The Banks Fall First: Bangladesh's Power-Sector-to-Banking-Sector Transmission Chain, FY27–FY30

1. Verdict

Partially confirmed — and further along the transmission chain than the original anchor figures suggested.

The thesis holds directionally: Bangladesh's power sector is generating fiscal and liquidity stress that is transmitting into the banking system, and the banking system is already the more acute point of failure relative to the sovereign, which remains current on external debt service. But two of the anchor figures were stale or wrong in ways that understate how far the chain has already progressed, and one load-bearing assumption (a live IMF program providing a stabilizing backstop) no longer holds as stated.

Strongest evidence for the thesis: The banking system's reported NPL ratio has continued climbing past the 32.26% anchor figure — to 32.78% (Tk6.07 lakh crore) by June 2026 [C], with total distressed assets (NPL + rescheduled + SMA) reported as high as ~60% of the loan book by mid-2026 [C][T]. (Caveat per Finding 2: the climb from 10.11% to 32.78% substantially reflects post-2024 AQR unmasking of previously hidden NPLs alongside fresh deterioration — the ratio evidences recognized stress, not purely new power-sector transmission.) Nineteen banks failed to maintain minimum capital adequacy at end-2025, and system-wide CRAR has been reported at negative territory in mid-2026 press accounts [C], a condition none of Bangladesh's regional peers (Pakistan 20.8% CRAR, Sri Lanka 19.4%, India 17.2%) shares. Simultaneously, BPDB's arrears to IPPs — which had been substantially cleared through 2025 bond issuance — reaccumulated to roughly Tk20,000–25,000 crore by mid-2026 [C], confirming the arrears-rebuild dynamic the thesis predicts rather than a one-off historical episode.

Strongest evidence against (or complicating) the thesis: Bangladesh's FY27 budget explicitly under-allocates the power subsidy (Tk37,000 crore against a BERC-estimated need of Tk41,000–56,000 crore) [C], meaning the fiscal gap the thesis describes is a known, budgeted, and politically chosen shortfall rather than a hidden one — which cuts both ways. It confirms the mechanism is live, but it also means the government has repeatedly demonstrated a working (if leaky) release valve: ad hoc special bonds, ADP reallocations, and emergency Finance Division releases have kept BPDB's arrears from compounding into full-blown IPP defaults so far, across at least three arrears build-up/clearance cycles since 2023. The system has proven more elastic under repeated stress than a linear transmission model implies.

Where we are in the chain (as of Q3 2026): Stage 2–3 of 4. Capacity-payment-to-subsidy-ceiling gap is open and structurally recurring (Stage 1, confirmed). BPDB arrears to IPPs are actively cycling — building, partially clearing via bond/ADP transfers, rebuilding (Stage 2, confirmed, and now on its third or fourth cycle since 2023). Bank-sector asset quality deterioration is severe and system-wide but is not yet demonstrably power-sector-led in the public data — it is concurrent with power-sector stress but the dominant documented drivers are politically directed lending (S Alam Group and similar), Islamic-bank governance failures, and broad-based post-2024 AQR unmasking of previously hidden NPLs (Stage 3, partially confirmed — power-sector contribution not yet isolable from public disclosures). Full capital-adequacy erosion specifically attributable to power-sector loan quality at the top 5 exposed banks (Stage 4) could not be confirmed or falsified from public reporting; bank-by-bank power-sector exposure disclosure does not appear to exist in the public domain in Bangladesh, which is itself a finding (see §2, Finding 1).

2. Findings

Finding 1 — Exposure concentration: NOT independently verifiable from public data

Evidence tier: [S]/gap. No public Bangladesh Bank disclosure, IMF staff report, or press investigation located in this research names bank-by-bank power-sector loan exposure, single-name IPP/sponsor concentrations, or expresses power exposure as % of capital for named banks. This is itself a material finding: unlike Pakistan (where circular-debt bank exposure by institution is disclosed via the 2025 18-bank financing consortium term sheet) or India (where PRAAPTI discloses genco-by-genco overdue amounts), Bangladesh does not appear to publish sector-level credit concentration data at this granularity. The closest available proxies:

• State-owned commercial banks (Janata, Agrani, Rupali, Sonali, BASIC) carry the largest aggregate NPL stocks in the system (Tk1.47 lakh crore combined as of ~Sept 2026) [C], and SOCBs have historically been the primary lenders to BPDB and public/quasi-public power entities, making them the most probable concentration point, but this is inference, not disclosure.
• BPDB's own arrears are to IPPs directly (trade payables), not bank loans per se; the banking-system transmission runs through the IPPs' own debt service to their lenders, which is a second-order and even less transparent linkage.
• Confidence band: 30-40% that the true top-5 bank exposure profile resembles the anchor hypothesis (SOCB-concentrated, single-name IPP sponsor concentration); this is a plausible-but-unverified prior, not a finding.
• Falsification condition: Bangladesh Bank publishing a sectoral credit concentration report (it publishes NPL-by-bank but not loan-purpose-by-bank at this granularity in the sources reviewed) would resolve this either way.

Finding 2 — True vs. reported asset quality: the forbearance-adjusted estimate is now largely merged into the reported figure, and it is worse than the anchor assumed

Evidence tier: [C]. The anchor figure (32.26% reported NPL) is stale by one quarter. As of June 2026, reported NPLs stood at 32.78% (Tk6.07 lakh crore of Tk18.5 lakh crore total loans) [C], and multiple outlets report Bangladesh has the world's second-highest NPL ratio after Ukraine. Critically, the forbearance-vs-reported gap the thesis instructs researchers to estimate has already been substantially closed by policy action, not by continued concealment: post-August 2024, Bangladesh Bank dissolved boards at 14 banks and ran KPMG/EY-conducted Asset Quality Reviews (AQRs) through 2025 that forced recognition of previously hidden NPLs [C]. This is why the ratio jumped from 10.11% (June 2023, pre-AQR classification) to 20.20% (Dec 2024, Basel III reclassification) to 32.78% (June 2026) — the jump reflects unmasking, not fresh deterioration alone, though fresh deterioration (Tk31,500 crore added in Q1 2026 alone) [C] is also occurring concurrently.

• Including restructured loans and Special Mention Accounts, total stressed assets reached an estimated Tk11.2 lakh crore, or ~61% of the total loan portfolio, as of mid-2026 [C].
• Provision shortfalls reached Tk205,665 crore by March 2026 [C].
• Confidence band: 75-85% that the officially reported figure is now within 5-8 percentage points of a fully forbearance-adjusted figure, given the AQR process — a much narrower gap than the pre-2025 baseline, because the concealment mechanism (rescheduling, evergreening) has been substantially — though not completely — dismantled by the 2025-26 reform push.
• Falsification condition (from the original brief) — status: "Audited system NPL falls below 20% with forbearance fully unwound" has not occurred; the ratio has moved in the opposite direction. This falsification condition is decisively not met, which is evidence for the thesis.

Finding 3 — The transmission chain, quantified (FY27 data point; FY28-30 requires modeling assumptions not available in public sources)

Evidence tier: [C] for FY26-27 data; [T] for forward projection.

• FY27 capacity payments: Tk52,608 crore, up from Tk48,261 crore in FY26 [C] (anchor figure confirmed).
• FY27 subsidy allocation: Tk37,000 crore [C] (anchor figure confirmed) against a BERC-estimated need of Tk41,000 crore even after a 16.7% tariff hike already implemented in mid-2026 [C], implying a BPDB operational gap of roughly Tk15,000-19,000 crore for FY27 alone, before any further tariff or FX shocks.
• BPDB's real-time payable position to IPPs stood at approximately Tk20,000-25,000 crore as of mid-2026 [C] — reconstituted after a partial clearance funded by Tk26,500 crore in special bonds issued in 2025, most of which was absorbed by the power-sector backlog rather than resolving it [C]. This is the third documented arrears cycle since the August 2024 transition (arrears build → partial bond/ADP-funded clearance → arrears rebuild), a pattern consistent with a structural, not episodic, gap.
• BPDB is contractually required to pay interest (1-year T-note yield + 4%) on late payments but "has rarely honoured this clause" according to industry sources [C], and has additionally been deducting liquidated-damages penalties from IPPs for outages caused by BPDB's own fuel-payment delays — a dynamic actively litigated at BERC and threatened for the High Court as of late 2025 [C]. This is a second, harder-to-quantify channel of IPP financial distress beyond the headline arrears figure.
• What the public record does not support: a quantified IPP debt-service-coverage-ratio path, or a bank-by-bank provisioning/CAR erosion projection. No source located models FY28-30 capacity payments against contracted-plant schedules with the granularity the original brief requested; BPDB's own generation expansion plan and PPA schedules are not fully public. This sub-question is not resolvable from open sources at the precision requested; any FY28-30 numeric path offered elsewhere should be treated as a rough extrapolation, not a sourced finding.
• Confidence band: 65-75% that the capacity-payment/subsidy gap remains structurally open (Tk10,000-20,000 crore/year order of magnitude) through FY28-30 absent a step-change in either tariffs or contract renegotiation, based on the pattern of three consecutive fiscal years (FY25-27) each showing a materially under-funded subsidy line relative to BERC's own cost estimates.

Finding 4 — The tariff-hike paradox: collection efficiency data not located; qualitative confirmation only

Evidence tier: [C] for the tariff hikes themselves; gap on collection-efficiency quantification. BERC implemented a wholesale/transmission/retail tariff increase in mid-2026 (a Tk1.52/unit retail increase confirmed) [C], projected to raise ~Tk14,200 crore in revenue against the ~Tk56,000 crore three-year subsidy need — closing perhaps a quarter of the gap [C]. BERC simultaneously cut the guaranteed rate of return to IPPs from 12% to 6% [C], which reduces the capacity-payment burden mechanically but also directly threatens IPP debt-service coverage ratios (a lower guaranteed return on the same debt load mechanically tightens DSCR), reinforcing rather than resolving the transmission-chain mechanism. No source located provides a collection-efficiency percentage for the FY26 tariff hike specifically (distinct from generation-cost recovery, which BERC does track). This is a gap, not a resolved finding.

• Falsification condition status: "A tariff path closes the capacity-payment gap with collection efficiency above 95%" — not met; BERC's own numbers show the FY26-27 tariff actions close at most ~35% of the projected three-year gap (Tk14,200cr of Tk56,000cr) even before considering collection losses.

Finding 5 — Capacity payment trajectory: FY27 confirmed; FY28-30 schedule not public

Evidence tier: [C] FY26-27; gap FY28-30. As above — FY26 (Tk48,261cr) to FY27 (Tk52,608cr) is a confirmed ~9% year-on-year increase [C]. No public, dated capacity-payment schedule extending to FY28-30 broken out by fuel/contract type (coal, LNG, Adani, Rooppur, renewables) was located. Rooppur's addition to the capacity-payment base is a new and significant unknown for FY27-28 (see Finding 8) since the plant is only now entering commercial operation and its capacity-payment terms under the Russian financing structure are not the same PPA-with-IPP-sponsor model as the rest of the anchor figures — this is a sovereign-to-sovereign debt service obligation (to Russia, repayment beginning 2027) layered on top of, not inside, the BPDB-IPP capacity payment system. The thesis as originally framed does not clearly distinguish these two payment obligations, which is a modeling gap worth flagging: Rooppur stress is a separate transmission channel (sovereign external debt service to Russia) from the IPP-arrears channel the rest of the thesis tracks.

Finding 6 — Precedents: see §4 table below.

Finding 7 — Adani as a swing variable: materially updates the anchor figures

Evidence tier: [C]. This is the single largest correction to the original brief. The anchor's "$400-500M/year" overcharge figure is superseded by a formal government finding: Bangladesh's National Review Committee (NRC), in a report dated January 20, 2026 (reviewed by Reuters, not yet public in full), found the Godda plant priced power at a 39.7% premium over its nearest private-sector comparator, and separately — in earlier, lower-confidence reporting — that pricing ran as much as 50% above what it should be [C][T], driven by two specific mechanisms: (1) Adani passing Indian corporate tax liabilities through to the Bangladesh-side tariff, and (2) "excessively priced" coal inputs. The NRC explicitly recommended renegotiation of "the most fiscally damaging provisions" [C]. Separately, and confirming the arrears-cycle dynamic: BPDB's overdue payment to Adani reportedly reached $850 million at the peak of a 2024 dispute (later resolved with a partial-payment/continued-supply arrangement) [C], and Adani has cut Godda output roughly in half at least twice (Oct-Nov 2024, and again per a July 2026 report tied to a technical/payment issue) [C] — direct, repeated evidence of the arrears-to-curtailment transmission mechanism the thesis describes, already realized rather than merely modeled.

• Dollar-value relief estimate: If the NRC's ~40-50% overpricing finding were substantiated and renegotiated down to the comparator benchmark, plausible annual relief is in the $150-300M/year range on an estimated $600M-1B/year Godda contract value — a lower bound than the original $400-500M anchor but directionally consistent and now backed by an official (if not yet public) government finding rather than press triangulation alone.
• Feasibility: Political feasibility appears to have increased materially since the anchor was drafted — this is now a formal interim/post-transition government finding, not merely activist or opposition pressure, and the Bangladesh Supreme Court had already ordered a review of all Hasina-era Adani agreements in November 2024 [C]. Contractual feasibility (arbitration under the 25-year PPA vs. negotiated amendment) is not resolved in available sources.
• Confidence band: 55-65% that some negotiated relief (likely below the full 40-50% overpricing gap, given arbitration risk and India-Bangladesh bilateral energy-security considerations) is achieved within the FY27-30 window; 20-25% that the relief exceeds $300M/year (the stated falsification threshold), given the gap between "recommended for renegotiation" and an actually executed, quantified amendment.
• Falsification condition status: Not yet met ("delivers >$300M/year verified relief" — no renegotiated outcome has been announced as of the latest sources reviewed); under active review, not resolved.

Finding 8 — Rooppur dispatch risk: anchor figure (1.66s UFLS margin) not independently verifiable; underlying risk is real and newly live

Evidence tier: [S] for the specific 1.66-second figure — not located in any source reviewed and should be treated as unverified pending a primary technical source (Power Grid Bangladesh / BAERA grid-code documentation). [C] for the underlying risk. Rooppur Unit 1 began fuel loading April 28, 2026, and BAERA's own chairman publicly identified grid stability as "the biggest immediate challenge" following fuel loading, explicitly noting that nuclear generation requires frequency control precision ("50 hertz") different from gas/coal plants, and that Power Grid Bangladesh had been "repeatedly warned" to meet international standards [C]. This is a live, acknowledged, technical risk as of the most recent reporting (April-September 2026), not a resolved or hypothetical one. Total project cost has also risen to ~Tk1.39 trillion (from FX depreciation, not scope change) [C], with Bangladesh's 10% domestic contribution obligation and Russian loan repayment beginning in 2027 — the same fiscal year the capacity-payment gap is already projected to be unbridgeable on current subsidy allocations. This stacks a new sovereign-to-Russia debt-service obligation directly onto the FY27 stress window the thesis is testing.

• Commercial operation (full ~1,200MW per unit ceiling) is not expected until late 2026/early 2027 at the earliest, with Unit 2 pushed to a December 2027 (or, per a further extension, June 2028) target [C] — meaning the "full export" dispatch-risk scenario the anchor hypothesis assumes is a near-term but not yet realized condition; the grid has not yet been tested at scale.
• Who bears capacity payment if Rooppur cannot evacuate power: not addressed in any source reviewed — this is a genuine gap. Given the debt structure (sovereign loan from Russia, not a standard IPP PPA), it is likely the Bangladesh government/BPDB bears curtailment cost directly rather than an IPP-style capacity-payment dispute, but this is inference.
• Confidence band: low (informational, not probabilistic) — the specific UFLS margin figure should be dropped from any published version of this analysis unless independently sourced; the qualitative risk (nuclear grid integration risk in a chronically underinvested transmission system) is well-evidenced.

Finding 9 — IMF program interaction: the anchor's "current program" framing is outdated; Bangladesh is between programs

Evidence tier: [E]/[C]. This is the second major correction to the original brief. The ECF/EFF/RSF arrangement referenced in the anchor (approved January 2023, ~$5.5bn, of which ~$3.65-4.5bn had been disbursed across successive reviews) [C] is not an ongoing, open-ended backstop — its combined Third/Fourth Review concluded June 2025, a Fifth Review mission concluded November 2025, and as of June 3, 2026, the IMF confirmed Bangladesh has formally requested a new, successor arrangement, with staff "engaging on the reform agenda" rather than administering an active program [E][C]. The IMF's own mission chief explicitly named "banking-sector weaknesses" as a driver of the need for a renewed program [E]. This means:

• The banking-sector reform conditionality the original brief assumes is "current" (recapitalization targets, NPL recognition standards) was designed for the old program and its enforcement mechanism has lapsed pending a new agreement — a materially different, and more exposed, position than "IMF conditionality is currently disciplining the banking sector."
• The June 2026 Article IV / program-request context explicitly flags that Bangladesh met its primary fiscal deficit target only "through significant cuts in capital and social spending" [E] — i.e., fiscal space is already being consumed by austerity, which narrows the room to absorb a power-sector-driven banking shock via further capital injection without a new external anchor.
• A parallel domestic reform track (Bank Resolution Ordinance 2025, mandated IFRS-9 adoption by December 2027, risk-based supervision from January 2026, a June 2026 NPL deadline of <5% for private banks / <10% for SOCBs) is running independent of and ahead of the IMF successor-program negotiation [C] — the June 2026 NPL deadline was, per the same data reviewed, missed by a wide margin (32.78% actual vs. targets of 5-10%), which is a direct, dated falsification-adjacent data point: the domestic reform program's own headline target failed on schedule.
• Confidence band: 70-80% that a successor IMF program is agreed within the FY27-28 window (both sides have publicly signaled intent and the alternative — no program during acute banking stress — is a tail risk neither side wants), but program banking-sector conditionality and power-sector arrears resolution are very likely to be negotiated as separate, only loosely coordinated tracks, based on the pattern that BPDB subsidy/arrears decisions have continued on a purely domestic budget cycle (Finance Division, BERC) throughout the IMF program gap, without visible Fund sign-off gating them.
• Falsification condition status: "IMF program recapitalizes the top 10 banks above 12.5% CAR on a clean book" — not met and, given system CRAR reported as negative in mid-2026, currently a distant prospect; no evidence located of a program-linked recapitalization plan with that specific target for a successor arrangement (the prior program's provisions do not appear to have included this either, based on sources reviewed).

Finding 10 — Early-warning dashboard

See §5.

3. The Transmission Chain — Dated, Quantified Flow (as far as public data supports)

Read of the chain: the loop (arrears build → partial clearance via bond/budget transfer → arrears rebuild) has now completed at least three observable iterations since August 2024 on roughly 9-12 month spacing (three data points — treat as an observation, not a measured cycle). Each iteration appears to shorten the "clean" period between clearance and rebuild. That acceleration — not the absolute Tk figures — is the strongest quantitative signal that the transmission mechanism is intensifying rather than stabilizing.

4. Precedent Table

Synthesis: None of the three precedents is a clean match. Pakistan is the closest structural analogy (banks absorbing circular debt directly) but shows the mechanism can be repeatedly "managed" without resolution for years rather than producing a discrete crack. Sri Lanka shows the reverse sequencing is also plausible regionally. India shows what institutional buffers (transparency portal, LC mechanism, dedicated financing vehicle) look like — and Bangladesh's absence of all three is arguably the most important comparative finding of this research: the risk is not just the size of the numbers, but the absence of the disclosure and payment-security infrastructure that let Pakistan and India manage (if not solve) the same problem.

5. Early-Warning Dashboard

6. Sources (accessed September 2026)

• The Daily Star — "Tuku signals review of capacity charges for private power producers"; "Bangladesh enters nuclear energy era"; "Bangladesh Energy Budget 2026-27"; "Largest budget in the works"
• The Business Standard (tbsnews.net) — BERC subsidy estimate reporting; "Bangladesh's next financial crisis may begin in the power sector"; "BPDB sits on power producers' bills"; NPL/CRAR reporting (multiple); "Nearly half of loans in 10 banks flagged as risky"; "Grid stability now biggest task — Rooppur power supply"; "Six state-owned banks pile up Tk1.47 lakh crore in bad loans"
• New Age (newagebd.net) — "Govt to face tricky situation to release budget support"
• CPD (Centre for Policy Dialogue) — FY27 budget power-sector analysis; NPL trend reporting
• Dhaka Tribune — "NPLs surge further by Tk31,500cr in Q1'26"; "Distressed loans exceed Tk10 trillion"; "Rooppur races to join global nuclear boom"
• Bonikbarta — "Bangladesh power sector back in arrears despite multiple repayments"; "Finance Division imposes strict conditions on power subsidy disbursement"
• IANS / Bizzbuzz / Munsifdaily — syndicated reporting on Tk25,000cr BPDB-IPP arrears
• Business Today (India) / Business Standard (India) / TRT World — National Review Committee Adani Godda pricing findings, Jan 2026
• South Asia Monitor, Model Diplomat, Adani Watch — Adani Godda PPA history and 2024 payment dispute
• World Nuclear Association; Prothom Alo (En) — Rooppur commissioning timeline and cost
• IMF — Press Releases 26/182, 26/029, 25/369, 25/213; 2025 Article IV Staff Report; Country Report 25/150
• BBF Digital; M A Fazal & Co. — 2025-26 banking-sector reform chronology (AQRs, Islamic bank merger, Bank Resolution Ordinance)
• Pro Pakistani, Arab News, Pakistan Today (Profit) — circular debt figures and bank-financing package, 2024-2026
• Wikipedia / NUS-CRI / Model Diplomat / Ade Rana / FT.lk / Bloomberg Law — Sri Lanka 2022 default chronology and bank credit-health data
• Mercom India, Business Today India, Deccan Herald, Business Standard India — DISCOM dues (PRAAPTI) data, 2019-2021

Flagged as unresolved / not publicly obtainable within this research pass:

• Bank-by-bank power-sector loan exposure (Finding 1) — no public disclosure located.
• FY28-30 capacity-payment schedule broken out by contracted plant (Finding 3/5) — not public.
• Collection efficiency of FY26-27 tariff hikes, specifically (Finding 4) — not located.
• The 1.66-second UFLS margin figure (Finding 8) — could not be verified in any source; should be dropped or re-sourced from a primary Power Grid Bangladesh / BAERA grid-code document before further use.
• Bangladesh Bank's own forbearance-history circulars with dates, 2020-2026 (originally requested in the brief) — only referenced in secondary synthesis (AQR chronology), not compiled circular-by-circular here; a dedicated Bangladesh Bank document review would be needed to complete this.