Layer 2 · In depth
Circular dependencies and where to cut them
Bangladesh's problems feed each other in loops, so fixing one alone tends to fail. Ten loops, how they connect, and the links that are cheapest and most promising to cut first.
On this page
- The short version
- Why look for loops at all
- The ten loops
- Loop A. Jobs and skills
- Loop B. Banks and politics
- Loop C. Tax and the state
- Loop D. Winner-take-all politics
- Loop E. The party machine
- Loop F. Norms and women’s work
- Loop G. Exit
- Loop H. Energy, heat and productivity
- Loop I. Confidence of outsiders and skilled people
- Loop J. Patronage and competence
- How the loops connect
- How the three mechanisms interact
- A common chain of reasoning, tested: skills, happiness, prosperity
- Three rules for cutting a loop
- Which link to cut, and how likely it is to work
- The party-machine loop up close
- What the loops imply for the order of work
- What is uncertain here
This page is the detailed reasoning behind the short summary of the loops. It asks one question: why do well-meant reforms in Bangladesh so often fail, and where can a loop be cut so that the change lasts?
The short version
Bangladesh’s problems feed each other in loops. Weak banks starve firms of credit; firms that do not grow create few jobs; jobless young people are recruited into party machines; party machines raise business costs and need money from the banks. Fixing any one link alone usually fails, because the other links pull it back.
Loops can be cut. The rule of thumb on this page is to look for the link that is (a) cheapest to change and (b) held by someone who gains from changing it, and to bring in an outside deadline or outside money when nobody inside the loop will move first. On that rule the best first cuts are the banks, the November 2029 graduation deadline, party finance and extortion, and insurance for election losers. Culture, motivation and education come later in the chain than they first appear.
The ranking of first cuts is our assessment
The choice of banks, the graduation deadline, party finance and election-loser insurance as the best first cuts, and the claim that culture and motivation follow rewards more than they lead them, are our reading of the evidence (the garment-jobs study is the strongest single support). Alternatives are discussed on objections and critique. Reasonable analysts weigh education or religion more heavily.
Nobody in these loops is a villain
Each loop is a set of sensible choices that add up to a poor result. A politician who needs money, a young man who needs an income, a bank officer who is told to lend, a family that buys private schooling: each is responding to the payoffs in front of them. The loops are properties of the system. Changing the payoffs changes the behaviour (see group incentives).
Why look for loops at all
Problems that cause each other cannot be listed and fixed one by one. The most discussed problems (a bureaucracy that is slow to decide, education that rewards certificates, a political order in which the state is the biggest prize, and conservative social norms) can be read as the visible surface of one self-reinforcing equilibrium. The bureaucracy produces graduates who reproduce the bureaucracy; education produces partisans as well as professionals; winner-take-all politics gives every government a reason to capture institutions; and conservative framing can give cover to avoid reforms that threaten established rents. A reform that touches only one corner leaves the other three intact, and they restore the equilibrium.1
In our judgement three mechanisms recur: corruption (rent extraction and patronage), exclusionary norms (hierarchy, restricted voice, gender restrictions) and capability deficits (weak learning and management). They cause and strengthen each other. None is “the” root cause, and none is an unchangeable attribute of Bangladeshis.2 The diagrams below show how.
At the level of national politics and public finance, four loops make reform self-defeating: power and accountability (reforms that constrain the executive need the executive’s consent), money (captured banks fund politics, and politics protects the banks), capability (a weak state cannot collect the taxes that would pay for a stronger one) and norms and jobs.3
The ten loops below describe these in more detail, with their branches.
The ten loops
Each loop is described by what feeds what. Evidence for the individual links is on the economy, banking and what the numbers miss pages.
Loop A. Jobs and skills
Few good jobs → skill is poorly rewarded → the able chase government posts, party posts or visas → firms are short of skilled, ambitious people → productivity stays low → few good jobs.
The evidence suggests the loop is entered from the jobs side. When garment factories opened within reach of Bangladeshi villages, girls stayed in school longer and married later, with a bigger effect than a large cash-transfer programme.4 Meanwhile 31% of young people say their education prepared them “not at all” for work, unemployment is highest among the most educated, and in a 2018 survey 47% of youth with higher-secondary education or more named a government job as their main goal.5 Young people are not unmotivated. They are aiming sensibly at the three things that pay today: a government post, a party connection or a visa.
A first-principles view adds a related pull: the most able science and engineering graduates are drawn to the general civil-service exam because of its prestige, security and administrative power, which leaves factories short of engineers while graduates stamp paper in offices (one source, citing a 2026 study; unverified here).6
Loop B. Banks and politics
Politics needs money → the connected get bank licences and loans → loans are not repaid → banks cannot lend to good firms, and the losses help fund politics → politicians come to depend on the borrowers → no one resolves the banks.
With about half of MPs carrying loans or liabilities, the loop outlived the change of government.7 Four branches hang off it:
- Credit and demand. Banks weakened by bad loans cannot finance productive firms; firms facing uncertain gas and power hesitate to invest; weak investment suppresses jobs; job losses reduce demand; weak demand adds to defaults.8
- Crowding out. Weak revenue forces the government to borrow from banks (81.5% of its financing, on the World Bank’s figures), which crowds out private credit.3 The interest bill then absorbs more revenue, which leaves less to spend on anything else.
- Stay orders. Large borrowers can obtain long court stay orders that delay recovery for years, and lawyers’ fees rise with delay, so the courts add friction to the loop (one source proposes an escrow rule; see strategy and deals).9
- Rolling rescheduling. Each new rescheduling window teaches borrowers that waiting pays better than repaying. (Banking facts are on the banking crisis page.) The cuts are the compact’s rules that public office is a public trust and that numbers are open, with sunset clauses and correction of mistakes (see ground rules).
- Politics needs money
- The connected get licences and loans
- Loans are not repaid
- ✂ Break the loop here: Cut: one settlement window, a bankruptcy law and no new credit for wilful non-payers
- Good firms cannot borrow; losses help fund politics
- Politicians depend on the borrowers
- No one resolves the banks
Loop C. Tax and the state
Low tax → thin budgets, low pay and poor services → those with money buy private schools, clinics, generators and guards → they see no reason to pay tax → low tax. Elites that do not use public schools, hospitals or transport have little stake in the state.10
Two side branches matter. First, with tax under 7% of GDP, the state taxes trade instead of incomes, which protects producers for the home market and penalises exporters who need imported inputs (the same lobby that gains from protection resists direct taxation).11 Second, thin revenue means education and health stay starved (one source gives education near 1.5% and health under 1% of GDP; the education figure is disputed, see the disputed-figures page), which holds down skills and so productivity and so the tax base.6 Naive fixes fail: raising rates mostly pushes more activity into the cash economy or into bribery of tax officials, which is why automating collection and trimming exemptions are proposed before rate rises.
Loop D. Winner-take-all politics
Winning power gives control of police, courts, banks, jobs and contracts → losing is dangerous → whoever holds power takes what it can now and removes the checks that might later be used against it → the prize gets bigger → the next fight is harsher.
The key output of this loop is short time horizons. A ruler who expects to stay, or to leave safely, protects the economy he or she taxes; one who expects to be destroyed on losing strips it (a standard idea in political economy, cited from memory).12 Recent facts fit the pattern. First-past-the-post turned about half the vote into about 70% of the elected seats.13 Parliament reviewed 133 interim-era ordinances: 110 were approved, 7 repealed and 16 lapsed, and the ones lost included those on judicial appointments, the Supreme Court secretariat, police reform and anti-corruption oversight.14 None of this needs a villain. Any government that expects the next government to use the state against it has a reason to keep the state under its own control. The cut is the compact’s rule that losing power is safe, with justice through fair, timely institutions and never retaliation (see ground rules).
One refinement: in 2026 much of the violence is mob killings, custodial deaths and clashes inside and between governing and allied camps, so the model needs a second term. The state as a prize drives conflict within the winning coalition as well as between winners and losers.3
Loop E. The party machine
Parties have regulated legal funding channels (declared donations and grants are permitted), but the evidence suggests that disclosure and enforcement are weak and that informal funding fills the gap; whether lawful income could match costs is untested31 → they rely on businessperson-candidates and on young cadres paid in “territory” (markets, bus stands, tenders, hall seats) → collections raise business costs and firms shrink → fewer jobs → more idle young men → a larger pool of cadres. The jobless young man is the raw material of the machine, and the machine damages the jobs that would pull him out of it. The cuts are the compact’s rules that losing power is safe and that justice is individual, fair and never retaliatory, which lower what muscle is worth to a party (see ground rules). Details are below.
Loop F. Norms and women’s work
Few suitable jobs for women → families see little return to a daughter’s career → early marriage and restricted mobility → fewer women available for work → employers do not design jobs for women. Female participation fell from 42.8% to 38.4% between 2022 and 2024 on the national definition.15 (The ILO definition gives a much lower level, 21.3% in 2022; both are on the disputed figures page.) The loop can be broken from the jobs side, as the garment evidence shows.
Loop G. Exit
Poor prospects and poor governance → the able and the frustrated leave → remittances arrive and prop up consumption and the currency → less pressure for reform, and fewer reform-minded people at home → poor prospects. Remittances are both a lifeline and a sedative: they stabilise the taka while lowering the pressure to fix what pushes people out (our judgement).
Loop H. Energy, heat and productivity
Expensive, unreliable power → little cooling in hot, humid months → lost working days and poorer sleep and health → lower output per worker → less ability to pay for a better grid. Heat cost an estimated 250 million lost workdays and about 0.3-0.4% of GDP in 2024.16 This loop needs a correction to the common story. The shortage is not of power plants: about 29-30 GW is installed against peak demand near 17 GW. The shortage is of fuel and money, and the country pays large capacity charges for idle plants, so more plants would make the loop worse. Cheaper cooling (efficient fans, cool roofs, factory cooling that pays back in about three years) and renegotiated capacity contracts address it.17
Loop I. Confidence of outsiders and skilled people
Visible arbitrary state power or social policing → investors, skilled professionals and diaspora feel less safe → less foreign investment and skilled talent stays away or leaves → fewer high-wage jobs → hardship and resentment. This loop is the most debated.
How strong is the ‘social freedom to investment’ link?
One source argues that moral policing and arbitrary raids directly deter capital and skilled people, citing the summary sentencing of 85 people after a police raid on a private party as a signal of unpredictable rules.18 Another view finds that the event is documented but that the measured growth payoff of cultural freedoms is not established: the economically material freedoms are women’s mobility, safety and access to work, and freedom from mob and state coercion, while nightlife, alcohol rules and similar questions have no measured growth effect in the available evidence.19 We keep the loop as a hypothesis and place the stronger, evidenced part (predictable rules and protection from coercion) in the plan. See culture and religion for the wider evidence.
Loop J. Patronage and competence
Appointments and credit follow connections → weaker service quality → lower public expectations → patronage persists. Capability is both produced by institutions and shapes them: a talented engineer cannot compensate for a procurement process that buys inferior equipment, and a poorly prepared worker improves quickly in an organisation with clear standards and feedback.2 In education the loop shows up as posts and hall seats that are valuable to parties, which helps explain why quality is tolerated.1
How the loops connect
Blue boxes are economic links, violet are political, grey are social. Green boxes are the proposed cuts, and each points to the link it would break. Loops F, G, H, I and J are left out to keep the picture readable.
Economic link
- Loans not repaid
- Losses help fund politics
- No credit for good firms
- Firms do not invest
- Few good jobs; skill poorly rewarded
- Idle young men join party machines
- Low tax take
- Weak services, police and courts
Political link
- Short horizons: take now, remove checks
- State is the prize; losing is dangerous (checks removed, prize grows)
- Banks lend to the connected
- Idle young men join party machines
- Banks lend to the connected
- Firms do not invest
- The better-off buy private services
- Extortion and violence
Social link
- Extortion and violence
- Low tax take
Proposed cut
- Loans not repaid (cuts)
- Few good jobs; skill poorly rewarded (cuts)
- Idle young men join party machines (cuts)
- State is the prize; losing is dangerous (cuts)
- Weak services, police and courts (cuts)
How the three mechanisms interact
The framework can be read as a second map of the same system. It is a hypothesis about interacting pathways, not a measurement of any arrow’s size.
Starting forces
- Patronage, corruption, rent extraction
- Exclusionary norms and enforced orthodoxy
- Climate, density, trade and geopolitical shocks
First effects
- Weaker state capacity and enforcement
- Appointments and credit driven by connections
- Reduced women's participation; high cost of dissent
Capability and output
- Low-quality services and infrastructure
- Lower human capital and organisational capability
- Lower productivity and weaker job creation
Feeds back to
- Poverty, insecurity and dependence on patrons, which strengthens patronage again
- Political and institutional
- Social norms
- Economic
- External shocks
The three mechanisms
- Corruption: rent extraction, regulatory capture, politically directed finance
- Exclusionary norms: hierarchical deference, restricted voice, gender and religious coercion
- Capability deficits: weak learning, poor management, limited technical execution
Pairwise overlaps
- Corruption + norms: clientelism legitimised by status; selective enforcement
- Corruption + capability: patronage hiring; low-quality procurement; unqualified supervision
- Norms + capability: rote learning; suppressed feedback; talent lost through exclusion
All three together
- Weak accountability, low trust, poor adaptation, reduced productivity
A third diagram from the same essay shows how religious teaching becomes economically relevant: through interpretation by people and institutions, shaped by local patriarchy, hierarchy and party politics, which produces laws, expectations and enforcement, which in turn shape participation and capability. Because interpretation and enforcement can change, this is the alternative to cultural determinism. It is developed on the culture and religion page.
Inputs
- Religious teachings and traditions
- Local patriarchy, hierarchy and kinship
- Party politics, state incentives, coercive power
Interpretation
- Interpretation by people and institutions
Enforcement
- Laws, social expectations, enforcement
Outcomes
- Participation, freedom of inquiry, incentives, and so human capabilities
A common chain of reasoning, tested: skills, happiness, prosperity
A common way of describing the problem is as a chain: higher productivity needs skilled people, which needs happy and motivated people, which needs economic and cultural prosperity, which needs productivity.
The chain is real, but one link is weaker than the others, and that is useful. The evidence does not say people must be happy before they will learn and work. It says people invest in skills when skills are visibly rewarded, as the garment-jobs evidence in Loop A shows: the jobs came first, and the motivation and change in family norms followed.4 That is why the plan puts banks, trade and extortion ahead of curriculum reform, even though education matters more in the long run. Foundational learning still gets started early because its payoff is slow (see roadmap).
A caution from the same line of work: hours of work are not the missing ingredient. Bangladeshis work long hours (about 2,383 hours per person engaged in 2023 on the Penn World Table 11.0 series as published by FRED, with comparability limits for informal work), so the shortfall is output per hour: capital, power, skills and management.2 See what the numbers miss.
Three rules for cutting a loop
- Find the link held by someone who gains from cutting it. A reform with no self-interested sponsor does not happen. The Finance Ministry needs working banks to fund its own budget. Exporters need tariff and port reform before 2029. A prime minister who hopes to complete a term and hand over peacefully needs the economy not to collapse.20
- Borrow commitment from outside. When everyone inside a loop gains by waiting, an outside deadline or condition can move first: LDC graduation in November 2029, which starts the clock on preferential market access, an IMF programme, the conventions required for EU GSP+ and the labour commitments in the US framework, and UN peacekeeping vetting.21
- Start a small good loop beside the big bad one. Do not try to reverse the whole system at once. Make five industrial zones, two ports, the commercial courts and the central bank work properly so that productive firms have somewhere to grow. Success there creates constituencies that want more of it. This is the logic of aiming anti-corruption effort at sectors where powerful actors benefit from the rules being followed.22 It also answers a warning from the comparative literature: in thoroughly corrupt systems, fixes that depend on a trustworthy principal fail unless a trustworthy principal exists.23
Which link to cut, and how likely it is to work
| Loop | Weakest link | Who gains from cutting it | Outside anchor | Chance it can be cut within 5 years (judgement) |
|---|---|---|---|---|
| A. Jobs and skills | Demand for skilled labour | Exporters, new investors, youth | 2029 graduation deadline; Japan and EU deals | Moderate |
| B. Banks and politics | Loans not repaid, no exit for dead firms | Finance Ministry, depositors, good firms | IMF programme | Moderate |
| C. Tax and the state | Visible unfairness: big defaulters and evaders untouched | Finance Ministry, salaried taxpayers | IMF programme | Low to moderate |
| D. Winner-take-all | The cost of losing | The governing party itself, later; the opposition, now | January 2027 amendment window | Low to moderate |
| E. Party machine | Cash collections and jobless cadres | Businesses, transport owners, party leaders embarrassed by cadres | Investor and buyer pressure | Low to moderate |
| F. Women’s work | Jobs within reach, safe travel | Exporters, families once income arrives | Buyers’ standards | Moderate |
| G. Exit | Skills and cost of migration; reasons to return | Migrants, their families | Labour deals with Japan, Korea, Europe | Moderate |
| H. Energy, heat | Fuel finance, idle-plant payments, cooling | Finance Ministry, factories, workers | Lenders; buyers’ climate rules | Moderate |
| I. Outsiders’ confidence | Predictable rules, protection from coercion | Investors, professionals, diaspora | Buyer and investor due diligence | Low to moderate |
| J. Patronage and competence | Rule-based hiring and published results | Honest officers and teachers; citizens | Donor-funded learning programmes | Moderate for learning, low for campuses |
Confidence levels
The “chance” column is our judgement, not a measurement. It is lower where a loop’s cut needs the ruling party to give up a source of control (loops C, D, E and I) and higher where a cut matches the interests of exporters, the Finance Ministry or outside lenders.
The party-machine loop up close
This loop gets less attention than banks or tax, yet one mainstream think tank lists extortion among the reasons the recovery has stalled.24 Research on student politics explains why it persists. Each time power changes, the winning party’s student wing takes over the dormitories.25 Students name the shortage of hall seats as their biggest campus problem, which is the hook the wings use.26 One study found that using violence helps a student politician win protection from the party-state; in effect violence works as a job application.27 The pattern is the same whichever party holds office, and a court confession in November 2026 captured the incentive in a sentence: “I am the son of a poor man. I could not resist the temptation of money” (as reported in one source, unverified).28
A governing party that says it has acted against several thousand members, issued 71 central expulsion notices between February and September 2026 and is now reported to be considering taking back more than 500 expelled leaders shows the limit of discipline alone.29 A party that expels its foot soldiers without replacing what they do, or what they earn, ends up taking them back. A ban on one party’s student wing tends to hand the same turf to the next party’s wing unless the transaction itself changes.30
- Elections cost moneylegal funding exists but is thin on disclosure; informal funding fills the gap
- Party needs businesspeople and street muscle
- Cadres paid in territorymarkets, bus stands, tenders, hall seats
- ✂ Break the loop here: Licensed, digital fee collection; hall seats by public list
- Extortion: a private tax
- Higher costs; firms shrink or close
- Fewer jobs for young men
- Party offers income, protection, status
Each link has its own cut: legal party funding with disclosure; neutral election management; licensed digital fee collection and hall seats by public list; fast courts and victim hotlines; apprenticeships and migration finance for the same cohort. The designs are on Deal 4.
Policing alone fails if the cadre has no other income. Jobs alone fail if extortion still pays more. That is why the plan treats this as one package, run first in a few measurable places (see the cadre-to-career deal). A deeper view of the incentives of cadres, parties and students is on the group incentives page.
What the loops imply for the order of work
- First: the links with a sponsor and an outside anchor: banks, energy arrears, the 2029 graduation deadline and the constitutional insurance deal in January 2027.
- Alongside: the extortion package in a few places where it can be measured (five industrial zones, the two main highways, Dhaka’s wholesale markets).
- Then: tax fairness and pay tied to accountability in the civil service, which need the first group to have built some credibility. Cut exemptions and automate collection before raising rates.
- Throughout, but slow to pay off: foundational learning, women’s work, migration skills, cooling and cheaper power.
- Last: arguments about culture. The evidence is that norms follow opportunity more reliably than opportunity follows norms.
The full sequence with dates is on the roadmap.
What is uncertain here
- The arrows in every diagram are hypotheses about interacting pathways. Some are well evidenced (bank lending to connected borrowers, the party-wing pattern on campuses); some are inferred (the exit loop, the outsiders’ confidence loop).
- Nobody measures the total size of extortion or how many cadres there are, and there is no complete account of what parties spend on elections. These unknowns set the cost of the party-machine package (see objections and critique).
- The statistical backdrop is noisy: the IMF rated Bangladesh’s national accounts “C” in January 2026, so level comparisons carry measurement error (see what the numbers miss).
-
Acemoglu and Robinson, Why Nations Fail (2012); Acemoglu, Johnson and Robinson, Institutions as a Fundamental Cause of Long-Run Growth. ↩↩
-
FRED, annual hours worked per person engaged, Bangladesh (Penn World Table 11.0, Groningen Growth and Development Centre and UC Davis; the 2023 value is 2,383.18). ↩↩↩
-
Prothom Alo on World Bank, FY26 outlook (government bank borrowing 81.5% of financing); TBS on ASK, violence in the first half of 2026. ↩↩↩
-
Heath and Mobarak, Manufacturing growth and the lives of Bangladeshi women, NBER. ↩↩
-
Financial Express on the SANEM survey, youth support in Bangladesh (31% “not at all”); Bonik Barta, graduate unemployment; BRAC, 2018 youth survey (47%). ↩
-
Unverified: no primary source located (education spending levels and the civil-service exam pipeline). ↩↩
-
Dhaka Tribune, TIB: 79.46% of MPs in the 13th Parliament are crorepatis (half carry loans or liabilities totalling Tk 11,356 crore). ↩
-
SANEM’s Mahtab Uddin, as quoted in press coverage. Unverified: no primary source located. ↩
-
Unverified: no primary source located (the 25-50% escrow proposal; the 4.855 million case backlog is attributed to the Supreme Court and Daily Star, November 2026, without a link). ↩
-
Jyoti Rahman, interview, Prothom Alo, 2 August 2026. ↩
-
Interim government Task Force, final report, April 2025. ↩
-
Olson (1993), cited from memory in the analysis; not re-checked. ↩
-
Wikipedia, 2026 general election (49.97% of votes, 209 of 300 elected seats); see also the disputed-figures page for vote and seat counts by counting stage. ↩
-
Al Jazeera, Is Bangladesh killing reforms introduced after student-led protests?, 22 April 2026. ↩
-
Jyoti Rahman, A State of Economic Emergency, Counterpoint, 20 October 2026, citing official data; ILO-definition figure: UN Women, State of the Bangladesh economy from a gender lens. ↩
-
World Bank press release, Bangladesh faces health and economic risks from rising temperature, 16 September 2025. ↩
-
IEEFA, Fostering Bangladesh’s energy transition, May 2026; just-style, factory cooling; IEEFA, efficient fans. ↩
-
Counterpoint, Zia Hassan profile (article not readable by the analysis); event: Dhaka Tribune, Bhatara raid: 85 jailed. ↩
-
Cato Institute, Human Freedom Index 2025 (personal-freedom score 5.16). ↩
-
Jyoti Rahman, The Politics of Synthesis 2.0, Counterpoint, 1 June 2026. ↩
-
Daily Star, twin-track approach to retain EU duty-free access; USTR, US-Bangladesh framework fact sheet. ↩
-
Khan and Roy, Making anti-corruption real, SOAS ACE. ↩
-
Persson, Rothstein and Teorell, Lund University record. ↩
-
CPD, Six Months of the Government, 24 August 2026. ↩
-
Kuttig, LSE South Asia blog, 2019. ↩
-
IRI, focus groups. ↩
-
Suykens, “A hundred per cent good man cannot do politics”, Modern Asian Studies. ↩
-
Unverified: no primary source located (court confession, November 2026). ↩
-
Dhaka Tribune, BNP steps up action against controversial leaders; Daily Sun, reported plan to take back expelled leaders. ↩
-
Representation of the People Order, Article 90F (permits registered parties to receive donations and grants within limits); TIB, Institutional governance of political parties, executive summary, 2026. ↩
Go deeper
- Key statisticsData & sourcesThe important figures from the research, grouped by theme in tables, each with its value, date, source and link. Where sources disagree, see the disputed-figures ledger.
- Disputed figuresData & sourcesThe ledger of figures where our sources disagree, with both values, the likely reason and how this site handles each, plus a list of known gaps.
- SourcesData & sourcesThe consolidated bibliography of every external source cited in our research, grouped by type and deduplicated, with the links as given.
Summarised in
Where this comes from
Synthesised from 9 research reports, checked against the sources cited above and reviewed before publication. How we work · All sources.