VH2 — Friedrich Vorwerk Group SE

Friedrich Vorwerk Group SE (VH2.DE) — deep-value analysis

Date: September 17, 2026 · Reference price: … · Market cap: … (20,000,000 shares) Exchange: XETRA · Reporting and trading currency: EUR (no conversion risk) GBL score in the tracker before this analysis: …% — BUY (Graham 5.5 / Buffett 6.5 / Lynch 9.5; total 21.5/30; F-Score 9), evaluated September 6, 2026 from source rescor_dovezi. Today’s mechanical re-scoring has already dropped to 71.7% — INTERESTING (Graham 5.0 / Buffett 5.5 / Lynch 9.0), detailed in the last chapter.

The refresh trigger: the mechanical deep_delta.py triage flagged on September 15, 2026 the thesis from August 23 as broken on the falsifier fcf_yield > 0.10. As I demonstrate arithmetically below, that falsifier wasn’t measuring the business — it was measuring the price — and it was already true the day it was written. A full refresh was nonetheless justified: the price fell from … (August 30) to … intraday (September 14) and stands today at … and in the meantime three new pieces of information were added (the Meter-Q acquisition, the parliamentary status of the grid law, management’s August share purchases).

Note on the filings file: Friedrich Vorwerk is a German SE listed only in Frankfurt/XETRA, with no US listing and no SEC CIK. The “EDGAR file” step doesn’t apply and was intentionally skipped (the data pack records this explicitly). The primary documents used here are downloaded directly from the issuer and verified line by line: the 2026 semi-annual report (published August 13, 2026, 29 pages, unaudited), the 2025 annual report (audited, Nexia GmbH), the 2025 semi-annual report and the 2023 annual report (for the five-year series). Every figure in this report has its source next to it: “RS 2026” = 2026 semi-annual report, “RA 2025” = 2025 annual report, “RA 2023”, “RS 2025”; “data pack” = rapoarte/deep/data-pack-VH2-20260917.md.

A price note, so there’s no confusion later: the tracker’s indicator snapshot for September 17 records … while the data pack regenerated at 19:01 that same day reads … I use … in all models, because it’s the most recent quote and the one the mechanical thesis check reads too. The 2.9% difference moves the Monte Carlo median from … to … — it doesn’t change the verdict, but it needs to be stated, otherwise two documents from the same day would appear to contradict each other.

Executive summary (1 page: thesis, estimated value, verdict)

The thesis, in one paragraph. Friedrich Vorwerk is an outstanding operating business — probably the best construction-engineering company in Europe by margin and capital discipline — whose stock, however, offers no usable margin of safety at … The reason isn’t that the business has deteriorated: over the last four quarters revenue was …, EBITDA … (27.3%), and net profit …, all records, and management raised its guidance on July 22, 2026 from …–180 to …–200 mil. EBITDA (RS 2026, p. 11). The reason is that the free cash flow any screener sees — …, a …% yield at today’s price — is, after normalization, roughly …, i.e. 5.5%. The … difference isn’t a guess; it reads directly from the consolidated cash flow statement and notes, as a non-repeating … working-capital release, … of cash tax paid against a … accounting expense, and … of profit recognized from project joint ventures (ARGE) of which … was collected.

The answer to the question that triggered this refresh, put up front because it’s the most important thing in this report. The broken falsifier was fcf_yield > 0.10. This isn’t a structural deterioration and isn’t even quarterly noise — it’s a construction error in the falsifier, demonstrable with two figures. The numerator didn’t move at all: TTM free cash flow was … in the August 23 thesis and is … in today’s data pack, to the decimal. What moved is the denominator: market cap fell from … (…) to … (…). The yield rose from 10.38% to …% purely because the stock lost 12.5%. The falsifier triggered on the price decline, in a thesis whose verdict was “don’t buy, it’s too expensive” — i.e. it triggered exactly in the direction that confirms the thesis, not refutes it. Moreover, it was already true the day it was written: the August 23 thesis records in its own piata_la_data field an fcf_yield of 0.102822, above the 0.10 threshold it itself defines. It was born triggered. The old thesis stated in so many words that the reported 11.4% FCF yield is inflated; a threshold that triggers when the reported yield is high can’t refute that claim — it illustrates it. Today’s new thesis replaces the falsifier with one measuring the business, not the quote.

Estimated value. Five triangulated models give a wide range, from … (DCF bear, operating margin reverting toward 17%) to … (DCF bull, margin held at 23% and 10% growth). The base case — normalized owner earnings of …, 7% growth for five years then 3.5%, discount rate …, terminal growth 2.0%, conservative net cash of … accessed at 60% — gives an intrinsic value of …, i.e. … versus price. The Monte Carlo simulation over 20,000 scenarios on the same model confirms: median …, interquartile range from … to …, undervaluation probability …%. In other words, at … it’s a coin flip slightly weighted in your favor, but nothing resembling a margin of safety.

What changed versus the August 23 analysis. I raised the owner-earnings base from 68 to … (…) — not because the price fell, but for two methodology corrections: (1) the guidance raised July 22 is now backed by six months of realized figures (H1 EBITDA …, …), so 2026’s base is no longer a guess; (2) the prior analysis applied an aggregate 24% EBITDA margin AND a haircut on ARGE profit, which double-taxed the same uncertainty. I now decompose the margin into the operating component (22.8% on TTM, normalized to 21.0%) and the ARGE component (…, normalized to 30 with a 15% haircut), and the two independent calculation paths — from cash flow and from the income statement — both converge on … I’ve, however, added a constraint the prior analysis didn’t have: λ = 0.60 on net cash, because with a cumulative payout of 18.5% over four years and no buyback whatsoever after a 36% stock decline, the implicit assumption that the … reaches the minority shareholder in full isn’t free. The two adjustments nearly cancel out; what moves the verdict is the price: the Monte Carlo median goes from … (August 23, …) to … today, at …

Verdict: NEUTRAL / WATCH, no position. Not because the business is bad — on Buffett and Lynch criteria it’s among the best in the universe — but because at …x reported profit you’re buying an operating margin at its all-time historical peak (22.8% versus 4.6% in 2023), an own-backlog that fell from … to … in 18 months, and a 54% order-book exposure to the Electricity segment, exactly the one German legislators are currently reconfiguring. The price at which the thesis would change: below … (nearly 25% margin of safety versus the base case; the Monte Carlo probability of exceeding 30% is today …%), or at any price if two quarters simultaneously show operating EBITDA margin above 23% and ARGE distributions above …/year.