Thin Capital: The Smallest Cushion Under the Largest Loan Problem

Executive Summary. Bangladesh's banking system runs the thinnest capital buffer and the heaviest declared loan losses of any economy in its comparison group. Bank capital was 2.29% of assets in 2024, last of six Asian economies and about 27% of the peer median of 8.55%. Non-performing loans stood at 9.57% of gross loans in 2023, the highest of the six and 2.85 times the peer median of 3.36%. The ratio of the two is 4.17, which is not a comfortable number in any accounting convention. The lending side has already responded: domestic credit to the private sector peaked in 2015 and had fallen to 34.5% of GDP by 2025, fourth of the five countries reporting that year. And the price of credit leaves nothing to rebuild with. The real interest rate was 0.63% in 2023, the lowest of the four countries reporting, against a peer median of 7.01%. A banking system this thin does not have to fail to impose a cost. It only has to stop lending, which is what the credit series shows it doing.

The buffer and the problem, side by side

Bank capital to assets is the plainest solvency measure there is: what fraction of the balance sheet is owned rather than owed. It is the money that absorbs losses before depositors, the deposit insurer, or the treasury do.

In 2024 that fraction was 2.29% in Bangladesh. The comparison group: Indonesia 13.44%, Thailand 11.13%, Vietnam 8.55%, India 8.41%, Pakistan 4.98%. Bangladesh is last of the six, and the distance is not a rounding matter. The peer median is 8.55%, so Bangladeshi banks hold roughly 27% of the median comparator's cushion, a gap of 6.26 percentage points of the balance sheet.

Source: World Bank World Development Indicators, bank capital to assets ratio (FB.BNK.CAPA.ZS), 2024.

Two honest qualifications before this figure is put to work. First, the WDI series is an unweighted leverage measure, book capital over total assets, not the risk-weighted capital adequacy ratio that supervisors publish. The two are different instruments and the regulatory number is normally the higher of the pair. Second, the 2024 reading is a sharp break from the same series' own recent history, which ran materially higher through 2023. A fall of that size in a single year usually reflects a change in what is being recognised, tighter loan classification or provisioning against exposures that were previously rolled forward, at least as much as a change in the underlying balance sheets. Neither qualification changes the ranking: the series puts Bangladesh last either way.

Bad loans are a multiple of the cushion

The other side of the ledger points the same direction, from an independent series and a different year.

Non-performing loans were 9.57% of gross loans in 2023, the highest of the six. India 3.36%, Pakistan 6.63%, Indonesia 1.96%, Thailand 2.76%, Vietnam 5.41%. Bangladesh sits 6.21 percentage points above the peer median of 3.36% and carries 2.85 times the median share of bad loans.

Source: World Bank WDI, bank nonperforming loans to total gross loans (FB.AST.NPER.ZS), 2023.

Divide the loan-loss share by the capital share and you get 4.17: the fraction of the loan book that is non-performing is more than four times the fraction of the balance sheet held as capital. Read that carefully, because it is easy to over-claim. Loans are not the whole of assets, provisions already booked offset part of the exposure, and collateral recovers some of the rest, so this is not a literal statement that bad loans exceed capital fourfold. The two readings are also a year apart, 2023 for loans and 2024 for capital. What survives all of that is the direction and the order of magnitude: the declared problem is a large multiple of the declared buffer, and no other country in the group is close to that configuration.

None of the World Bank series used here carry provisioning or coverage data, and that absence is the single biggest thing this comparison cannot settle. Provision coverage is precisely the figure that would convert 4.17 from a striking ratio into a measured net exposure, and it is not in the indicator set.

Credit stopped deepening a decade ago

A thin, loss-laden banking system eventually does the arithmetic and lends less. The credit series is consistent with that.

Domestic credit to the private sector was 34.5% of GDP in 2025, fourth of the five countries reporting that year (Vietnam has no 2025 observation, so this comparison rests on five, not six). India 44.03%, Indonesia 36.28%, Thailand 143.1%, Pakistan 10.73%. Among the five reporting, only Pakistan sits lower than Bangladesh. Against the peer median of 40.15%, Bangladesh is 5.65 percentage points short, at about 86% of the median.

Source: World Bank WDI, domestic credit to private sector (% of GDP, FS.AST.PRVT.GD.ZS), Bangladesh against comparators, 1990-2025.

The long arc matters more than the level. Private credit rose from 21.78% of GDP in 2000 to 34.5% in 2025, a gain of 12.72 percentage points, taking it to 1.58 times its 2000 level, or 58.43% more credit per unit of output across a quarter century. All of that net gain was earned before 2015, and part of it has since been given back. The series peaks in 2015 and has drifted down since, so a country whose economy grew 4.22% in 2024, and which now has 173.6 million people and a US$450 billion economy, is intermediating a smaller share of its output through banks than it did a decade ago.

Three confounders before this is read as pure credit rationing. The ratio's step down in 2016 coincides with a large upward shift in the level of measured GDP in the same source, so part of that break is a denominator effect, not loans disappearing. Credit that migrates to non-bank finance, capital markets, or informal lenders leaves this series without leaving the economy. And a falling credit-to-GDP ratio can be healthy deleveraging after an overhang rather than a supply constraint. None of those explanations is testable inside a six-country panel, and this essay does not claim otherwise. What the data does establish is a level and a turning point: at 34.5% of GDP, bank credit is a smaller part of the Bangladeshi economy than the peer median, and smaller than it was at the 2015 peak.

The price of credit rebuilds nothing

Banks with thin capital have two ways back: raise new equity, or retain earnings. The interest rate data speaks to the second, and it is discouraging.

The nominal lending rate was 7.57% in 2023, third of the four countries reporting, below Vietnam's 9.32% and Indonesia's 8.93%, above only Thailand's 4.29%, and 1.36 percentage points under the peer median of 8.93%. So borrowers were not paying an unusually high headline rate.

Deflate it and the picture changes completely. The real interest rate was 0.63% in 2023, the lowest of the four countries reporting, against a peer median of 7.01% and a gap of 6.38 percentage points. Bangladesh's real rate was roughly 9% of the median comparator's. Indonesia earned 7.29% in real terms, Vietnam 7.01%, Thailand 3.0%.

The mechanism is not subtle. A lending rate that is barely positive in real terms means the loan book generates almost no real return, and a bank earning almost nothing in real terms cannot rebuild capital out of retained earnings, however long it is given. Combine that with a non-performing share of 9.57% and the recapitalisation path narrows to new equity, from shareholders or from the state.

The confounder is important and cuts against over-reading this. The WDI real rate deflates the lending rate by the GDP deflator, so a near-zero reading in 2023 reflects the price level that year as much as the pricing of loans. Only four of the six countries report the series, and one high-inflation year is not a policy regime. The nominal series is the more stable claim: at 7.57%, credit in Bangladesh was not expensive by group standards in 2023.

The branch network is not the binding constraint

It is tempting to read all of this as a reach problem, a system that has not built out far enough. The physical footprint says otherwise.

Bangladesh had 9.14 commercial bank branches per 100,000 adults in 2024, fourth of the six, 2.01 below the peer median of 11.15 and at about 82% of it. That is mid-pack, not an outlier: India 14.62, Indonesia 11.23, Pakistan 11.15, Thailand 8.02, Vietnam 3.097.

Vietnam is the instructive case: it runs the thinnest branch network of the six, 3.097 per 100,000 adults, and still holds capital of 8.55% of assets, exactly the peer median, with non-performing loans of 5.41%. Branch density and balance-sheet strength are not the same problem, and in this group they do not move together. Bangladesh's constraint is not how many doors the banks have. It is what has been lent through them and what stands behind it.

What would change this conclusion

A specific, checkable test. If, by the 2030 WDI vintage, Bangladesh's bank capital-to-assets ratio has recovered to the current peer median of 8.55% while the NPL ratio has fallen below the current peer median of 3.36%, with no loosening of classification rules in between, the argument here is wrong and the 2024 reading was a recognition artefact that the system absorbed. If instead the NPL ratio is still near 9.57% and capital is still nearer 2.29% than 8.55%, the decade will have been spent running a banking system on a cushion smaller than the losses it has already declared.

Three moves, with owners and signals.

1. Recapitalise against losses already recognised, before writing the next tranche of credit. Capital is the constraint on lending capacity, so an undercapitalised system rations credit whether or not anyone instructs it to. Owner: Bangladesh Bank, with the Financial Institutions Division on the ownership question. Success signal: bank capital to assets above Pakistan's 4.98% in the next WDI release, which would move Bangladesh off the bottom of the six without yet approaching the peer median of 8.55%.
2. Let the loan book earn a positive real return. A real lending rate of 0.63%, the lowest of the four reporting, closes the retained-earnings route to recapitalisation entirely. Owner: Bangladesh Bank. Success signal: the WDI real interest rate moving off 0.63% and holding above Thailand's 3.0% across two consecutive vintages.
3. Publish provision coverage as a standing series. The 4.17 ratio in this essay cannot be netted down without it, and it is absent from the World Bank indicators used here. Owner: Bangladesh Bank, in the regular banking statistics release. Success signal: a published coverage ratio that lets any outside analyst compute net non-performing exposure against capital without guessing.

The counterargument

The strongest objection is technical and it is a good one. The WDI capital-to-assets ratio is crude leverage, not risk-weighted capital adequacy; the 2024 Bangladesh figure is a single-year break in a series that ran far higher through 2023; and national accounting conventions for what counts as capital and what counts as non-performing differ enough across six countries that comparing to two decimal places is false precision. On this reading, Bangladesh looks worst partly because it recently started counting honestly, and a country that recognises more bad debt scores worse than one that recognises less.

Three answers. First, the two indicators are independent. Capital and non-performing loans come from different constructions and different years, 2024 and 2023, and both put Bangladesh at the extreme of the group, last on capital and first on bad loans. One artefact does not produce two.

Second, if the fall in measured capital was driven by tighter recognition, that is not a reason to discount the level. It is the level becoming visible. A buffer whose true size is only now being reported is not less alarming for having been reported late.

Third, the concession the objection earns. If the 2024 capital print is a definitional break rather than an economic one, then 4.17 overstates the true relationship between losses and buffer, and nothing in this data settles by how much. That is exactly what a published provision-coverage series would answer, and its absence is why the objection cannot be closed here. Six countries is a comparison, not a sample, and none of the associations in this essay identify a cause.

---

Data sources: World Bank World Development Indicators, retrieved from the BDPolicyLab data lake, 2026-08-10. Comparators are India, Pakistan, Indonesia, Thailand and Vietnam; "peer median" is the median of the comparators reporting in the stated year, excluding Bangladesh. Years differ by indicator because coverage does: bank capital and branch density are 2024, non-performing loans and both interest rates are 2023, private credit is 2025. Credit to the private sector reports for five countries in 2025 and both interest rate series for four countries in 2023; every comparison drawn from them is stated on that basis.

Sources

1. World Bank WDI, Bank capital to assets ratio (%), FB.BNK.CAPA.ZS: https://data.worldbank.org/indicator/FB.BNK.CAPA.ZS
2. World Bank WDI, Bank nonperforming loans to total gross loans (%), FB.AST.NPER.ZS: https://data.worldbank.org/indicator/FB.AST.NPER.ZS
3. World Bank WDI, Domestic credit to private sector (% of GDP), FS.AST.PRVT.GD.ZS: https://data.worldbank.org/indicator/FS.AST.PRVT.GD.ZS
4. World Bank WDI, Lending interest rate (%), FR.INR.LEND: https://data.worldbank.org/indicator/FR.INR.LEND
5. World Bank WDI, Real interest rate (%), FR.INR.RINR: https://data.worldbank.org/indicator/FR.INR.RINR
6. World Bank WDI, Commercial bank branches (per 100,000 adults), FB.CBK.BRCH.P5: https://data.worldbank.org/indicator/FB.CBK.BRCH.P5