64850 — FnGuide Inc.

Deep-value reanalysis: FnGuide Inc. (064850, KOSDAQ)

Date: 2026-09-18 · Price: … · Market cap: … (≈…) · Reporting currency = price currency = KRW

Trigger: rerun requested by the mechanical escalation of 2026-09-17, with a single falsifier reported as triggered — marja_operationala. Documents: research-064850-20260918.md, data-pack-064850-20260918.md, mc-064850-20260918.json, the prior thesis teza-064850-20260917.json, the one before that teza-064850-20260824.json.

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

The first thing to say, because it changes the nature of the whole task: the falsifier did not trigger on the current thesis. The check run today, python teza_deep.py --check 064850, returns 0 triggered, 0 unverifiable, 7 hold, and marja_operationala reads [0.5108; 0.5023] against a threshold > 0.55. The 0.55 threshold was written by the 2026-09-17 analysis. What triggered on 09-17 was the falsifier from the previous thesis (teza-064850-20260824.json), which had the threshold at > 0.50 for two consecutive quarters — and which had already been recalibrated by the 09-17 run, precisely in response to that trigger. Today’s escalation is a rerun on a thesis already superseded: the queue pulled the same event twice. I flag this explicitly instead of simulating a discovery, because it is the only honest conclusion about the trigger — and because, if it isn’t stated, it will repeat tomorrow.

The second thing, and it matters more than the first: the falsifier was poorly built, regardless of the threshold. The load-bearing claim said “the 52.7% operating margin from Q2 2026 is a cycle peak, not a sustainable level.” The attached falsifier — “the margin exceeds 0.50 for two quarters running” — triggers on the thesis’s own stated fact: the margin was already above 50% in both H1 2026 quarters at the moment the thesis was written. A statement of the form “level X is a peak” cannot be falsified by the observation “level X persists for two quarters,” when the peak already has two quarters of age. The falsifier was guaranteed to fire on the first check. Worse: the metric it relies on is, for this issuer, nearly unusable — the marja_operationala in the code reads yfinance’s EBIT (pretax income + interest expense), which includes both financial results and “unusual items” lines. For Q2 2025 the metric gives 48.2%, while the company’s reported operating margin in the same quarter was 27.1% — a 21.1-percentage-point error, entirely from a one-off item of +…. A falsifier built on a series with ±21 pp errors does not measure the business — it measures the data vendor’s normalization noise.

The third thing, the answer to the underlying question — structural deterioration or quarterly noise? Neither: the margin has not deteriorated, it has expanded, and the expansion is real and verifiable in the reported figures, not in yfinance’s EBIT. Reported operating margin has climbed for five quarters running: 27.1% (Q2 2025) → 32.0% → 40.2% → 40.7% → 52.7% (Q2 2026), and at the annual level 21.4% (2022) → 20.5% (2023) → 24.2% (2024) → …% (2025) → 47.4% (H1 2026 annualized). The incremental cost measured at the two ends of the range is 20.8% (revenue +7.32 bn, operating expenses +1.52 bn between Q2 2025 and Q2 2026) and 6.7% over the last quarter. This is genuine operating leverage, produced by the mix shift toward index licensing, where an extra billion of AUM costs almost nothing. The falsifier fired in the direction of “the business is doing better than you assumed,” not “the business is breaking” — and that doesn’t break the thesis, it puts pressure on the valuation side.

The thesis, rewritten after verification. FnGuide sells three things: financial information (DataGuide, WiseReport, QuantiWise) with recurring revenue and near-zero growth in a normal regime; index licensing to ETF managers, paid as an implied fee of ~4.75 basis points on indexed AUM; and fund evaluation, in decline. The index segment has become, for the first time, the majority (53.18% of H1 2026 revenue) and is the engine of the entire stock re-rating. But it is not a business that grows through the company’s own effort — it is a royalty on the level of the Korean stock market: indexed AUM climbed from ~25 to 77 trn KRW in eight months, on a KOSPI that rose 112% and then gave back part of it. Consequently, the 52.7% margin, the 16.11 bn TTM profit, and the tracker’s 0.07 trailing PEG are all functions of the same variable, and the entire valuation reduces to a single question: at what AUM level does the business settle?

The estimated value. I normalize to an indexed AUM of 45 trn KRW (… versus today’s estimated level, … versus the Q2 run-rate, but … above September 2025), which gives normalized revenue of 45.6 bn, operating profit 18.79 bn (margin 41.2% — note: this is not a return to the historical 21% margin, because the mix has changed structurally) and owner earnings of ~15.3 bn, rounded to …. Five independent models, at today’s price: DCF bear …, DCF base …, DCF bull …, EPV Greenwald …, 5-year historical multiples …. Value range 11,949 – …, median … = … versus …. The Monte Carlo simulation over 20,000 scenarios, with λ = 0.30 on net cash, gives a median … and undervaluation probability …%.

Verdict: NEUTRAL, leaning AVOID, at … — unchanged versus 2026-09-17, but for an added reason. An undervaluation probability of …% is literally a coin flip. It is not a distribution you build a concentrated position from, and Radu’s plan explicitly rejects the alternative (diversification). The threshold at which the business again becomes an acquisition: …. What concretely changed versus yesterday: the price rose 1.1% (17,860 → …), the λ actually applied in the simulation dropped from 0.45 to 0.30, i.e. to the company’s own calibration that the prior report had computed but not applied (−2.2 pp of margin), and the multiples model was rebuilt on a reworked normalization with an incremental cost of 20.8% instead of 14.3%. Net effect: the simulation median moves from … to …, and the undervaluation probability from 52.3% to …% — i.e. below the 50% threshold, for the first time in this ticker’s analysis series.