase-marker]] had already rejected hyperscaler financing as a phase marker. The 2026-11 tripwire-diagnostics batch in [[2026-05-18-memory-cycle-v1.1]] has been retiring or demoting rows that fail out-of-sample, and this brief is the same kind of test.
What the web says
The primary evidence is the filings themselves. This was a data reconstruction, not a literature review.
- Data reconstructed (SEC EDGAR, public filings only). AMAT (CIK 6951), LRCX (707549) and KLAC (319201):
- Pre-XBRL era (FY1998/99 through FY2009/10): parsed from about 148 full-text 10-Q/10-K/10-K405 submissions in
edgar/data/<cik>/<accession>.txt. Q4 values were derived as annual minus 9-month YTD. Annual OCF was cross-checked across consecutive 10-Ks and the MD&A text. - XBRL era (2010 through Sept 2026): from the companyfacts API (AMAT, LRCX, KLAC). Quarterly OCF was derived by differencing YTD facts.
- Validation: the two eras overlap at AMAT Q3 FY2009, and both sources give the same values: revenue $1,133.7M, AR $842.2M, OCF $193.7M.
- Definitions: DSO = quarter-end AR / quarterly revenue x 91. Each firm's fiscal quarter is mapped to the nearest calendar quarter (AMAT's January quarter becomes the prior calendar Q4).
- Pre-XBRL era (FY1998/99 through FY2009/10): parsed from about 148 full-text 10-Q/10-K/10-K405 submissions in
- AMAT's January 2001 quarter is the most Lucent-like print in the sample, and F2 misses it by 0.1 day. Q1 FY2001 (10-Q, acc. 0001136261-01-500004) had revenue of $2,731M, AR of $2,470M and operating cash flow of -$4.3M. DSO jumped from 71.9 to 82.3. But the prior quarter's DSO had dipped from 72.02 to 71.91, so the "2 consecutive rises" leg failed. The rule is brittle exactly where it should bite.
- Receivable factoring breaks the DSO leg at the largest name. AMAT's FY2001 10-K (acc. 0000006951-02-000002) states it sold $953M of receivables in FY1999, $1.5B in FY2000 and $1.2B in FY2001. It also states that "Receivable sales have the effect of increasing cash and reducing accounts receivable and days sales outstanding." The factoring peaked in the boom year, so it suppressed both legs of F2 exactly when the parent brief expects the signal.
- Revenue-recognition changes create DSO artifacts. Lam adopted SAB 101 retroactively to the start of FY2001 and booked a $122.1M cumulative-effect charge (Lam 10-Q, Dec 2001). Combining the original Q1-Q3 10-Qs with the restated FY2001 10-K produces a spurious Q4 FY2001 DSO of 132 days. KLA's 2019 DSO jump to 79 days (one of its F2 firings) coincides with the Orbotech acquisition closing, when acquired receivables arrive before a full quarter of acquired revenue.
- The accruals literature predicts this failure mode. Sloan (1996) found that high working-capital accruals predict lower subsequent returns. Fairfield et al. reframe that as a general growth effect, since growth in net operating assets predicts about as well (Lewellen, Tuck; NBER w13525). "Receivables outrun cash" is a growth signature first. That matches the finding below that F2's OCF leg fires in nearly half of all expansion quarters. Caveat: these are secondary summaries, not Sloan's original paper.
Table 1 - Fire rates by window class (firm-quarters; labeled quarters through CQ2 2025)
Windows are my dating from the three firms' revenue peaks, not an external series. 2001 runs pre-break CQ1-CQ4 2000, then downturn to CQ1 2002. 2009 runs pre-break CQ4 2006-CQ3 2007, then downturn to CQ2 2009. 2019 runs pre-break CQ3 2017-CQ2 2018, then downturn to CQ2 2019. The "other" mini-cycles are 2011-13, 2015-16 and 2022-23, each with a 4-quarter pre-window. "Clean" means every other quarter. CQ3 2025 onward is unlabeled because there are not yet 4 quarters of hindsight.
| Variant | Pre-break (3 episodes) | Downturn (coincident) | Clean (false-positive pool) | Other mini-cycle pre / down | Fisher p, pre vs clean |
|---|---|---|---|---|---|
| F2 as specified (DSO up 2 qtrs AND TTM OCF growth < TTM revenue growth) | 1/32 = 3% | 4/48 = 8% | 12/141 = 9% | 4/36, 1/45 | 0.47 |
| F2, quarterly YoY OCF leg | 4/36 = 11% | 6/48 = 13% | 14/147 = 10% | 5/36, 1/45 | 0.76 |
| "Lucent shape" (TTM AR build > TTM OCF, no DSO leg) | 4/36 = 11% (all 2000) | 0/48 | 8/149 = 5% | 0/36, 0/45 | 0.25 |
Component rates in clean quarters, F2 TTM: the DSO-up-2-quarters leg alone fires 22/141 (16%), and the OCF-lags-revenue leg alone fires 66/141 (47%). In pre-break quarters the rates are 8/32 (25%) and 11/32 (34%). The OCF leg is less frequent before breaks than in ordinary expansions.
Table 2 - DSO (days) by calendar quarter around each designated break, with F2 firings
T = F2 as specified fired. q = only the quarterly-YoY variant fired. n/e = not evaluable (insufficient trailing history).
| CQ | AMAT | KLAC | LRCX | Firing |
|---|---|---|---|---|
| 2000Q1 | 69 | 89 | 86 | n/e for all three |
| 2000Q2 | 72 | 91 | 79 | none (AMAT n/e) |
| 2000Q3 | 72 | 93 | 82 | none |
| 2000Q4 | 82 (OCF -$4.3M) | 96 | 78 | KLAC q |
| 2001Q1-Q4 | 77 / 90 / 52 / 71 | 93 / 78 / 67 / 82 | 78 / 132* / 54 / 36 | none |
| 2002Q1 (trough) | 76 | 70 | 70 | AMAT T |
| 2006Q4-2007Q3 | 82 / 76 / 80 / 79 | 63 / 63 / 72 / 81 | 66 / 65 / 55 / 45 | KLAC q (2007Q2) |
| 2007Q4-2008Q4 | 88 / 73 / 78 / 75 / 87 | 83 / 87 / 76 / 63 / 76 | 66 / 89 / 66 / 64 / 93 | KLAC q (07Q4, 08Q1) |
| 2009Q1-Q2 (trough) | 82 / 68 | 71 / 68 | 103 / 106 | LRCX T, T |
| 2017Q3-2018Q2 | 54 / 47 / 53 / 63 | 63 / 69 / 59 / 55 | 56 / 80 / 66 / 63 | AMAT T (2018Q2, the peak quarter); KLAC q (2017Q4) |
| 2018Q3-2019Q2 | 56 / 59 / 58 / 61 | 50 / 53 / 79** / 72 | 72 / 67 / 57 / 56 | KLAC T (2019Q1) |
* SAB 101 restatement artifact. ** Orbotech acquisition close.
Sources: 2000-2002 from AMAT acc. 0000006951-00-000002, -00-500005, -01-500003, 0001136261-01-500004, 0000006951-01-500005, -01-500008, -02-000002, -02-000010; KLAC acc. 0000891618-00-002841, 0001095811-00-003641, 0000891618-00-005067, -01-000175, -01-500693, 0001095811-01-505048, 0000891618-01-502223, -02-000664, -02-002335; LRCX acc. 0000891618-00-002570, 0000707549-00-500008, -00-500013, -01-500006, -01-500009, -01-500016, -01-500019, -02-000003, -02-000007. 2006-2009 from the corresponding 10-Q/10-Ks (e.g. LRCX 0001193125-09-105515 and 0001193125-09-182356; AMAT 0000950134-09-004336). 2017-2019 from XBRL companyfacts.
Table 3 - Where F2 (as specified) fires outside the designated breaks (the false positives)
| Class | Firings |
|---|---|
| Clean (12) | KLAC 2004Q1, 2006Q1; AMAT 2005Q3, 2019Q3, 2024Q2, 2024Q3, 2024Q4, 2025Q1; LRCX 2013Q4, 2019Q4, 2020Q1, 2024Q4 |
| Other mini-cycle (5) | KLAC 2014Q3, 2014Q4, 2023Q1; AMAT 2015Q2; LRCX 2022Q2 |
| Unlabeled, CQ3 2025+ (2) | KLAC 2026Q2, LRCX 2026Q2 (both cleared in CQ3 2026) |
AMAT fired four quarters running in 2024-25 with no capex break following. That is the cleanest single false-positive run in the sample.
Coverage: what I could and could not reconstruct
- Reconstructed: AMAT from FY1999 Q1 (Jan 1999), and KLAC and LRCX from the Sept 1998 quarter, through the latest filed quarter (AMAT Jul 2026; KLAC and LRCX Sept 2026). That totals 341 firm-quarters. F2 TTM needs 8 trailing quarters, so evaluation starts around mid-2000.
- Not evaluable: AMAT's entire 2000 pre-break window under the TTM rule. The FY1998 10-K405 is a 1998 text filing whose cash-flow table did not parse, and I did not hand-extract it. Hand-extracting it would make AMAT 2000Q3-Q4 evaluable. Those were the quarters that mattered, but the DSO leg failed there anyway.
- Basis mixing, flagged rather than fixed:
- AMAT Q4 FY1999 is on Etec-pooling restated figures, while Q1-Q3 FY1999 are original.
- LRCX FY2001 mixes pre- and post-SAB 101 figures (see Table 2 note).
- KLAC Q4 FY2005 revenue is derived from a $M-rounded annual figure (about ±$0.5M).
- KLAC FY2004-06 straddles the stock-option restatement.
- XBRL "latest filed" values can be recast (e.g. ASC 606).
- Not done: Novellus, Kulicke & Soffa, ASML and TEL. ASML and TEL file 20-F/6-K or report in Japan, so they have no 10-Q quarterly series. I also did not hand-check every parsed quarter. Extraction was regex-based with about 40 manual Q4 and AR overrides, and the totals tie to the 10-K annual figures checked.
Convergences and contradictions
- Contradiction with the parent brief: it expected F2 to have a "demonstrated real-time trace" by analogy to Lucent. At semicap it does not. The Lucent print was a vendor-financing story at a telecom OEM that was effectively lending to CLECs. Semicap customers in 2000, 2007 and 2018 were mostly investment-grade fabs paying on terms. The receivable channel never carried the financing stress, and the one firm big enough to show it (AMAT) factored it away.
- Convergence with the NVIDIA brief: both find that a DSO move is uninterpretable without the company's own explanation of terms. NVIDIA's 15-day jump came with a disclosed cause ("extended payment terms"). Semicap DSO moves of 10-30 days happen every few quarters with no financing story at all, from shipment linearity, factoring, SAB 101 and M&A. Disclosed terms language is the signal; the ratio is noise.
- Convergence with the accruals literature: "AR grows faster than cash" is mostly a growth or expansion property. That is why the OCF-lags-revenue leg fires in 47% of clean quarters, and why the Lucent-shape variant fires in Lam's 1999 and 2009-10 recoveries as readily as in the 2000 boom.
Synthesis for RDCO
F2 should not go on the memory-cycle tripwire list, not even as a co-indicator. The parent brief set the bar ("a feature that fires everywhere is not a tripwire"), and F2 fails it in the strongest form: it fires less often in the four quarters before a capex break (3%) than in ordinary expansion quarters (9%). At the episode level, a random clean 4-quarter window has about a two-in-three chance of showing at least one firing across AMAT, LRCX and KLAC, while the designated pre-break windows showed one in three. The one pre-break hit (AMAT, April 2018) landed in the peak quarter itself, so it provided no lead. The trough-adjacent firings (Lam 2009Q1-Q2, AMAT 2002Q1, KLAC 2019Q1) are coincident or lagging: DSO rises because revenue has already collapsed, or because an acquisition closed. None of the p-values is anywhere near significance, so the honest reading is "no detectable discriminating power," not "anti-signal." At n=3 episodes, nothing stronger can be claimed.
The mechanism explains the failure, and the explanation matters more than the counts. The OCF-lags-revenue leg is a growth-phase working-capital property: in boom quarters inventory and receivables absorb cash, so it fires in nearly half of all expansions. The DSO leg is dominated by things that have nothing to do with customer financing: quarter-end shipment timing, receivable sales (AMAT moved $1.5B off balance sheet in FY2000, explicitly to cut DSO), revenue-recognition regime changes (SAB 101 in 2001, ASC 606 in 2018), and acquisitions. The financing stress the parent brief worried about would have to pass through the supplier's receivables to show up in F2. At semicap it historically did not, because the customers were not the ones being financed. The Lucent-shape variant (receivable build exceeds trailing OCF) did light up in 2000 at KLAC and Lam. It also lit up in Lam's 1999 and 2009-10 recoveries, never at AMAT, and not before 2009 or 2019, so it is a one-episode boom detector.
What to do with the slot instead. If the financing-led capex-leg block wants a supplier-receivables feature, it should be disclosure-keyed, not ratio-keyed. The candidates are company-stated "extended payment terms" language (the NVIDIA case), disclosed receivable-sale or factoring volumes, and ≥10%-customer AR concentration. All three are discrete, explained events with a cause attached, which a ratio is not. That is the same lesson [[2026-10-25-nvidia-vendor-financing-exposure-capex-indicator]] reached from the other direction. F1 (customer financing as a share of revenue) and F5 (refinancing-calendar stress) are untouched by this result.
Current reading, for the record and not for posture. F2 fired at KLAC and LRCX in CQ2 2026 and cleared in CQ3 2026. AMAT's DSO has risen two quarters running (65 to 73 to 77 days through July 2026), and the quarterly-YoY variant fires there, with the October-quarter 10-K not yet filed. Against a 9-10% clean-quarter base rate per firm, a three-name flicker is roughly what noise looks like. It should not move the memory-cycle thesis in either direction. This is not a position recommendation.
Why this is in the vault
This closes the "Does F2 have discriminating power?" follow-up in [[2026-09-26-financing-led-capital-cycle-phase3-precedent]]. It decides that F2 is rejected (not demoted) from the proposed F1-F5 financing block before that block is wired into the [[2026-05-18-memory-cycle-v1.1]] risk register, and it redirects the supplier-receivables slot to disclosure-keyed features.
Open follow-ups
- Disclosure-keyed replacement: does the first appearance of "extended payment terms" or a step-up in disclosed receivable sales in AMAT/LRCX/KLAC/NVDA 10-Q text lead capex breaks, and how often does it appear in non-break quarters? This is the same false-positive frame applied to language instead of ratios.
- Customer-side mirror: do memory makers' days-payable (Micron DPO, from 10-Qs) stretch ahead of 2001, 2009, 2019 and 2023? If financing stress lives anywhere in filings, it should show on the buyer's balance sheet, not the vendor's.
- F3 under the same test: does AMAT/LRCX backlog or deferred revenue (and ASML bookings from 6-K) roll over ahead of the three breaks, and what is its clean-quarter fire rate? It is the next-cheapest candidate in the F block.
- Close the AMAT 2000 gap: hand-extract the FY1998 10-K405 cash-flow statement so the AMAT 2000 pre-break window becomes evaluable under the TTM rule. This is low priority because the DSO leg already fails there.
Related
- [[2026-09-26-financing-led-capital-cycle-phase3-precedent]] - parent; defines F2 and asked this question
- [[2026-05-18-memory-cycle-v1.1]] - the live thesis whose tripwire and risk register this feeds
- [[2026-10-25-nvidia-vendor-financing-exposure-capex-indicator]] - NVIDIA DSO jump with disclosed terms change; convergent conclusion
- [[2026-07-15-capex-financing-layer-as-memory-cycle-phase-marker]] - grandparent; registered financing-led Phase 3 as an unevidenced risk
- [[2026-09-07-memory-supplier-inventory-days-dio-anchor]] - the sibling supplier-ratio series (DIO), worth subjecting to this same false-positive test
- [[2026-09-29-phase3-memory-cycle-rotation-basket]] - earlier finding in the same spirit: rules that look good from hindsight dates fail from their own firing dates
Sources
- Vault:
06-reference/research/2026-09-26-financing-led-capital-cycle-phase3-precedent.md;01-projects/investing/theses/2026-05-18-memory-cycle-v1.1.md;06-reference/research/2026-10-25-nvidia-vendor-financing-exposure-capex-indicator.md;06-reference/research/2026-07-15-capex-financing-layer-as-memory-cycle-phase-marker.md;06-reference/research/2026-09-07-memory-supplier-inventory-days-dio-anchor.md;06-reference/research/2026-09-29-phase3-memory-cycle-rotation-basket.md - SEC XBRL companyfacts: https://data.sec.gov/api/xbrl/companyfacts/CIK0000006951.json, https://data.sec.gov/api/xbrl/companyfacts/CIK0000707549.json, https://data.sec.gov/api/xbrl/companyfacts/CIK0000319201.json
- SEC submissions indexes: https://data.sec.gov/submissions/CIK0000006951.json (and -submissions-001.json), likewise for 707549 and 319201
- SEC full-text filings, 1998-2010, at
https://www.sec.gov/Archives/edgar/data/<cik>/<accession-no-dashes>/<accession>.txt. Key accessions are cited inline above, notably AMAT 0001136261-01-500004 (Q1 FY01) and 0000006951-02-000002 (FY01 10-K, receivable-sale disclosure); LRCX 0000707549-02-000003 (SAB 101) and 0001193125-09-182356 (FY09 10-K); KLAC 0000891618-01-000175 (Dec 2000). - Lam SAB 101 10-Q: https://sec.gov/Archives/edgar/data/707549/000070754902000003/0000707549-02-000003.txt
- Accruals anomaly (secondary summaries): https://faculty.tuck.dartmouth.edu/images/uploads/faculty/jonathan-lewellen/Accruals.pdf ; https://www.nber.org/papers/w13525.pdf
- Computation: scratch scripts and intermediate data are in /tmp/dso-edgar (ephemeral, not vaulted). Fisher exact tests computed in-script.