CAAP — Corporacion America Airports S.A.

CAAP — Corporación América Airports S.A. · deep-value analysis (REFRESH)

September 19, 2026 · reference price … (NYSE) · 163.178 mn shares (165,219,146 issued − 2,040,816 treasury, Item 7, 20-F FY2025) · market cap 3,854 mn USD · reporting and trading currency: USD

REFRESH regime. The reference analysis is rapoarte/deep/2026-08-28-deep-CAAP.md (08/28/2026, verdict AVOID at …). The mechanical triage the quarter-over-quarter table requested a re-run for a single new filing: 6-K of 09/16/2026, accession 0001104659-26-108115 — the August 2026 traffic release. Delta package: rapoarte/deep/delta-CAAP-20260918.md.

What I explicitly inherit, without re-deriving (no new financial filing from 08/28/2026 to today — the latest statements remain the condensed interim ones as of 06/30/2026 from the 6-K of 08/18/2026, accession 0001104659-26-098141/-098142, and the FY2025 20-F of 03/17/2026): the business architecture and the concession table; the 2021-2026 capital allocation history; the five O’Glove checks on receivables, inventory, debt, discretionary expenses and accruals; the Thorndike grid; the eight accounting flags; the FCF and owner earnings bridge of USD … mn; the g1/r/gt/nd assumptions; the intrinsic values of models 1 (bear), 2 (base) and 5 (bull).

What I re-derived, because a fact moved: (1) July and August 2026 traffic, broken down by country and by domestic/international/transit segment — new evidence; (2) the actual USD 150 mn dividend payment on 09/10/2026, confirmed from the 2Q26 results release, with pro-forma balance-sheet effect; (3) model 3 (EPV Greenwald) and model 4 (historical multiples), both recalculated on post-dividend net debt; (4) the Monte Carlo simulation, at the new price; (5) the GBL scorecard, against the tracker score updated 09/06/2026.

A process correction, before anything else. The Monte Carlo block in the delta package is unusable: it reports a median intrinsic value of USD …/share and a median MOS of …. The cause is a wrong unit — the mechanical re-run passed the Monte Carlo DCF model the argument --oe … (thousand USD, as owner earnings appears in the thesis bridge), but the script expects millions, so it modeled a company with USD … billion of annual cash flow. The simulation in this report is re-run correctly, with --oe …, and the reference file is the Monte Carlo simulation. No figure from the delta package’s MC block is used here.

Executive summary

The thesis remains the one from August, unchanged in substance. CAAP operates 52 concessioned airports in six countries and looks, on any screener, like a deep-value opportunity: trailing P/E 13.5x, company-reported net leverage 0.5x EBITDA, “FCF yield” …%, and only 7 analysts covering it, with an average price target of …. I continue to argue that this profile is an illusion produced by two things automated indicators don’t see, and that at … the stock is fairly valued to slightly expensive.

First: IFRIC 12 accounting. Investment in concession assets — USD 232.0 mn over the last twelve months — is reported inside operating cash flow, as a separate line, while the “Capital Expenditure” line in investing is only USD 20.0 mn. Whoever applies the CFO − capex reflex gets 454.5 mn and an …% yield, but that is a flow before interest (75.6 mn), before lease principal (5.8 mn), and before the subsidiaries’ minority co-owners’ share (… mn). The issuer is IFRS and takes all three out of the operating section.

Second: the suspended Brazilian concession fee. ANAC suspended, in December 2025, the enforceability of a fixed contribution of R…lion at Brasília. The amount was not forgiven — it was deferred, appears in the cash flow statement as a positive add-back to the “Unpaid concession fees” line, and accumulates on the balance sheet under “Concession fee payable”: 748.5 mn (Dec. 2024) → 918.8 (Dec. 2025) → 1,012.8 mn USD (06/30/2026). Roughly USD 57.0 mn a year of CAAP’s “flow” is an unpaid bill. The net debt the company puts in the headline of its release — 381.0 mn, 0.5x leverage — ignores a contractual obligation exceeding one billion dollars, with 1,449.1 mn undiscounted due beyond five years.

The FCF bridge, TTM July 2025 – June 2026, unchanged: CFO 474.6 mn → simple FCF 454.5 → FCFE 338.5 → normalized FCFE … mn USD, i.e. 7.3% on today’s market cap. The gap to the market-published figure remains 61.5%.

What the new filing added. The 09/16/2026 release shows August 2026 traffic down 0.9% year-over-year (7.871 vs 7.940 million passengers). Combined with July (…), the third quarter is running at … year-over-year over the first two months — 15.930 vs 15.890 million — versus … in the first half and … in the second quarter. The breakdown matters more than the aggregate: domestic traffic fell 7.6% in July-August (versus … in H1), international grew 5.8% (versus … in H1), and transit — the passengers who bring the least unit revenue — accelerated to …. Argentina, 55% of EBITDA, went from a flat half-year (…) to … in July-August, with aircraft movements at … in July and … in August. This is not an inflection that changes the thesis; it is a confirmation of it, on exactly the channel I flagged as the signal to watch.

The second new fact, more important for the balance sheet: the dividend was paid. The board approved it on 08/18/2026 (150 mn USD, ~…/share), the record date was 08/31/2026, and the distribution took place on September 10, 2026. The 150 mn actually left the balance sheet nine days ago. An arithmetic consequence I must state explicitly: the … price is ex-dividend, so the … move reported by the delta package is not a repricing — adjusted for the dividend, total return since the reference analysis is … (… + 0.908 = 24.53 vs …). The market did not react to anything; it collected.

Estimated value. The five models, repriced at … and with models 3 and 4 recalculated on the post-dividend balance sheet, give a range of 13.27 – …/share, median …: EPV Greenwald with concession debt included …, DCF FCFE bear …, DCF FCFE base …, own 5-year historical multiples …, DCF FCFE bull (Brazilian renegotiation succeeds) …. Monte Carlo on 20,000 scenarios: median …, probability of undervaluation …%, probability of a margin above 30% …%.

Note on why the base MOS improved from … to …: not from fundamentals. The DCF models run with nd = 0, meaning I attribute no value to the cash on the balance sheet (FCFE is already post-interest, so subtracting the debt too would double-count). When the company takes out 150 mn of cash my model was valuing at zero, the price falls by … and the intrinsic value stays the same — so the MOS mechanically improves by 3.9 percentage points without anything having changed. Of the 2.1 points of improvement observed, all of it comes from here, and the rest is price movement in the opposite direction. Anyone reading only the final figure sees an approach to value that doesn’t exist.

Verdict: AVOID at …; WATCH below …. Unchanged. Not because the business is bad — the concessions are real, Armenia grows 14% a month, Italy and Uruguay deliver — but because today’s price simultaneously assumes the Brazilian bill will never be paid, that Argentina stops deteriorating, and that a committed capex program of over a billion dollars over five years funds itself without touching the flow to shareholders. The event that could flip the thesis in a single step — the tender for 100% of Inframerica (Brasília), expected by December 2026 — is now about three months away, a month closer than at the reference analysis.