9450 — Kyung Dong Navien Co., Ltd.

Deep-value analysis (REFRESH) — 009450 · Kyung Dong Navien Co., Ltd. (경동나비엔)

Analysis date: September 14, 2026 · Exchange: KOSPI, code 009450 · Yahoo symbol: 009450.KS Price: … (09/13/2026) · Market cap: … · Shares outstanding: 14,452,932 (issued 14,568,592, treasury 115,660) Reporting currency = trading currency = KRW · PRIOR GBL score in tracker: 73.3% — 22.5/30 (Graham 8.5 · Buffett 6.0 · Lynch 8.0), F-Score 4/9, scored 09/06/2026 Reference analysis: 08/23/2026 (verdict SELECTIVE ACCUMULATE) · deep_delta triage: REFRESH — 2 unverifiable falsifiers (crestere_venit_yoy with one quarter missing, pb empty in yfinance)

What I inherit and what I re-derive. Refresh regime. Inherited without re-derivation: the business description and moat verdict (narrow, slowly eroding), the leadership structure with three co-CEOs, the bear/bull scenario architecture, and four of the five valuation assumptions (oe …, r 11%, gt 2%, nd 0). Fully re-derived: the size of the U.S. tariff reimbursement, normalized operating margin, organic versus acquisition-driven growth, the FCF bridge, the earnings-quality chapter, the red flags, the valuation, and the growth assumption g1 (7% → 5%).

Why the refresh became a deep re-derivation anyway: the delta package said “new filings: none (issuer that doesn’t file with the SEC).” Correct for EDGAR — wrong as a picture of the world. The company filed with DART the semi-annual report 반기보고서 (2026.06) on 08/14/2026, nine days before the reference analysis, and that report couldn’t read it and flagged as “missing data” exactly the most important open question. I downloaded it in full (opendart.fss.or.kr, rcept_no 20260814003740, 280,685 characters), and it answers all three remaining open points. deep_delta.py doesn’t scan DART; I note this as a tool gap, not an issuer gap.

The price move (… → …, …) is context, not the reason for the redo. The new thesis isn’t built around it.

Executive summary

The thesis in brief. Kyung Dong Navien is the best Korean manufacturer of condensing boilers and tankless water heaters, with 62.1% of H1 2026 revenue generated in North America (…) and a leading position in U.S. condensing tankless water heaters built on a network of trained installers, not on patents. At … it trades at 5.53x trailing-four-quarter profit, 6.93x normalized profit and 0.986x book value — below book for the first time in the analyzed series. The reference analysis’s conclusion remains structurally valid: the displayed multiples are optically distorted, and the real discount is more modest than the raw figures suggest. What changes is the size of the distortion — larger than I estimated in August — plus the fact that the question that could have killed the thesis closed favorably.

The most important correction: the tariff reimbursement is worth about …, not 26. The August analysis estimated the Q2 2026 non-recurring effect at ~…, through the margin difference versus Q1. The semi-annual report allows three independent measurements, which converge tightly: (1) the company explicitly states, in note 25 “classification of expenses by nature,” that the reimbursement of American customs tariffs was reflected as a reduction of cost of goods sold — the first primary-source confirmation; (2) the U.S. subsidiary Navien Inc. went from a net profit of … in H1 2025 to … in H1 2026, a jump of 42.86 bn, entirely in Q2 (44.93 bn in Q2 2026 versus 2.45 bn in Q2 2025), on quarterly revenue falling (226.90 versus 232.40 bn); (3) consolidated gross margin jumped from 45.3% in Q2 2025 to 57.0% in Q2 2026, i.e. +… of gross profit on lower revenue. A fourth check, on the cost/revenue ratio: at the H1 2025 cost structure, H1 2026 expenses should have been …; they were 665.45 bn, 50.7 bn less, even though the cost base grew with a consolidated acquisition. I use …, in Q2 2026.

What remains after normalization — more than I thought. The half-year’s operating profit was … (reported margin 18.20%). Subtracting 45 bn, the half-year’s normalized margin is 12.67%, and the trailing-four-quarter one …%. The five-year average is 9.2%. On the quarterly series, Q1 2026 delivered 15.0% with no reimbursement (Navien Inc. lost 0.87 bn that quarter), and normalized Q2 gives 10.1%. The anchor I use is 11.5% — 0.3 points above August’s assumption, justified by the fact that FY2025 (9.54%) was itself distorted downward by tariffs paid and only now reimbursed. Normalized owner earnings: …, the same figure as in August, but rebuilt from different components (higher margin, higher depreciation, higher maintenance capex).

The external validation that was missing in August now exists. The reference analysis had to defend a base of … against a TTM free flow of −91.1 bn. The half-year closes the gap: CFO 122.09 bn, capex 66.21 bn, lease principal 5.47 bn → FCFE 50.41 bn over six months, i.e. … bn annualized, an 11.5% yield on market cap. Three consecutive quarters of positive free flow (+40.62, +29.64, +26.24 bn). The 96 bn figure is no longer a model construct; it’s approximately what the company produced in cash in the first half, doubled.

The question that could have killed the thesis closed benignly, with evidence. The Q1 2026 provision explosion (+19.4 bn on receivables, +13.3 bn on inventory) was “the single most important unverified element in the whole analysis.” The semi-annual report resolves it arithmetically: the doubtful-receivables provision roll-forward table shows opening balance 27,802 mil → +24,953 mil “other changes” (Comax consolidation) → −2,259 mil bad-debt expense, i.e. a REVERSAL → closing balance 50,496 mil. On inventory, note 7 literally says: “in the current period a valuation loss of 5,706 mil won was recognized, and through business combination the valuation provision increased by 10,879 mil won.” So two-thirds of the inventory provision and the entire increase of the receivables provision came through acquisition accounting, not the P&L. The reddest flag in the previous report falls. What remains in its place is an observation about the quality of the asset purchased, not about the quality of Navien’s earnings.

The new counter-fact, which keeps the verdict in check: organic growth is negative. Comax is consolidated from January 1, 2026 (deemed acquisition date, not February), so both 2026 quarters contain it. The Comax group brought … of revenue in the half-year (36.19 + 2.52). Subtracting it: half-year organic revenue grew …, not …, and Q2 2026 fell organically by 5.8%, not by 1.0%. Korea, the region hosting Comax, was flat at 228.58 bn only because of the acquisition — organic was down roughly 17%. North America grew 7.1% for the half-year, but fell 3.9% in Q2. From this comes the only assumption change: g1 falls from 7% to 5%.

Estimated value and verdict. Five models: bear DCF …, EPV Greenwald …, historical multiples …, base DCF …, bull DCF …. Median …, versus … in August — the move comes from the price (…), the higher book value (…/share) and the higher normalized profit, not from a more generous assumption: g1 fell. The two methods without a growth assumption independently converge again: 74,315 and …/share. Monte Carlo over 20,000 scenarios: median …, probability of undervaluation …%, 10th percentile at ….

Verdict: ACCUMULATE — small-medium position (2… of portfolio). One step up from “selective accumulate, 1…” in August, and only one. What justifies the upgrade: two of the three conditions the previous report set as triggers have been met — the provisions turned out to be acquisition accounting, and free flow turned positive and sustained, with 50.41 bn FCFE over six months. What prohibits more: organic revenue is falling, capex is still running at 2.4x depreciation, and the simulation’s 10th percentile is below zero. The trigger that would justify a large position remains 2027 announced capex below … at a tariff-stripped operating margin of at least 11%.