Chapter 6

Figures converted from Indonesian rupiah at historical FX rates — see data/company.json.fx_rates. Balance-date figures use their own period-end rate; year-on-year comparisons and single reconciliations use the later period's rate throughout so the percentages hold; the share price and per-share net asset value in the discount table use the 27 July 2026 rate. Ratios, margins, and multiples are unitless and unchanged.

Marks Without Prices

Roughly $543 million of Saratoga's $3,640 million portfolio has no quoted price behind it. The audited notes describe the methods in three short paragraphs, disclose no discount rate, multiple or range, and state that a sensitivity analysis would not be practicable. That slice has tripled as a share of the portfolio since 2022. Writing all of it to zero moves the discount to published net asset value from 60% to 53%.

Sources: FY2025 audit report, key audit matter [1]; FY2025 Annual Report, Note 17 Fair Value of Financial Instruments [2]; discount derived from the $0.0968 close of 27 July 2026.

How big the unpriced book is

The auditor draws the line for you. At 31 December 2025 the Group held $2,073 million of Level 2 and Level 3 investments, including those measured at cost — 55.26% of consolidated assets — and of that, $1,529 million sits in entities that themselves own directly-quoted listed shares [3]. Those are the Bersama Digital and Adaro Strategic holding vehicles, whose value moves with TBIG, ADRO and AADI screen prices. The residual — $543 million — is the part that does not trace to any quoted instrument. It is 14.9% of the $3,640 million of investments the balance sheet carries, and 38% of Saratoga's $1,313 million market capitalisation.

Four consecutive audit reports state the same two figures, so the series can be built on the auditor's own definition rather than a constructed one.

Sources: key audit matters, FY2022 [4], FY2023 [5], FY2024 [6] and FY2025 [7]; portfolio totals from the fair value hierarchy tables, FY2023 [8] and FY2025 [9].

Both halves of the ratio moved. The unpriced balance rose from $199 million to $543 million over three years while the portfolio itself first shrank — $3,999 million at end-2022 to $3,186 million at end-2023 as listed marks fell — then recovered to $3,640 million [10] [11]. The management discussion reports the same trend in its own words: non-listed entities $308.5 million at end-2025 against $218.8 million a year earlier, up 41% [12].

Where the 2025 increase came from

Level 3 investments in shares went from $114.9 million to $308.6 million during 2025 [13]. Note 5's position-by-position table lets that $193.8 million increase be rebuilt, and most of it is accounting category rather than value.

Source: derived from Note 5 Investments in Shares, FY2025 Annual Report [14] [15] [16], and Note 17 [17].

$157.7 million of the increase — 81% — is positions arriving at their existing carrying values. Growth companies 9 and 10 were carried at cost of $88.3 million and $15.6 million at end-2024 and are Level 3 at end-2025 [18]. Growth companies 11 and 12 had no 2024 line at all; the footnote says their value was previously inside the Bersama Digital mark, and the BDIA footnote puts the non-tower assets held there at $53.7 million [19] [20].

The two positions that came off cost are now carried 24.0% and 22.8% above their prior-year figures — $24.8 million between them, 10% of the $246.5 million of fair-value gains booked on investments in shares for the year [21]. How much of that is fresh cash and how much is the change in measurement basis is not disclosed. Note 5's roll-forward runs by strategy bucket, not by hierarchy level, and its 2025 presentation merges additions with reclassifications into a single column where the 2024 version showed additions alone [22].

The timing points to the two investments made in 2024. Saratoga acquired a majority stake in the Brawijaya hospital chain in the first quarter of 2024 and a significant minority of the food-ingredients maker Foodex in the first half [23]. The cost approach applies for twelve months after acquisition [24], the two positions carried at cost at end-2024 hold ownership bands of above 50% and 20-50% matching those two deals, and both moved onto models a year later. The note anonymises, so this is an inference from ownership band and timing rather than a disclosure.

What stands behind the numbers

Note 17 sets out three methods. Anything bought in the last twelve months is held at cost. Where an investee's own net assets are themselves Level 1, Saratoga uses the carrying amount of those net assets. Everything else is valued by management using discounted cash flow and comparable-company multiples [25]. No discount rate, growth rate, multiple or valuation range appears anywhere in the accounts, and no external valuer is named. The note then closes the question explicitly: because the investees span varied industries with unique assumptions, "providing an aggregated disclosure of sensitivity analysis on the key inputs used would not be practiceable nor meaningful" [26].

That is a standing position, not a new one. The same statement sits in the FY2021 accounts [27] and in the Q1 2026 interim statements [28]. The reasoning is defensible for a portfolio of a dozen unrelated private businesses; the consequence is that the 15% of net asset value most exposed to judgment is the 15% for which no quantified sensitivity exists.

The contrast within the same set of accounts is sharp. Note 18(e) discloses, holding by holding, what a 10% share-price move does: $112.0 million for Tower Bersama, $85.6 million for the Adaro complex, $64.8 million for Merdeka and $14.6 million for Mitra Pinasthika — $277 million in total [29]. Price risk is quantified precisely for the 85% that is observable and not at all for the 15% that is not.

Two further features limit what an outside reader can do with the unlisted table. The anonymised labels are not stable identifiers: the FY2023 report shows Growth company 9 at $8.8 million at 31 December 2023 [30], while the FY2024 report's comparative column for that same date shows Growth company 9 at $20.7 million and Growth company 6 at $8.8 million [31]. Tracking one private position across two annual reports is not possible from the published numbering.

And the investor-facing net asset value table takes the presentation one step further out. Every listed holding is shown with its effective ownership and its closing share price; the unlisted book appears as a single "Others" line under Growth Focused of $377.9 million, against $259.8 million a year earlier, with no price, no ownership percentage and no components [32]. That table also sits outside the audit. The auditor's report covers the consolidated financial statements only and records that the 2025 Annual Report "is expected to be made available to us after the date of this auditors' report" [33]. The $0.2444 per share against which the whole discount is measured is management's own presentation, built from audited components.

What the marks have done when they moved

The one unlisted bucket old enough to show a full cycle is digital technology, and it has been marked down almost to nothing.

Sources: Note 5 Investments in Shares, FY2021 [34], FY2023 [35] and FY2025 [36].

From the $34.1 million peak at end-2022 the bucket is down 90%, and 2025's $14.4 million decline was $14.2 million of fair-value change against $0.2 million of divestment [37]. The exits from that book, examined in Realisation Record, were almost entirely markdowns rather than sales. What that record establishes about the estimates themselves is that they are revised, and revised hard, when the underlying businesses disappoint — a point in favour of the marks as much as against them.

The stronger evidence for management on this question comes from 2023. Saratoga reported an $897.7 million net loss on investments in shares and other securities that year, the largest in its listed history [38]. Level 3 investments in shares nonetheless rose from $90.7 million to $105.2 million over the same twelve months [39]. The write-down came from quoted prices, not from a discretionary sweep of the unlisted book while attention was elsewhere. Nor do the growth marks run one way: the growth-focused bucket took $36.9 million of negative fair-value change in 2024 before adding $68.1 million in 2025 [40].

What a haircut does to the discount

The unpriced book is $543 million at the 31 December 2025 rate, or $0.0367 per share on 13,564,835,000 shares at the 27 July 2026 rate used below. Applying a haircut to all of it and holding the rest of the published net asset value constant gives the following.

Source: derived from the FY2025 fair value hierarchy [41], the audit report's Level 2 split [42] and the published net asset value table [43]; share price at the 27 July 2026 close.

The read this supports: the unpriced book is not what creates the discount, and it is not large enough to remove it either. A complete write-off of every dollar that does not trace to a quoted price still leaves the shares 53% below the remainder. For an investor working from a margin-of-safety frame, the estimation risk in the notes is bounded at roughly seven points of a sixty-point gap.

The fact that cuts hardest against reading the unpriced book as immaterial runs the other way, and it is about direction of travel rather than level. The three net asset value lines outside the blue-chip block carrying no share price — growth-focused "Others", digital fund investments, and other shares and securities — went from $460.4 million to $576.1 million during 2025, an increase of $115.7 million [44]. That is $0.0078 of the $0.0255 by which published net asset value per share rose during the year — 31% of the gain, from lines an outside reader cannot verify. The slice is small enough to survive a write-off and large enough to move the reported number, and both statements matter.

What would change the read in either direction is the same event: a transaction. A listing or trade sale of one of the larger private positions would price the model against a market for the first time since the digital book began its decline, and would say more about the reliability of the remaining $543 million than any further disclosure. Failing that, a Level 3 reconciliation separating transfers from revaluation, or the unobservable-input table the accounts currently decline to give, would let the next reader do the work this chapter had to do by inference.