Layer 2 · In depth

Strategy - nine deals and the order to do them in

A plan built as exchanges. Each deal gives a group something it values for a change in behaviour that makes production pay better than taking. Design rules, nine deals with their prices, the hardest four in detail, and what not to do.

Updated 21 Nov 2026 · 32 min read

On this page
  1. The short version
  2. Design rule zero: the compact
  3. Seven design rules
  4. Step zero: the straight story
  5. The nine deals
  6. The ground rules in each deal
  7. The four hardest deals in detail
  8. Deal 1: pay-to-exit for bad debt
  9. Deal 2: the insurance pact
  10. Deal 4: cadre-to-career and legal party money
  11. Deal 7: the madrasa livelihoods compact
  12. The other five deals
  13. Deal 3: pay for accountability
  14. Deal 5: the 2029 export compact
  15. Deal 6: the small-business compact
  16. Deal 8: barracks for budget
  17. Deal 9: the migrant and diaspora compact
  18. Reforms that support the deals
  19. What not to do
  20. What is uncertain here

This page is the detailed reasoning behind the path forward. It takes the loops on circular dependencies and the group payoffs on group incentives and asks what set of exchanges could re-aim enough groups to move the country. The dates and phases are on the roadmap. The facts about banks are on the banking crisis page and are not repeated here; this page owns the reform design.

The short version

The plan is a set of deals. Each one gives a group something it values in exchange for a change in behaviour that raises the return to making things or lowers the return to taking them. The order matters more than the list. Start by telling the public the truth about the economy and resolving the banks, because every other deal needs credit and credibility. If the constitutional amendment the government plans to table in January 2027 proceeds (the date is an assumption; see the fallback), use it to make losing power safe. Use the November 2029 graduation deadline to push the export agenda. Run the extortion and youth package in a few measurable places before scaling it. Expect little growth payoff before 2029, and say so.

A country with limited administrative capacity cannot pass a hundred reforms at once. Confronting the bureaucracy, protected firms, political cadres and religious institutions on the same day builds a coalition of the opposed. The deals are therefore sequenced, and the hardest ones are piloted.

Design rule zero: the compact

Every deal below is built on a short set of ground rules for working together across differences: shared prosperity with fairly shared costs; one law for everyone including the state; persuade, never coerce; justice through fair, timely institutions, never retaliation; losing power is safe; public office is a public trust; open numbers, sunset clauses and correcting mistakes. A recommendation is credible only if its sponsors would accept it under their strongest rival. The rules are in ground rules and the reasoning in the compact across differences. The table below says where a rule changes a decision.

Seven design rules

  1. Ask no group to act against its interest. Change the interest. Appeals to patriotism have been tried for fifty years.
  2. Pay in the cheapest currency. Status, security, legal income and a safe exit cost the country little. Impunity, monopoly and vetoes cost a lot. Offer the first kind to buy back the second.
  3. Make the legal path cheaper than the illegal one. Tax must cost less than extortion. Repaying must cost less than defaulting. Passing the exam must be easier than paying the bribe.
  4. Compensate losers once, in the open. A one-time settlement is cheaper than a permanent rent, and a public deal can be judged where a private one cannot.
  5. Borrow outside deadlines. The 2029 graduation clock, the IMF programme and EU conditions do what a domestic coalition cannot yet do.
  6. Build where it can be measured. A few zones, ports, courts and highways, with numbers published every quarter.
  7. Sunset everything. Every concession should expire unless renewed, so that buy-outs do not turn into new entitlements.

One more principle from the comparative evidence: Bangladesh already has many of the laws it needs. An anti-corruption commission has existed since 2004, judicial independence is in the Constitution, and a right-to-information law has been on the books since 2009. What is missing is an enforcement equilibrium: the laws bite on the outgoing party and are otherwise dormant.1 Each deal below is therefore judged by whether it leaves a rule behind (a bankruptcy law, a term limit, a posting rule, a published fee) and the institutional conditions under which that rule will actually bind.

Step zero: the straight story

Before any deal, the government needs to tell what one economist calls a “straighter” story: declare an economic emergency, publish the losses in the banking system, replace the 6.5% and 8.5% growth targets with ones the forecasts support, and pass an austerity-minded supplementary budget.23 The leading think tank asks for the same thing in its own words: a documented baseline, a realistic core budget and an integrated reform package presented to Parliament.4

Why this comes first, in incentive terms: every group in the incentive map is deciding whether the government will still mean it in two years. Investors, bankers, officials and cadres all discount promises from a government whose numbers do not add up. Candour is the cheapest credibility there is. Related items in this step:

  • Monetary and fiscal credibility. An independent, predictable central bank and a clear exchange-rate regime. A trusted central bank is the scarce asset: remittances and reserves rose when the previous governor was believed, and a successor seen as the government’s choice risks losing that belief even when decisions are right.5 No stimulus, and no rate cuts from a central bank that people do not trust. (Economists differ on how fast rates should fall; they agree that credibility is the binding constraint.)
  • Energy. The shortage is of gas and money, not plants. Net off the arrears, publish every contract, fix a three-year tariff and subsidy path, restart stalled drilling (offshore gas bids now close on 31 March 2027, after an extension from 30 November 2026), and review the power purchase agreements; idle capacity of 7,700-9,500 MW is reported by the review committee.246 Cut the duties on rooftop solar components, run an efficient-fan programme and finance factory cooling.7
  • Tax before rates. Cut discretionary tax exemptions (estimated at 3.6% of GDP in one source; unverified) and automate collection through e-invoicing and QR-based payment before raising any rates. Phase out blanket fuel and fertiliser subsidies only when targeted cash transfers (paid through one card and mobile money) are ready for the poorest households.8
  • Targets. Replace growth targets that the forecasts do not support (see targets and outlook).

The nine deals

# Deal The group gets The country gets Price and tradeoff Main risk
1 Pay-to-exit for bad debt (large borrowers, bank owners) A one-time settlement window that can resolve civil and financial liability; a legal bankruptcy route Banks that can lend again; dead firms closed; an end to rolling rescheduling Partial civil forgiveness; public money for several years (size disputed, see below) The window is extended again and again; settlement used selectively against opponents
2 Insurance pact (ruling party and opposition) Ruling party: safety after office, a legacy. Opposition: caretaker elections, oversight chairs, upper house by vote share Lower stakes, longer horizons, less violence around elections The same fair-process guarantees cover today’s rulers if they lose power; no immunity Amendment passed in a hollow form; opposition boycotts
3 Pay for accountability (civil service, police) The rest of the 100-142% pay rise; fixed tenures; protection from political transfers Asset declarations; published service standards; postings by rule; digital services; deemed approvals Fiscal cost already committed; risk of strikes Conditions quietly dropped; pay rises without change
4 Cadre-to-career and legal party money (parties, youth wings) Parties: audited public funding. Cadres: apprenticeships, migration finance, open hall lists, elected student roles Less extortion and campus violence; lower business costs Public money for parties and for some people who behaved badly Funding taken and extortion continues
5 2029 export compact (exporters, workers, new investors) Exporters: tariff cuts on inputs, faster ports, gas priority, trade deals with the EU and Japan. Workers: a predictable yearly wage rule Fewer shutdowns like November 2026, labour standards the EU requires, more women hired, new export lines Tariff revenue lost until other taxes replace it; higher labour costs Protected industries block tariff cuts; buyers move orders anyway
6 Small-business compact (small firms, traders, transport) A simple low tax paid by phone; a payment record that builds a credit history; protection from extortion whether or not they register A wider tax base; formal firms that can borrow and grow Lower tax rates at first; the state must deliver protection Protection fails, and the tax becomes one more collection
7 Madrasa livelihoods compact (Qawmi boards, imams) Per-student funding for voluntary add-ons; skilled-migration routes; a transparent imam welfare fund; a seat at the table Employable graduates; access for women and safeguards against coercion; no mob enforcement State entanglement with religion; family-law reform deferred Strengthens Islamist parties; boards take money and refuse testing
8 Barracks for budget (military) A predictable budget, peacekeeping, businesses left in place Out of policing and politics; published accounts for the businesses; no new sectors A military business base stays Budget pressure breaks the deal
9 Migrant and diaspora compact Lower recruitment costs, skills and language training, investment channels Higher remittances per worker; returning skills and capital Short-run loss of talent Gulf demand falls, as it has since the Iran war

The deal numbers are not an order of work. The order is on the roadmap. Deals 3, 5, 6, 8 and 9 are shorter to describe because their design is more settled; deals 1, 2, 4 and 7 are the hardest and are set out in detail first.

The ground rules in each deal

This is the one place where each reform is mapped to the rules it depends on. Rules are named in short form (1 shared prosperity, fairly shared costs; 2 one law for everyone; 3 persuade, never coerce; 4 fair, timely justice, never retaliation; 5 losing power is safe; 6 public trust; 7 open numbers, sunsets, correcting mistakes). Only the changes to implementation are listed; the deal designs above and below already include them.

Deal or reform Depends on What concretely changes in implementation
Step zero: straight story 1, 7 Publish aggregate bank, guarantee and energy-arrears exposure promptly; time institution-level disclosure to protect depositors and financial stability; show who bears the costs by region and group
1. Bad-debt deal 2, 6, 7 Charge owners by the legal loss hierarchy; independent valuations and conflict checks before any bad-bank transfer; settlement of civil liability only, with independent criminal investigation continuing; statutory, costed limits on public support; stay-escrow only with a hardship exemption and judicial discretion41
2. Insurance pact 2, 5, 6 Publish a provision-by-provision map from proposal to enacted text, commencement and pending litigation; hearings, committee scrutiny and transparent appointment standards first; a handover inventory; compare opposition-led committees, caretaker arrangements, Article 70, term limits and a second chamber against both executive dominance and deadlock; prosecutorial decisions stay independent
3. Pay for accountability 2, 6 Pay tied to published service deadlines, written reasons and appeal channels; verification designed before performance pay; no arrest or conviction quotas
4. Cadre-to-career, party money 3, 5, 6 Hall allocations with appeals; apprenticeships open by need and merit across affiliations; lawful student association preserved; lawful fees inventoried and receipted, complainants protected and individual conduct investigated; party funding on neutral eligibility with a fair chance for new entrants
5. Export compact 1, 6, 7 A market-by-market transition calendar; time-limited, audited adjustment support and worker representation; published bids and contracts for ports and zones; residents’ land and compensation rights protected; protection from arbitrary interference applies nationwide, not only inside zones
6. Small-business compact 2, 6 In-person and appeal routes for people who cannot use digital systems; protection from extortion regardless of registration; overlapping fees reduced before any new payment
7. Madrasa compact 3, 1 Voluntary, accredited skills modules under the same grant rules for any eligible provider; students and women included in course design; the right to decline or dissent for individuals; basic protections never depend on community endorsement
8. Barracks for budget 2, 7 Statutory conditions for any domestic military role; civilian audit of budgets and commercial interests, with a narrow basis for confidentiality
9. Migrant compact 6, 1 Verified recruitment charges, contract terms and licensed providers published; accessible complaint channels; diaspora investors get the same approvals as comparable applicants
Courts and justice 2, 4 Victim and witness protection; preserved evidence; independent investigation of credible allegations across affiliations; accountability by individual proceedings, never collective punishment
Protest and assembly 2, 3 Facilitate peaceful assembly (which may disrupt ordinary activity), give reasons for restrictions and act against specific unlawful conduct on evidence42
Energy, heat, health 1, 7 A costed tariff path with targeted support and an appeal route before vulnerable households lose essential service
Education and learning 1, 3 Published learning data; women and students in design; support allocated by open criteria

The rival test applies to every row: would the sponsors accept the same rule, appointment process or charging standard if their strongest rival held power?

The four hardest deals in detail

Deal 1: pay-to-exit for bad debt

Problem. A third of all loans are bad, concentrated in a few thousand borrowers, and each new rescheduling window teaches them to wait. One economist thinks only 10 to 12 of the 61 banks can survive over the long term.9 (The figures are on the banking crisis page.)

Design.

  • An independent banking reform commission with a ten-year mandate, which first publishes the losses bank by bank.9
  • A bankruptcy and restructuring law, so that a failed firm can be closed or sold instead of being kept alive on fresh loans.2
  • One settlement window with a fixed end date written into law. Inside it: repay principal with interest relief, or convert debt to shares that are then sold. Outside it: no new credit, no bank directorships, no state contracts, and cases sent to dedicated benches. Independent criminal investigation continues where evidence warrants; a settlement resolves civil and financial liability and is not immunity. Honest firms hit by shocks should be offered structured restructuring on different terms from wilful non-payers.
  • Court rules that match. Dedicated commercial and bank-recovery benches; a rule that a stay on loan recovery requires an escrow deposit (25-50% of the disputed amount is a proposal), but only with access-to-justice safeguards: a hardship exemption, judicial discretion, review for proportionality and preserved High Court remedies; paid mediation as a new fee stream for lawyers. Commercial-court reform (90-day trials, capped adjournments) is covered under supporting reforms below.
  • An end to rescheduling before default except under the new law.
  • Merge, sell or close the weakest banks, with insiders’ shares wiped out before public money goes in. A public-private asset-management company (a “bad bank”) is one way to hold toxic loans so that healthier banks can lend; it is a structure that the commission could choose, not an alternative to the settlement window. Retail depositors should be protected so that mergers do not trigger runs (a proposal is full protection for small savers; the threshold is unverified).

How much will the clean-up cost?

Estimates range widely and are made on different bases. One economist puts the clean-up at about 1% of GDP a year for several years, citing the central bank governor (Tk 60,000 crore a year) and the Finance Minister ($40bn over several years).3 Another analyst puts the possible cost of the five merged Islamic banks alone at a quarter of government revenue, or nearly 2% of GDP.10 A third source reports a rating agency’s view that recapitalisation would need about 10% of GDP.11 A fourth estimate assumes a recapitalisation bond costing 3-5% of GDP amortised over 10-15 years.8 These are not the same measure (annual flow, total stock, a single group of banks, the whole system), but the headline is the same: the bill is large and cannot be avoided. See disputed figures.

What the Section 18(a) episode teaches. A version of pay-to-exit has already been tried and rejected. In April 2026 the government wrote into law a route for former owners to take their banks back by repaying the state’s money, 7.5% up front and the rest within two years. One analyst defended it as the least bad fiscal option.10 Public, opposition and banking-industry anger forced its repeal by September.12 Four lessons follow.

  • Control cannot go back to the people who caused the losses. Deal 1 is a settlement of borrowers’ debts. Bank owners and directors lose their shares and do not return.
  • The terms must not be payable with more bank money. Critics saw that an owner could pay 7.5% and borrow the rest.13 Settlements have to be funded from outside the banking system, with the source disclosed.
  • It must be done in the open. The clause appeared without the central bank’s knowledge.14 A deal that looks like a private favour dies, and deserves to.
  • The bill is still there. Repeal did not pay it. The honest list of who pays is: owners’ equity first, then recovered assets and new investors, then part-conversion of the largest deposits into shares (already used at Sammilito Islami Bank), and taxpayers for the remainder.15

There is also good news in the episode: the constituency for fairness has real weight. It reversed a law backed by a two-thirds majority within five months.

Why the powerful might accept it. The Finance Ministry needs lending to restart to fund its own budget. The government borrows heavily from banks, which starves private firms. Good firms and depositors gain directly. For a defaulter, a certain civil settlement can beat years of litigation and a frozen firm, provided the window really closes; it does not buy immunity from independent investigation.

The tradeoff to state plainly

This forgives part of what was taken, and the public pays for years. The alternative is not full recovery. It is the present situation, in which about 1% of the top defaulters’ debt has been recovered at the state banks.16 Part of the loss is permanent either way.

Deal 2: the insurance pact

Problem. Power is winner-take-all, so each government strips the checks that might be used against it, and each opposition expects ruin. The interim reforms that would have freed judges, police and the anti-corruption body from the executive were repealed or lapsed.17 As of mid-October no police commission existed, the anti-corruption commission had reverted to its 2004 law with executive appointment, and caretaker restoration and a prime-ministerial term limit were promised only through the planned amendment.18

Design. Use the constitutional amendment that the government’s own committee plans to table in January 2027.19 Put into it the items both sides have already accepted in principle: a prime-ministerial term limit, a caretaker arrangement for elections, a relaxed floor-crossing rule (Article 70) and an opposition deputy speaker. Add opposition chairs for the oversight committees and, by ordinary law, a police commission with a binding shortlist for the top post. A menu of instruments, with a verdict on each from the comparative evidence:

Instrument What it does Verdict
Term limit; caretaker or neutral election management Lowers the stakes of losing; lowers the value of muscle Core of the pact
Opposition chairs for public-accounts and oversight committees; opposition seats on selection committees for judges, the anti-corruption commission and the Election Commission Gives the opposition real stakes and reduces the payoff from winning Core; durable if written into standing orders and statute
Ring-fenced autonomy for the anti-corruption commission, Election Commission, Public Service Commission, Bangladesh Bank, judiciary and auditor-general, with fixed non-renewable terms, cross-party selection and budget lines to parliament The highest-leverage single reform in the electoral architecture Core; harder to reverse if constitutional
Proportional element (upper house by vote share, a national top-up list or mixed-member system) Gives a party with 30% of the vote roughly 30% of the seats Bargaining chip for 2-5 years; the governing party prefers allocation by seats
Super-majority for constitutional change, electoral law and the autonomy of the bodies above Prevents a simple majority rewriting the rules Useful with the above
Constructive no-confidence vote (a government can be removed only by electing a new one) Mid-term accountability without instability Preferred to staggered elections
Staggered local elections; recall at constituency level with high thresholds Continuous feedback without destabilising parliament Local elections are a 0-2 year deliverable because they spread patronage
Shorter term (five to four years) Slightly more responsiveness, slightly less continuity Small effect; not decisive
Staggering half of the lower house every two years Intended to limit how long a poor government can ride out its term Weak: government formation needs a fresh majority; no prominent parliamentary democracy does it

Three further provisions repeatedly recommended: an independent prosecution service, committee-based appointments to constitutional bodies, and an independent vetting and oversight body for the new special-response force.

Why the ruling party might accept it. It spent 2009-2024 on the losing side of a politicised state. The opposition won 31.8% of the vote. Local elections will show how fast incumbency erodes. A party that can imagine losing in 2031 has a reason to want neutral referees while it still has the votes to create them.20 The window is now: the governing party can pass the amendment with its own votes.

Risks. The amendment becomes a facade, as an earlier amendment committee did; the opposition may reject it as illegitimate or boycott it; it could face a basic-structure challenge; the military could re-enter politics given its history in 1975, 1982 and 2007; the barred Awami League could become a destabilising force outside the system.21

The tradeoff to state plainly

The same fair-process guarantees apply to today’s leaders if they lose power, and properly brought proceedings continue: the pact is not immunity. Confidence is low: the governing party’s actions this year suggest it currently values control over insurance. If the amendment fails, a narrower package by ordinary law (a police commission with a binding three-name panel for the top post and protected postings, committee-based appointments, an independent prosecution service) is the fallback.

Problem. The party-machine loop (see circular dependencies).

Design. A package in a few places first: five industrial zones, the Dhaka-Chattogram and Dhaka-Mymensingh highways, and Dhaka’s wholesale markets.

  • Demand side: paid apprenticeships with employers in those zones, public works and migration finance, targeted at men aged 18-30 in the same districts. One source proposes a national programme of 200,000 paid campus and industry placements; the cost and the pipeline are not established.
  • Collections: collections are lawful only with legal authority, for specified services, at transparent cost-based charges with an appeal route, and are collected digitally so that cash at the roadside becomes the exception; compulsory association payments may not masquerade as fees. About 932 associations already collect “donations”; the aim is to cap and expose them.22
  • Enforcement: a victims’ hotline with response-time targets and dedicated fast benches, with mobile units along highways and in wholesale markets.
  • Campus: hall seats by public list or lottery run by professional wardens, not party wings; student-union elections run by the university, with term-limited leaders and protected speech. Apply the same principle to faculty: individual political expression and lawful association by teachers are protected, and the target is coercion, patronage in appointments and misuse of university resources; any organisational restriction on party groups must be necessary, proportionate and open to review. The phrase to use is “separate the party machinery from educational institutions”, not “ban student politics”. At least 33 higher-education institutions have banned party-affiliated student politics since August 2024, and in an online poll of about 350,000 people about 93% favoured ending it.1 The same source warns that new wings are repositioning on campuses.
  • Party money: audited public funding tied to votes won and to published accounts, with donation disclosure, and a condition that armed or extortionist wings are dissolved.
  • Measurement: a quarterly extortion survey of firms and transport operators, published.

Why party leaders might accept it. Cadres are an embarrassment as well as an asset; the governing party says it has acted against several thousand members, and it has dissolved city committees.23 If elections are run neutrally (deal 2), muscle is worth less and delivery is worth more.

The tradeoff to state plainly

Some public money reaches people who extorted. A lawful, cost-based fee replaces a payment that began as coercion, and risks being read as legitimising it unless the authority, service and appeal are real. No rigorous evaluation of this kind of package in Bangladesh has been found, so it should be run as a pilot with honest measurement, not rolled out on faith.

Deal 7: the madrasa livelihoods compact

Problem. A large and growing stream of young men is educated for a labour market whose pay is not measured (a survey of 300 volunteer imams found an average near Tk 10,000 a month, which is not evidence on all graduates), and its leaders can mobilise against gender and curriculum reform.24

Design. Everything voluntary. The madrasa boards are one eligible delivery route among others, with independent accreditation and access for individuals that does not need a gatekeeper’s endorsement.

  • Per-student grants for add-on courses in maths, English, digital and vocational skills (solar installation, accounting, health and logistics are examples), with independently accredited tests recognised by employers.
  • Skilled-migration routes where Arabic is an advantage.
  • A transparent welfare fund for imams and muezzins, tied to community roles they already perform.
  • A standing consultation channel on matters within their field, including joint campaigns against corruption, adulteration and loan default.
  • Equal rules: grants have neutral eligibility for any accredited provider; peaceful disagreement is protected; coercion and exclusion face the same rules whoever practises them. Women’s right to work and learn and every individual’s right to decline or dissent do not depend on community endorsement.

Why the ulama might accept it (a hypothesis the pilot will test). If graduates are indeed poorly paid and insecure (the evidence is one small survey of imams; graduates’ earnings are unmeasured), the offer would answer a real need. It leaves the religious curriculum and governance alone, which was the fear in past reform attempts. It has worked before: public subsidies drew many traditional madrasas into the registered system and into teaching girls.25

The tradeoff to state plainly

It defers liberalising changes that some would like to see, and that deferral is paid for mostly by women. It also puts public money and legitimacy behind religious institutions. The argument for accepting that price is that the freedoms with a measured economic payoff are women’s work, mobility and safety, and those are inside this deal. Compare the dispute on personal-freedom zones in the next section.

The other five deals

Deal 3: pay for accountability

The pay rise (100-142% for 2.4 million employees, phased to July 2027 with allowances from 2028) has no conditions attached yet. Later phases can be tied to asset declarations, published service standards and postings by rule. Four design additions from the wider evidence:

  • Remove the discretion that produces the bribe. Much corruption is structurally produced by systems that require physical presence, paper files, discretion and repeat visits. Digitising land records, trade licences, customs and tax assessment (faceless e-assessment), extending e-procurement (single-bid share was 20.7%26) and introducing deemed approval for standard licences remove the clerk’s leverage without a single prosecution.
  • Pay targeted at the highest-risk posts. Raising pay is most effective where the chance to extract rent is greatest compared with salary (customs, tax, land registration, immigration, traffic police), together with procedure redesign. Doubling a teacher’s salary is good social policy but has little anti-corruption effect.
  • Protect honest officers. Statutory protection against arbitrary transfer, whistleblower protection, and a competitive senior-services pool with lateral entry.
  • Make enforcement visible and hit the top too. Weekly published prosecution data and mandatory published asset disclosure for senior officials.

Top-first shock or pilots and islands?

Comparative cases point in two directions. Singapore began with zero tolerance at the top (investigating a cabinet minister) with pay reform running in parallel; Hong Kong’s anti-corruption commission combined investigation, system prevention and community education with real independence and due process; Georgia dismissed its entire traffic police (about 16,000 officers in most sources, up to 30,000 in others), raised pay roughly tenfold and made enforcement visible, and bribery in routine police contact fell sharply.127 One reading is that gradual tightening is almost never what works when corruption is networked, so the first prosecutions should target former ministers, senior police, judges and top businessmen, including some from the governing coalition (otherwise reform becomes a cycle of political prosecution). The reading we adopt stresses that Georgia’s success came from a dominant executive, that democratic control of the police stayed weak,28 and that Bangladesh’s own history of supermajorities entrenching themselves produced violent cycles; it favours islands of effectiveness (central bank supervision, the large-taxpayer unit, Chattogram port and customs, a few zones, commercial courts) and a police commission with binding rules.29 We adopt the second approach for the first two years because it matches the political window and what the incumbents can accept, and we keep the first as a medium-term test of seriousness: any anti-corruption reform that reaches only the outgoing party is not anti-corruption. This is a judgement call and reasonable analysts disagree.

Deal 5: the 2029 export compact

Reading the 2029 date

The research treats November 2029 as a hard deadline. On the evidence checked, it is the start of a transition clock (about three years for EU access), with the exact dates for other markets unknown. We keep the urgency, because rules of origin, GSP+ eligibility and trade agreements take years to secure, but do not call it a cliff.

The anchor is LDC graduation in November 2029.30 Graduation starts a clock on preferential market access, but it is not itself a single cliff: the EU’s “Everything but Arms” access continues for a three-year transition after graduation (about November 2032 on the stated timetable, subject to the implementing decision), and other markets have their own arrangements that are not set out in the research.31 Urgency still holds, because the fallback must be secured before the transition ends and buyers are already positioning. About 73% of exports benefit from LDC preferences, and the fallback scheme (GSP+) is weaker because Bangladesh’s share of EU preferential clothing imports (about 50%) exceeds a 37% cap.32 The economics are on economy 2026 and targets and outlook. The design:

  • Tariffs. Tariffs on intermediate goods of 18.8% (about 29% with para-tariffs) compare with 7.4% in China.33 Reduce them on a published glide path (a proposal is to cut protective duties by a third a year from 2027 to 2029), with replacement revenue first: VAT-base widening (registrations from 800,000 toward 2 million) and income-tax gains must replace tariff revenue as they fall.
  • Equal export incentives. Extend duty-free bonded access beyond garments to other exporters. One proposal names five priority sectors (leather and footwear, light engineering and electronics assembly, pharmaceuticals, processed agro-food and aquaculture, and IT-enabled services) with illustrative export-share targets (8%, 8%, 7%, 6% and 5%); those targets are aspirational and unverified.
  • Trade deals. A dedicated chief trade negotiator’s office staffed outside normal rotation; EU GSP+ compliance (labour and environmental audits), ratify the Japan EPA, pursue South Korea and ASEAN-area access, and launch EU talks.
  • Zones and ports. Make five zones work rather than 100: reliable gas or power, customs time-release targets, a digital single window with deemed approval within a week, and commercial arbitration courts. One proposal would designate three “autonomous” enclaves on the Chattogram, Dhaka and Cox’s Bazar axes; others propose five zones. We use the common core (a few zones with guaranteed services and legal certainty) and treat the number as open.
  • Ports and process. The foreign-operator question at Chattogram illustrates how process decides whether a good policy survives. Container terminals show a real gap between local and frontier performance (turnaround, dwell time and logistics cost), yet three concerns are legitimate: process legitimacy (a deal started without open competitive tender), strategic geography (a terminal beside a naval base) and labour terms (verbal guarantees are not enforceable terms). The practical lesson is to run open international tenders for future concessions (Bay Terminal, Laldia, Matarbari and Mongla), publish concession contracts in full with performance-based tariffs, ring-fence labour terms, include review windows, carve out terminals next to sensitive assets, and keep regulatory capacity in a reformed port authority.1
  • Wage process and standards. A predictable yearly wage rule and the labour standards the EU requires, including women’s employment targets, as the workers’ side of the deal.
  • Performance-conditioned industrial policy, later. The interim government’s Task Force idea of support to about 1,500 non-garment exporters, conditional on export performance, and a backward-linkage fund for man-made fibre.34 The developmental-state record (conditional support, export discipline, withdrawal when targets are missed) is attractive, but without bank supervision it repeats the bank looting under a new name. It belongs in the 2-5 year window, after supervision and auditable data exist.1

Deal 6: the small-business compact

A flat, simple turnover tax paid by phone, with the payment record building a credit history. A complaint line on extortion with a response-time target must be open to every trader, registered or not, because protection that depends on registration or a receipt is not protection. It succeeds only if the state delivers protection; otherwise it becomes one more collection. Run it as a pilot alongside the extortion survey in deal 4.

Deal 8: barracks for budget

A predictable budget and continued peacekeeping in return for staying out of policing and politics (the army is being withdrawn from street policing).35 Publish the accounts of the welfare businesses and stop expansion into new sectors. Do not try a forced sell-off; Indonesia’s attempt failed.36 Modernise defence transparently and keep party politics out of promotions.

Deal 9: the migrant and diaspora compact

Break recruitment syndicates (reported at several times the legal fee; unverified) through direct government-to-employer digital portals; train workers for care, nursing, IT and technical trades in Japan, Europe and Korea; build investment channels for the diaspora. The skilled share of outflow was 19.1% in 2025.37

Reforms that support the deals

These are not deals with a single counter-party, but several deals fail without them.

  • Courts. Commercial and financial justice first, because it links the judiciary to the binding finance constraint: make the Commercial Court Ordinance 2026 operational (90-day trials, capped adjournments, pre-suit mediation) in Dhaka and Chattogram with published clearance data (whether the ordinance survived parliamentary review is a gap); raise mediation targets for defaulted loans well above the 1% pilot; recruit the 304 judge posts in process; review boards for undertrial prisoners (about 74% of prisoners are awaiting trial); village courts for small disputes; a law against false and vexatious cases. Structural independence (restoring the 1972 form of Article 116, a Supreme Court secretariat, an appointments council) goes into the January 2027 amendment, with a staged version (the secretariat first) as the fallback.38
  • A liberal floor. Replace the Cyber Security Ordinance and stop using the Anti-Terrorism Act against speech; prosecute mob violence and shrine attacks (11 cases for 97 attacks); equal protection for minorities. These improve investor and buyer confidence and are required for EU GSP+ conventions. Respect for freedom of conscience is a principle, and its institutional requirements (narrowing religious-insult provisions, protecting minorities and dissenters, equal personal law, decriminalising consensual adult conduct) are demanding; the pathway recommended is to lead with the economic, educational, women’s-rights and anti-mob points and build the institutions that make harder protections possible later.1
  • Feedback loops. No inclusive settlement survives without a free press and civil society. That means a strong right to information with an independent commissioner, editorial independence for public broadcasters and the national news agency, a narrow incitement and defamation law in place of broad cyber provisions, protection for sources and whistleblowers, and an independent statistics body so that failure is detected early.1
  • Management and professional standards. Industry accreditation, independent inspection, continuous training and evidence-based performance review in public and private organisations.

Personal-freedom zones or no cultural front?

An alternative proposes launching globally competitive investment and lifestyle zones within two to three years, with legal firewalls against moral policing, arbitrary raids and vigilante interference, running at the same time as the madrasa compact. Our recommendation is to secure the liberal floor and women’s economic freedom now, treating wider social liberalisation (alcohol permits, nightlife, weekend alignment) as opportunistic and zonal (tourism enclaves, export zones, diplomatic areas) and not as a national programme in the next five years, because it would hand a mobilising issue to groups whose acquiescence is needed on gender-and-jobs measures. There is no measured growth payoff for the cultural freedoms in the available evidence.38 The two agree on the core: private life should be free from vigilante and arbitrary state interference, and zones can host liberal norms. We follow the second on timing and treat zonal enclaves as an option, not a commitment. This is a judgement.

Long-run investments that run alongside every phase (foundational learning, women’s work, health and nutrition, cooling and power, climate adaptation) are on the roadmap.

What not to do

  • Do not stimulate. Cheaper credit and extra spending leak away until balance sheets are fixed.3 (This includes allowing stayed defaulters access to stimulus funds, which is said to have happened in November 2026; unverified.)
  • Do not keep rescheduling. Each window makes the next default more likely.
  • Do not expel cadres without replacing what they earn. The party’s own move to take expelled members back shows the result.39
  • Do not start state-directed industrial support before bank supervision works. Without discipline it repeats the lending failures under a new name.
  • Do not force the military out of business. Indonesia’s attempt failed.36
  • Do not let the state police private life arbitrarily. Mass summary sentences after a raid on a private party tell investors, professionals and the diaspora that rules are unpredictable.40 This asks the state to stop doing something, not to pass liberalising laws.
  • Do not open a cultural front in the next five years. It would unite religious forces against the economic agenda at the moment their acquiescence is needed.
  • Do not let the IMF be the author. A package the government owns is more likely to be carried out than one it can blame on outsiders.3
  • Do not announce targets nobody believes. See targets and outlook.
  • Do not try a hundred reforms at once. Order and pilots matter more than breadth.

What is uncertain here

  • The cost of each deal is unknown except for the bank clean-up and the pay scale. How many cadres there are and what they earn, what parties spend, and what madrasa graduates earn are all unmeasured.
  • Evidence from Indonesia, Andhra Pradesh, Georgia and Singapore may not travel; apprenticeship and public-works evidence in Bangladesh was not searched.
  • Whether the redesigned bad-debt deal will win consent after the 18(a) episode is untested.

For the strongest objections to the whole plan, see objections and critique.


  1. Our assessment. Comparative case details (Singapore, Hong Kong, Georgia, port case): Unverified: no primary source located. ↩↩↩↩↩↩↩

  2. Jyoti Rahman, The Economy Needs Credibility, Not Optimism, 16 September 2026. ↩↩↩

  3. Jyoti Rahman, A State of Economic Emergency, Counterpoint, 20 October 2026. ↩↩↩↩

  4. CPD, Six Months of the Government, 24 August 2026. ↩↩

  5. Counterpoint, Why the New Governor Won’t Work, 27 February 2026 (figures and the banker’s account are the author’s own; no primary sources given). ↩

  6. Prothom Alo, capacity payments review; Financial Express, offshore bid documents. ↩

  7. IEEFA, Fostering Bangladesh’s energy transition, May 2026; just-style, factory cooling. ↩

  8. Unverified: no primary source located (tax exemptions, bad-bank structure, escrow rule, apprenticeship scale, five sectors). ↩↩

  9. Jyoti Rahman, interview, Prothom Alo, 2 August 2026. ↩↩

  10. Counterpoint, The Bank Bailout Trap, May 2026. ↩↩

  11. Unverified: no primary source located (rating-agency recapitalisation estimate). ↩

  12. BSS, parliament session; bdnews24, Section 18(a). ↩

  13. TBS, critique of 18(a). ↩

  14. Prothom Alo, central bank not informed. ↩

  15. Dhaka Tribune, Bank Resolution Act: cabinet repeals controversial section. ↩

  16. Daily Star, top 20 loan defaulters, 1% recovered at six state banks. ↩

  17. Al Jazeera, Is Bangladesh killing reforms introduced after student-led protests?, 22 April 2026. ↩

  18. Financial Express, a year after July Charter signing, key reforms lie in limbo. ↩

  19. Prothom Alo, constitutional amendment committee. ↩

  20. Ginsburg, Economic Analysis and the Design of Constitutional Courts, 2002. ↩

  21. ConstitutionNet, Mandate deferred; ORF, civil-military equation. ↩

  22. Bonik Barta, transport associations. ↩

  23. Dhaka Tribune, BNP steps up action against controversial leaders; Prothom Alo, dissolved committees. ↩

  24. Study of imams’ income, EconPapers. ↩

  25. Asadullah and Chaudhury, Holy alliances, Education Economics, 2009. ↩

  26. TIB, e-GP main report. ↩

  27. Just Security, Georgia national police modernization (about 16,000 dismissals); some sources give about 30,000 (disputed). ↩

  28. LSE, research online record. ↩

  29. Khan and Roy, Making anti-corruption real, SOAS ACE; Khan (2010), core.ac.uk. ↩

  30. Prothom Alo, LDC graduation deferred to November 2029. ↩

  31. EU Regulation 2026/1395, Article 17(2): a transitional period of three years from the date a country stops meeting the LDC condition. ↩

  32. Daily Star, twin-track approach to retain EU duty-free access. ↩

  33. World Bank, Country Economic Memorandum 2022. ↩

  34. Interim government Task Force, final report, April 2025. ↩

  35. Dhaka Tribune, army to return to barracks. ↩

  36. Human Rights Watch, Unkept Promise, 2010. ↩↩

  37. TBS, overseas employment 2025. ↩

  38. Daily Star, judiciary rollback; Cato, Human Freedom Index 2025; HRW, speech arrests; Daily Star, 97 shrine attacks, 11 cases. ↩↩

  39. Daily Sun, reported plan to take back expelled leaders. ↩

  40. Dhaka Tribune, Bhatara raid: 85 jailed. ↩

  41. Bank Resolution Act, 2026, as amended in 2026 (Act No. 108 deletes section 18A; the 2025 ordinance is the predecessor): bdlaws and its 2026 amendment. ↩

  42. UN Human Rights Committee, General Comment No. 37 on the right of peaceful assembly; ICCPR text. ↩

Go deeper

Where this comes from

Synthesised from 11 research reports, checked against the sources cited above and reviewed before publication. How we work · All sources.