Variant Perception

Figures converted from KZT at historical FX rates — see data/company.json.fx_rates. Ratios, margins, and multiples are unitless and unchanged.

Where We Disagree With the Market

The market is pricing Kaspi.kz on a single consolidated 7.1x P/E as if the three engines — Payments, Marketplace, Fintech — carry the same regulator and rate risk; the upstream evidence says only one does, and the cheapest engine multiple is being applied to the highest-quality one. The sell-side has the direction right (every published price target is above spot) but is debating the wrong question: whether FY26 adjusted-EBITDA growth of +5% is "the floor," when the larger mispricing sits in the implied assumption that Payments take-rate compression and a Türkiye write-down are both already in train. Neither is supported by current evidence — Payments take rate slipped on internal Kaspi B2B mix, not regulator action, and Hepsiburada hit management's EBITDA-breakeven bar in Q4 25. Q2 2026 results plus the Rabobank A.Ş. BDDK ruling test both inside the next six months.

Variant Perception Scorecard

The variant is not "the stock is cheap" — every published price target shares that. The variant is what the discount is for. Our read: the 7.1x multiple is paying for two latent risks (PIX-style open-loop QR and a Türkiye write-down) that the live evidence does not yet support, and a third (Krivenok litigation tail) that is binary on a single docket event the market is treating as permanently unresolved. Evidence strength is high because the relevant signals — Payments take rate quarterly, Hepsiburada standalone EBITDA, Rabobank BDDK filings, Krivenok docket — are all observable. Sell-side price-target range ($90-125) brackets a 30%+ band, and reported short interest dropped 46% post-Tencent (15-29 May 2026).

Consensus Map

What the market appears to believe, where the belief is observable, and the assumption it embeds.

Consensus clarity differs by issue. On the headline multiple anchor and FY26 guide, consensus is sharp — every published PT and Goldman/Morgan Stanley note converges. On Payments take-rate path and Türkiye option value, consensus is more diffuse but the direction of the implied bear assumption is consistent enough to disagree with. On Krivenok, consensus is functionally "do not model" — itself a position, and an unusual one for a binary docket event 17 months past filing.

The Disagreement Ledger

Four ranked disagreements, ordered by what would most update a PM's underwriting if right.

Disagreement 1 — Payments take rate is internal-mix compression, not regulator-mandated commodification. Consensus says "PAGS is the cautionary tale and the 7x P/E compensates for the same NBK risk." The upstream evidence disagrees: Q1 26 compression is explicitly attributed to Kaspi B2B mix in management commentary, not a regulator rule, and no NBK consultation paper on payment interoperability has been published in 24 months. If right, the Payments segment ($832M FY25 NI, sub-1% variable cost, 78% population coverage) merits a payment-platform multiple closer to 12-18x rather than PAGS at 7x. Cleanest disconfirming signal: an NBK consultation document.

Disagreement 2 — Türkiye option value is being priced near zero against a deal structure that earns its own carry. Consensus says "Hepsiburada is a drag, Rabobank is binary, do not credit it." The evidence disagrees: management's Q4 25 EBITDA bar was met (TRY 1M), Q1 26 standalone loss narrowed substantially, Tencent and U.S. endowments took anchor stakes at $78 alongside Lomtadze on 20 April 2026, and the credit market funded the deal at a sub-6% coupon. Implied probability of Rabobank closing inside 12 months is likely 35-50%. If right, conversion would credit a multi-billion uplift on consolidated EV. Cleanest disconfirming signal: BDDK delay past Q4 2026 with no published rationale.

Disagreement 3 — FY25 underlying +18% NI growth is the right read, not the reported +1%. Consensus says "the 'underlying' carve-out is management spin." The evidence disagrees: each of the four itemised shocks is independently reconcilable — smartphone import disruption is documented in Kazakhstan trade data, +200 bps bank tax is a 1 January 2026 statutory step, +220 bps deposit funding is the NBK rate move from 15.25% to 18.0%, and Hepsiburada drag is consolidation accounting. Q1 26 +9% adj-EBITDA against a +5% guide is the first quarter of empirical support. If right, FY26 EPS estimates rise 12-20% and the 7.1x anchor becomes implausibly low. Cleanest disconfirming signal: Q2 op margin ≤52% with no offsetting Hepsiburada uplift.

Disagreement 4 — Krivenok MTD is binary and live, not permanent overhang. Consensus says "foreign-issuer securities class actions resolve glacially; price as permanent risk." The evidence disagrees: docket is 17 months past filing with no public ruling, Rosen Law was still soliciting class members in May 2026, and the typical MTD window is 12-24 months. A clean MTD dismissal removes the only discrete multiple-cap on the page. Cleanest disconfirming signal: denial of motion-to-dismiss — which would push discovery and force a quantified Russia exposure disclosure.

Evidence That Changes the Odds

Eight items from the upstream report that materially move the probability of the variant view. Each is auditable in a primary filing, transcript, or filing-equivalent source.

Each piece of evidence is an observable that the market has either ignored or assumed away. Item 2 (no NBK consultation paper) is the cleanest — a null result the consensus implicitly assumes will end. Item 4 (Tencent + endowment cohort) is the strongest external validation in the file; the stock's round-trip of the +30% rally in two weeks suggests the market has not fully priced strategic-anchor signaling.

How This Gets Resolved

Eight observable signals over the next 6 months that validate or refute the variant. The single-quarter inflection sits at Q2 26 results (~10 August), but the long-term thesis variables (Rabobank, Krivenok, NBK posture) all carry inside-12-month resolution windows.

The highest-conviction near-term test (Q2 op margin) lands inside 60 days, while the highest-thesis-impact resolutions (Rabobank, Krivenok) sit inside mid-2026 and a 12-month docket window respectively. A PM does not have to wait — Q2 26 results are the single forcing event, and a margin print at either extreme would rebase the multiple debate before either binary resolves. Signals 6-8 are continuous watchpoints that move slowly but compound.

What Would Make Us Wrong

The strongest case against the variant is that the market is correct to discount three latent risks into a single 7.1x multiple because they are correlated in ways the upstream evidence underestimates. A Krivenok denial that surfaces quantified Russia exposure could simultaneously trigger an OFAC inquiry, prompt Halyk to use the litigation overhang as a wedge for premium-deposit migration, and give NBK political cover to launch a payments-interoperability consultation. In that scenario, all three disagreements would be wrong at once — and the $97 → $78 round-trip in two weeks would be the market correctly recognising risk before the catalyst sequence unfolds.

Second, the "underlying" reading of FY25 net income is methodologically clean but discretionarily framed. Management chose which four shocks to itemise; the definition of "underlying" cannot be audited without trusting the management reconciliation. If three of four shocks recur in FY26 (rates held at 18%, bank tax at +200 bps, deposit funding elevated, Hepsiburada loss persisting), then "underlying +18%" was a one-time-use defence. The Q2 26 print is the first place the framing actually has to work.

Third, the Türkiye disagreement requires both Rabobank approval and Hepsiburada engagement convergence — a regulatory exercise and a competitive win against Trendyol. Hepsiburada is the #2-#3 platform with Alibaba-backed Trendyol leading; an approved licence with a Hepsiburada losing GMV share would be worst-of-both.

Finally, the most uncomfortable possibility: the 7.1x multiple accurately prices the combination of single-country regulator concentration, founder-controlled governance, an unresolved disclosure tail and an unproven cross-border expansion — and breaking the analysis down by engine is the "sum of parts that doesn't actually re-rate" trap that long EM ADRs habitually fall into. A confirmed lean requires Q2 op margin above 58% combined with at least one of Rabobank approval or a clarifying Krivenok docket event.

The first thing to watch is the Q2 2026 consolidated operating margin print, expected ~10 August 2026 — a reading above 58% with Hepsiburada standalone EBITDA positive is the cleanest single validation of the underlying-engine variant, and a reading at or below 52% is the cleanest single refutation.