Chapter 3
The Estimate Behind the Reserves
Qifu keeps three separate credit reserves, and all three are driven by one internal number: the expected default rate on its loan book. Deloitte flags that single estimate as the only critical audit matter in the FY2025 audit. It sets the allowance on Qifu's own loans, the reserve behind its loan guarantees, and — through variable consideration — part of revenue itself. The estimate is being made as the book deteriorates: 90-day+ delinquency rose to 2.71% and loan charge-offs reached ¥3.2 billion. The 2024 "record" profit leaned on an unusually light provisioning year; 2025's dip is largely that reversing.
Figures are in renminbi (RMB, ¥) with the FY2025 20-F's own US$ translations at the ¥6.99 = US$1.00 convenience rate shown alongside headline amounts; ratios are currency-independent. Each citation links to the RMB source page.
One estimate, three places it lands
The unit-economics picture (Unit Economics) established that Qifu's growth now comes from loans it either funds itself or guarantees — in both cases keeping the credit risk. That risk shows up in four provision lines on the income statement: a provision for loans receivable (on-balance-sheet loans), a provision for financial assets receivable and a provision for contingent liabilities (both tied to guarantees on off-balance-sheet loans), and a provision for accounts receivable and contract assets [1].
Underneath the four lines sits a single judgment. Deloitte identified exactly one critical audit matter for 2025: "the estimate of expected default rate," which the company uses to value its stand-ready guarantee liabilities (a discounted-cash-flow model), its contingent guarantee liabilities (an expected-credit-loss model), and its allowance for loans receivable. The rate is estimated "on a pool basis according to the historical net default rate by vintage," adjusted for industrial and macroeconomic factors. Deloitte needed its own fair-value specialists to test it [2]. The company describes the same machinery in its critical accounting estimates: default rates built from historical vintage curves, then adjusted for CPI, money supply and delinquent-loan collection rates [3].
Because one input drives everything, the disclosed sensitivity is the cleanest way to size the judgment. A 0.5-percentage-point move in the estimated default rate would change the loan-loss allowance by ¥459.4 million (US$65.7 million) and the contingent guarantee liability by ¥294.2 million (US$42.1 million) — ¥753.6 million (US$107.8 million) together, before tax [4]. Against ¥5,989.7 million (US$856.5 million) of 2025 net income, that half-point is worth roughly an eighth of a year's after-tax profit [5]. A year earlier the same half-point moved ¥694.0 million; the sensitivity has grown as the on-balance-sheet book has [6].
The two reserves the critical audit matter names, with the disclosed effect of a half-point change in the estimated default rate. Source: FY2025 20-F, critical accounting estimates [7].
The provisioning cycle reframes the profit line
Read as a group, the four provisions total ¥5.77 billion in 2023, then drop to ¥3.97 billion in 2024, then climb back to ¥5.85 billion (US$836.1 million) in 2025 [8]. Net income traces the mirror image: ¥4.29 billion in 2023, a record ¥6.26 billion in 2024, then ¥5.99 billion in 2025 [9]. The 2024 peak coincided with the lightest provisioning of the three years.
The swing sits almost entirely in one line. The provision for contingent liabilities — the reserve Qifu builds against guarantees it may have to honour — was ¥3.05 billion in 2023, collapsed to ¥478.4 million in 2024, and rebuilt to ¥1.67 billion (US$238.5 million) in 2025 [10]. Management's own bridge is blunt: operating costs and expenses rose from ¥9.64 billion to ¥12.53 billion in 2025 "primarily due to the increase in provision for contingent liabilities and increase in provision for loans receivable" [11].
The four credit-related provision lines. The contingent-guarantee line (red) collapses in 2024 and rebuilds in 2025. Source: FY2025 20-F, costs and expenses [12].
Two readings compete, and both have support. On one side, the low 2024 provision partly reflects a shrinking guarantee book rather than a deliberate release: the contractual balance of loans Qifu guarantees fell from ¥24.8 billion to ¥22.4 billion over 2025, and fewer new guarantees mean less to reserve against [13]. On the other, new guarantee origination did not shrink in 2025 — the stand-ready guarantee liability provisioned at inception actually rose from ¥2.35 billion to ¥3.55 billion — yet the contingent provision more than tripled, which points to a higher assumed loss rate on new business, not just more of it [14]. The reasonable read is that 2024's record was flattered by a benign default assumption, and 2025 is closer to a through-cycle provisioning level — which means the earnings base the market de-rated from was softer than the 2024 headline suggested.
Delinquency and charge-offs are rising; coverage is holding
The reserve build is not conservatism on a stable book. The overall 90-day+ delinquency rate fell to 2.09% at the end of 2024 but rose to 2.71% at the end of 2025, which management attributes to macroeconomic uncertainty and the April 2025 regulatory changes [15]. On the on-balance-sheet book, gross loan charge-offs climbed from ¥1.84 billion (2023) to ¥2.42 billion (2024) to ¥3.20 billion (US$457.1 million) in 2025, and the loan-loss allowance rose to ¥3.63 billion; net of ¥506.9 million of recoveries, net charge-offs ran near ¥2.7 billion on an average book of about ¥37 billion — a roughly 7% net loss rate [16].
Year-end 90-day+ delinquency, plus 31 March 2026. Sources: FY2025 20-F [17]; Q1 2026 results [18].
Whether the reserves are keeping pace is the point a skeptic presses, and the answer is mixed. On the on-balance-sheet book, the allowance as a share of gross loans barely moved — 8.4% at end-2024 to 8.6% at end-2025 — even as delinquency and charge-offs rose faster than that [19]. On the guarantee book, coverage rose more, from 7.3% to 8.4% of the guaranteed balance, so that reserve was rebuilt harder [20]. A high-yield, short-duration consumer book can carry a ~7% loss rate if pricing covers it, so the level is not alarming on its own; the direction is what warrants attention, and the on-balance-sheet allowance is the line where the cushion thinned relative to the deterioration.
The same dial also sets part of revenue
The default-rate estimate does not only build reserves — it feeds the top line. Qifu recognises loan-facilitation revenue using an estimated transaction price that includes variable consideration tied to future default and prepayment; when prior vintages perform better than modelled, the fee rate re-estimates upward and revenue is booked. In 2025 that catch-up added ¥1,374.5 million (US$196.5 million) of revenue "from performance obligations satisfied (or partially satisfied) in prior periods" — about 7% of total net revenue [21]. In 2024 the equivalent adjustment was disclosed as "not significant" [22].
That creates a tension worth naming plainly. The same estimation machinery was pushing reserves higher (a more cautious default view on the current book) while releasing ¥1.4 billion of revenue from favourable re-estimates of older vintages. The two are not contradictory — older, largely run-off vintages can perform better than newer ones being written into a tougher market — but they sit on opposite sides of one judgment, and a single set of assumptions determines both how much Qifu reserves and how much it earns. For a company already trading near book value on falling profits, the quality of that one estimate carries unusual weight.
What would decide it, and what to watch
The evidence does not settle the durability question by itself; it sharpens it. The reserve build is real and the book is deteriorating on the lagging measure, which argues that 2025's lower profit reflects genuine credit cost rather than accounting caution alone. But coverage ratios rose rather than fell, and the leading indicators turned up early in 2026 — the Chief Risk Officer reported Day-1 delinquency easing to 5.7% and the 30-day collection rate improving to 85.8% in the first quarter, with legacy loans running off [23]. The lagging 90-day+ rate, which the company itself calls "a backward looking indicator," kept climbing to 3.50% at 31 March 2026 precisely because it reflects loans that soured a quarter earlier [24] [25]. Whether 2025 proves a provisioning trough or the first leg of a longer deterioration turns on which of those signals dominates as the pre-cap vintages run off.
Three lines make that check falsifiable. The provision for contingent liabilities (¥1.67 billion in 2025) and the provision for loans receivable (¥3.63 billion) are disclosed each quarter; a further step-up would confirm the deterioration read, a fade would confirm the trough. The on-balance-sheet allowance ratio (8.6%) should rise if delinquency keeps climbing; a flat ratio against a rising 90-day+ rate is where under-reserving would first appear. And the disclosed sensitivity — ¥753.6 million per half-point of default rate — is the multiplier on any of it. What cannot be independently checked is the estimate itself, which is why Deloitte singled it out.