LPG — Dorian LPG Ltd.

Dorian LPG Ltd. (NYSE: LPG) — deep-value analysis, REFRESH

September 12, 2026 · reference price … (data pack 09/12/2026, yfinance) · market cap … · 42,782,681 shares · REFRESH REGIME relative to the analysis of 08/17/2026

What I inherit and what I re-derive. The reference thesis is the one in rapoarte/deep/2026-08-17-deep-LPG.md (verdict DO NOT BUY at … reference value …–38). The mechanical triage deep_delta.py triggered two falsifiers — pe (7.29 < threshold 8) and roe (26.0% > threshold 25%) — and flagged a single new filing, the 8-K from 09/04/2026.

I inherit unchanged, because nothing in the new filings touches it: the business and Helios pool description, the moat structure, the five-year capital allocation history, the O’Glove analysis of receivables/inventory/discretionary expenses across FY2022–Q1 FY2027, the list of accounting red flags 1–11, and the three pre-mortem scenarios. All the accounting figures from the 10-K FY2022–FY2026 filings and the 10-Q Q1 FY2027 are taken as verified, not re-read.

I re-derive, because the facts have moved: (1) the mid-cycle freight rate level, following the forward chartering press release; (2) the owner earnings bridge and, through it, all five valuation models; (3) the pro-forma NAV at 09/30/2026; (4) the management chapter, where the order for three new ships and the stock sales by three executives materially change the reading; (5) the Monte Carlo simulation, which had to be redone anyway (see the note below).

New sources verified directly in this session: 8-K filed 09/04/2026 (accession 0001596993-26-000041, items 1.01, 2.03, 7.01, 8.01) and its attached Exhibit 99.1; Forms 4 filed between 06/30/2026 and 09/11/2026 (accessions 0000919574-26-004215, -005033, -006221, -006248, -006249, -006250); Note 17 “Subsequent Events,” Note 4 “Helios Pool,” Note 5 “Vessels Held for Sale,” Note 8 “Long-term Debt,” and the depreciation sensitivity table in the 10-Q Q1 FY2027.

Mandatory technical note on the inherited simulation. The simulation from the delta package (mc-LPG-20260911.json) is corrupted: it was run with --oe … instead of --oe 103, i.e. owner earnings in absolute dollars fed into a script that expects millions. The result — median intrinsic value “…/share” and median MOS “ …” — is arithmetically impossible and would have passed silently into the report. I reran it, with the correct units and updated assumptions: mc-LPG-20260912.json. The DEEPMC marker at the end of the report is from the new file.

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

What changed in 26 days. On September 4, 2026, Dorian published three things at once, in a single press release: it ordered three 90,000 cbm dual-fuel Panamax VLGCs from Hanwha Ocean for ~…, with delivery in June, September, and December 2030; it refinanced four facilities into one of … over seven years at SOFR + 140 basis points; and it disclosed it had fixed 99% of the calendar days of the quarter ending September 30, 2026 at a rate above …/day, excluding demurrage. The third item is the one that matters for valuation: the record TCE in the company’s history, the …/day of Q1 FY2027, hasn’t just failed to normalize — it climbed more than 16% further, and this time it’s contracted, not hoped for.

Why the falsifiers broke, and why the break doesn’t overturn the thesis. pe < 8 and roe > 0.25 are both poorly calibrated falsifiers for a cyclical at its peak. A trailing P/E of 7.3x and a reported ROE of 28.3% are exactly what you see when peak earnings sit in the numerator: as profit rises, the P/E falls even if the price rises too — and it did: the price gained … while the multiple stayed below 8. Their breaking measures the persistence of the peak, not an error in the thesis. The falsifier that actually matters, revenue_growth_yoy > 0.8 over four consecutive quarters, is still unresolved: I have one quarter out of four (…), the second is practically contracted, two more are needed. What I’m changing is not the conclusion, but a number: the mid-cycle level. I move it from … to …/day — an OPINION move, disclosed as such, justified by five months of war-driven rates that no longer look like a spike, by a fleet cleared of its three oldest units, and by the structural rerouting of LPG flows from the Persian Gulf to the Gulf of Mexico.

What I found that wasn’t in the delta package. Between September 8 and 10, 2026, three of four executives sold stock on the open market: Chief Commercial Officer Tim Truels Hansen 50,000 shares at … (the third tranche of a 95,000-share series started at … on June 30), CFO Theodore Young 26,000 shares over three days at …–… COO Alexander Hadjipateras 5,000 at … None under a 10b5-1 plan — the aff10b5One field is “0” on all six Forms 4. The sales came in the first three trading days after the press release announcing the record quarter. The CEO sold nothing.

Estimated value. Five triangulated models give a range of … – …/share, with a median MOS of …. The pro-forma NAV at 09/30/2026, with vessels valued at the prices the company actually realized in May–July 2026, is … (it was … in August; the difference is the cash generated in the quarter). The DCF on mid-cycle FCFE of …, plus the present value of the contracted and probable cyclical surplus (…/share), gives …. Historical multiples at 6.0x cyclical EBITDA give … The Greenwald EPV, the honest floor, gives … Weighted 40% NAV / 35% multiples / 25% flow: reference value ~….

Verdict: DO NOT BUY at … Entry threshold moved up, from … to … The reference value has risen 11% (from …–38 to ~…) — rightly so, the facts improved. But the price rose 16.5% over the same 26 days, so the gap has widened, not closed: from −20…… to …. The consensus of the five analysts covering the stock has a target of … i.e. 0.4% BELOW today’s price, and that target predates the September 4 press release. At … you’re buying a quarter that has already been announced, at 8.8x mid-cycle EBITDA and 1.20x the ships’ sale value in the best VLGC market on record. The company remains above the sector average; the price remains one to refuse.