House view: the cross-section

1 · Banks: what the market pays per rupiah of demonstrated economics

The excess-return model values each bank from its own audited record: book equity plus the present value of returns above an ~11.2–11.3% cost of equity (Rf 7.26%, Damodaran EM bank betas). Against the 2026-07-31 snapshot the cross-section splits cleanly. The market pays HALF or less of modeled value for three of the fifteen: BBTN at 39% (P/B 0.47x vs 1.21x modeled), BJTM 44% and BBNI 48% (0.78x vs 1.65x; priced below book while the record models well above it). Five more sit between half and two thirds: BJBR 52%, BMRI 55%, BNGA 56%, BTPS 58% and NISP 59%; BBRI (72%) and BRIS (78%) sit above them. At the other end the market pays a premium over the model for BBCA (104%: priced almost exactly at its demonstrated economics), MEGA (109%) and BDMN (114%).

Two readings deserve care rather than excitement. First, the deep BPD discounts can be the market charging for things the model deliberately excludes: regional-government ownership risk, deposit concentration, liquidity of the shares, not an error by either side; the model states what the record earns, the price states what the market will pay for that record. Second, the most instructive inversion is PNBN: at 0.41x book it LOOKS like the cheapest large private bank, and on this snapshot it carries the lowest market P/B of all fifteen, but its own record (ROE ~5.8%, far below the cost of equity, by management’s stated choice of balance-sheet fortress over lending) models to just 0.33x; the market is actually paying a 22% premium over demonstrated earning power for optionality the record has not yet shown. Cheap-to-book is not cheap-to-economics. ARTO is the same lesson at the extreme: a startup-phase record that cannot yet justify any premium to book models to approximately zero excess value, while the market pays 1.92x book; a pure expectations position, which the slider on its report page makes explorable.

2 · Mining’s two tribes: harvesters priced below mid-cycle, builders priced on hopes

With cyclical names valued on NORMALIZED defaults (full-window mean margins, no cycle-position growth extrapolation), the mining cross-section splits into two tribes. The cash harvesters trade below even mid-cycle economics: ITMG’s normalized model sits ~240% above its price, with MEDC (+206%), PTBA (+175%), GEMS (+153%) and INDY (+119%) in the same direction; the market prices these as if mid-cycle margins themselves are unrepeatable. The builders invert it, and the model now says so in words rather than with a number: for MDKA, MBMA, CUAN, AMMN and PTRO the mechanical DCF is SUPPRESSED, because on default assumptions the modelled enterprise value falls below net debt and minority claims, leaving a negative equity residual. Equity cannot be worth less than nothing, so no per-share figure is published for them; the condition itself is the finding, and it is the construction-phase signature: losses and debt-funded capex on the record while the price pays for what the finished asset should earn. HRUM is the instructive exception and it moved a long way. Once terminal capex is set at replacement level rather than at the build-phase rate the last three years happened to capture, its gap flips from deeply negative to +78%. That single assumption is how much a construction-phase valuation depends on what you assume the company spends AFTER the build ends, which is why this document now states that convention openly in its limits. Neither tribe’s gap is a verdict: the harvester discount embeds real energy-transition terminal risk the perpetuity cannot see, and the builder premium is exactly what the reverse DCF frames it as; an explicit growth bet, checkable against each project’s milestones rather than against last year’s income statement. One harvester fact worth stating plainly, because it was verified from the filings rather than inferred: BYAN repaid its remaining $340m of bank loans during FY2025 and ended the year with NO interest-bearing debt at all; as did AALI in plantations, retiring Rp3.19tn. Two commodity names choosing the balance sheet over the growth option, in the same year.

3 · The property discount: where the market disbelieves the record most

The most consistent sector-wide gap sits in property: CTRA (+391%), SMRA (+321%), BSDE (+213%) and PWON (+166%) all carry forward-DCF defaults far above their prices. Some of this is the macro backdrop doing exactly what it should: with BI at 5.75% defending a record-low rupiah and the 10-year at 7.26%, rate-sensitive, pre-sales-driven businesses deserve compressed multiples, and a WACC built from today’s curve already carries part of that. But the size and uniformity of the residual says the market is pricing a worse future than the delivered margins and revenues describe. SMRA is the honest complication: its forward DCF screens cheap while its reverse DCF screens expensive (the price implies 17.7%/yr FCF growth against a −23.7% delivered FCF CAGR). Checked against the record, that divergence is largely the PRE-SALES CLOCK, not deterioration: FY2021–22’s fat operating cash (Rp2.4–2.7tn) was installments collected before revenue recognition, and FY2024’s +86% revenue jump is recognition of cash banked earlier, so the reverse DCF reads from a recognition-phase FCF base while the forward DCF reads P&L momentum. When two models disagree this hard, the disagreement itself is the finding: the equity story hinges on where the pre-sales cycle sits, not on either model alone.

4 · What prices demand versus what was delivered

The reverse DCF asks one question of every price: what constant FCF growth would justify it? The widest demand sits in healthcare: SILO’s price implies 26.5%/yr against a −17.0% delivered FCF CAGR (+43pp gap), MIKA +28pp. Checked against SILO’s record, this is NOT a bet on a capex cycle ending: capex is still rising (6.6%→13.9% of revenue, FY21–25) and FCF has sat flat near Rp0.7tn for three years because the hospital build-out absorbs all the operating-cash growth. What the price demands is that today’s expansion converts into tomorrow’s cash at a pace the record has not yet shown: a real expansion-payoff bet, stated as such. The opposite tail is where prices imply LESS than the record: ISAT; with the honest correction that its headline 55.4% delivered CAGR rides a merger-transition base year; measured clean from FY2022, delivery is ~20%/yr; against which the price still implies only 7.7%/yr. The market treats the post-merger FCF ramp as largely finished even on the conservative reading. Extreme negative gaps on GGRM, MYOR and ASSA (−132 to −171pp) should be read with the small-base caveat below before anything else.

5 · Model limits, named out loud

Where a gap says more about the model than the market, it is listed here, not buried. (1) JSMR’s +439% and ASSA’s +198% forward-DCF gaps are dominated by model shape: toll roads run on FINITE concessions a growing perpetuity flatters, and ASSA’s delivered growth is a transformation spike no default should extrapolate; read both pages’ assumption tables before reading the gap. (2) Delivered FCF CAGRs computed from near-zero bases (GGRM, MYOR, ASSA, PTPP) are arithmetic noise, not trajectories. (3) The mining “mid-cycle” normalization is only as mid-cycle as its window: FY2021–2025 contains one boom (2022) and no trough, so the full-window mean margins behind the harvester gaps (ITMG, MEDC) likely still sit ABOVE a true through-cycle level; the direction of the finding survives, the magnitude is flattered. ADRO’s normalized margin additionally spans the AADI perimeter change (pre-2023 years include the thermal business later spun off), which is why no ADRO forward-DCF figure is cited in this document. (4) Thirteen names have their mechanical DCFs SUPPRESSED by rule rather than shown, for two different reasons that should not be conflated: five because the record is pre-profit or corrupted (GOTO, BUKA, TPIA, PTPP, ADHI), and eight because the modelled enterprise value falls below net debt and minority claims, so the equity residual is negative (EXCL, HEAL, AMMN, MDKA, MBMA, CUAN, BUMI, PTRO). WIKA is a fourteenth abstention of a different kind and is counted separately below: not suppressed but unpriced. Equity cannot be worth less than nothing, so those pages state the condition in words instead of printing a negative price, with the components left visible so the arithmetic stays checkable. GOTO additionally prices outside the reverse-DCF’s solvable range, and is the only name on the roster that does. LSIP sat here at the previous snapshot and no longer does: at this one its reverse DCF solves, and it solves at −48%/yr against a delivered +13%/yr, so the price implies free cash flow roughly halving every year while the record grew it. That is a finding rather than an abstention, and it is what a solvable range buys you. (4b) Three model inputs are set by convention rather than by the raw history, and each is stated on the page that uses it. Terminal capex is held at REPLACEMENT level (equal to depreciation) rather than at the trailing three-year rate, because carrying a build-phase ratio into perpetuity charges a company forever for expansion it never receives, while carrying a capex pause credits it with never reinvesting again; this roster contained both distortions. The effective tax rate feeding NOPAT is capped at 35%, because above that the pretax approximation is picking up minority interests, which are already deducted separately from enterprise value, rather than tax. And five companies use a company-specific beta row instead of their industry default where the industry note already identified the mismatch (DCII, BREN, PGEO, PGAS, AMRT). (5) Ninety-nine of the hundred covered companies carry a dual-source-verified price at this snapshot. The hundredth is WIKA, and its price is WITHHELD rather than published: the stock has not traded since 17 February 2025, and the last quote sat unchanged for 349 sessions with volume in none of them. The dual-source check could not catch that by itself, because both sources derive from the same frozen quote and therefore agree perfectly; only the market SESSION reveals it, and the snapshot now gates on that. A price nobody can transact at is not a price, so rather than print a year-and-a-half-old market capitalisation next to current accounts, every price-derived figure on that page abstains and says why. Its audited operating record is untouched and still fully analysed, because none of it ever depended on the quote. This is the same discipline the ratio engine applies everywhere else: omit with a reason rather than publish a number that cannot mean what it appears to mean. (6) The bank model’s implied retention derives from delivered book-equity growth, which also carries OCI effects (FVOCI bond marks, revaluations): an approximation, disclosed on each bank’s panel. (7) Every number above moves with the sliders on its own report page; this document is the map, the pages are the territory.