Three Evidence Chains on Net Interest Margin: Debunking Bank Windfall Profits

This framework uses three mutually independent evidence chains — NIM (net interest margin) definitional disambiguation, revelation of selection bias in bank shares, and restoration of NPL ratio accounting conventions — to refute layer by layer the popular narrative of “bank windfall profits / monopoly-driven high financing costs,” arriving at the conclusion: “Under correct definitions and a full sample, the evidence chain for this popular judgment does not hold.”

The Framework As It Stands

This section is compiled from the research draft: the original framework’s structure, terminology, and key formulations are preserved, including editorial bridging and supplementary factual notes; diagrams are drawn by the compiler following the structure of the original text.

Core Agenda and Three Sub-Themes

On the surface, this framework answers whether “banks have the incentive or room to extend financing to small and micro enterprises”; in substance it is a profitability-side empirical refutation: the three popular claims — “banks rake in windfall profits, earning tens of billions per day,” “bank monopoly causes high financing costs,” and “bank NPL ratios are so low they look like fabricated accounts” — are each dismantled at the level of indicator definitions and statistical conventions, demonstrating that they all stem from definitional misreadings and sample/selection biases.

Main-line judgments:

NIM convention — Net Interest Margin (NIM) ≠ lending–deposit spread; it is the most accurate indicator of the cost–revenue relationship of banks. After interest-rate liberalization, the protection of a fixed spread disappeared, and commercial banks’ NIM has been declining continuously since 2013 to 2.2–2.3% (joint-stock banks at approximately 2%); in international comparison China is at or below the midpoint — even below the United States — and the sector as a whole has never exceeded 3%; with approximately 4,000 licensed banking entities, “monopoly-driven high financing costs” is hard to sustain.
The bank-share paradox — “Media cry windfall profits yet nobody buys bank stocks” is a statistical selection bias (buyers say it’s expensive; sellers say margins are thin), while bank dividend yields of approximately 4% were equivalent to bank wealth-management products at the time, making them a stable-income vehicle for low-risk-preference investors.
NPL ratio convention — The NPL ratio is “the balance of non-performing loans after write-offs and disposals / total loans”; the denominator (loan growth of 13%+) inflating and the numerator (disposals and write-offs) shrinking combine to compress the ratio to 1.7–1.8%; restoring the written-off and disposed NPLs would make the actual ratio significantly higher.

Sub-theme A — Three Independent Evidence Chains Jointly Refute “Bank Windfall Profits / Monopoly”

The entire framework uses three mutually independent sets of evidence (NIM, bank shares, NPL ratio) pointing to one conclusion — banks do not have the legendary “windfall profits,” and their profitability is not outlandish under international and full-sample conventions. This answers, from the profitability side, whether “banks have the room and incentive to provide financing to (small and micro) enterprises”: given no windfall profits or monopoly, the blame for “high financing costs” should not be laid simply at the banks’ door.

Sub-theme B — Methodological Main Line = Return to Indicator Definitions and Conventions

The three claims look different on the surface but are all problems of definition. The NIM claim misreads “spread” — the distinction between the lending–deposit spread and NIM must be drawn; the international comparison misreads “nominal numbers” — high NIM in Brazil and Russia reflects high inflation, a different unit of measurement; inflation must be removed. The NPL claim misreads “book figures” — the key phrase is “after write-offs and disposals”; the book NPL is the number after cleaning. The logical parallel to the definitional disambiguation in Total Social Financing and Broad Money M2: Accounting Definitions and Divergence: indicator numbers are meaningless in isolation; definitional conventions are the starting point of any judgment. Rule of judgment: when encountering the claim that “some indicator is abnormally high/low,” first ask “are the definitional convention, sample convention, and unit of measurement comparable?”

Sub-theme C — Selection Bias Runs Through the Entire Lecture

The systematic divergence between buyers and sellers on “is it expensive?” (buyers mostly say yes; sellers say margins are too thin to cover costs) explains why “media cry windfall profits” and “investors don’t buy bank stocks / dividend yield stable” coexist; the salad analogy (fat people eat salad to control calories) shows the need to understand causation beneath appearances and to remove sample bias; the same tool also cautions that NPL numbers must be restored by including the disposed-of/written-off sample.

The value of this framework lies in upgrading the emotionally charged narrative of “are banks rapacious / do they have room to serve (small and micro) enterprises” into a refutation methodology of “identifying indicator conventions → restoring to international and full-sample benchmarks → removing selection bias → grounding in data” — the conclusion is not that “banks are innocent,” but that “the evidence chain for windfall profits / monopoly does not hold under correct definitions.”

Thesis Distillation

1. NIM convention: NIM ≠ lending–deposit spread; it is the most accurate indicator of the cost–revenue relationship of banks.

Lending–deposit spread = lending rate minus deposit rate (e.g., lending 4%, deposits 2%, spread 200bp); Net Interest Margin (NIM) = the ratio of the difference between total interest income and total interest-bearing cost to interest-earning assets. Because banks do not deploy all funds as loans (some is invested in bonds and so on, where the margin is lower than on loans), NIM is typically smaller than the lending–deposit spread; NIM is the most accurate indicator of the cost–revenue relationship of banks.

2. Historical trend of NIM: After interest-rate liberalization, the protection of a fixed spread has vanished; NIM has been declining continuously since 2013.

Before 2010, the widely discussed issue was that banks were under administrative protection, with both lending and deposit rates set directly by the state, leaving NIM relatively wide. Since then, both lending rates and deposit rates have been fully liberalized, and the fixed spread protection no longer exists. The outcome of liberalization: commercial banks’ NIM has been declining overall since 2013. At its peak it was approximately 2.7%–2.8%; by the time of the lecture it had fallen to 2.2%–2.3%, with joint-stock banks even lower at approximately 2% and in certain periods only a little above 1%.

3. International comparison: China’s NIM is at or below the midpoint — even below the United States — and has never exceeded 3% across the sector; “monopoly-driven high financing costs” is hard to sustain.

Comparing major global economies (2008 to the lecture point), China is at or below the midpoint — neither the highest nor the lowest; far below Brazil and Russia — these countries have high inflation, high nominal interest rates, and naturally higher NIM figures (when inflation is 15%, lending rates are essentially above 15%, a different unit of measurement). China’s NIM is even below that of the United States, where the banking sector has recovered to a sustainably profitable level with NIM above 3% after full competition. China’s sector-wide NIM has never exceeded 3% since data have been recorded; in 2009 China was roughly in the middle of nearly 200 economies globally, and in 2016 it was also basically in the middle of the G7 and BRICS countries in World Bank statistics. Conclusion: the NIM does not reveal anything particularly outlandish about China’s financial sector; with approximately 4,000 licensed banking entities, it is very hard to ensure uniform behavior or the absence of competition, so the claim that “bank monopoly drives high financing costs” is difficult to sustain.

4. The bank-share paradox and selection bias: “Media cry windfall profits yet nobody buys bank stocks” reflects a systematic bias between buyers and sellers.

A paradox: media report daily that banks rake in windfall profits, earning tens of billions a day, yet investors do not buy bank stocks, even though banks pay out a dividend payout ratio of at least 30% or more. This is a manifestation of the statistical concept of “selection bias”: ask buyers “is it expensive?” and most will say yes; ask sellers “is it expensive?” and they say margins are so thin they barely cover costs. Removing selection bias has been a core challenge in statistics and econometrics (scholars have even been awarded prizes for it). The salad analogy: seeing an overweight person eating salad and concluding “salads make people gain weight” — in fact, they eat salad because they are already overweight and trying to control calorie intake; slim people are happily eating meat. Understanding the causation behind appearances and removing sample selection bias is very important.

5. Actual bank dividend-yield performance: approximately 4%, equivalent to bank wealth-management products at the time — a stable-income vehicle for low-risk-preference investors.

From a dividend-yield perspective (treating bank shares as bonds held for income, assuming the share price is broadly stable): the dividend yield on all A-share stocks is approximately 2%, above the overnight deposit rate; from 2011 to the lecture point, the dividend return on the Big Five banks was never below 4%, approaching 6% in 2013 and approximately 5% recently; joint-stock banks also offered 3%–4%; all A-share-listed banks in recent years have a dividend yield of approximately 4% (3% in lower years), equivalent to the return on bank wealth-management products at that time. For investors with lower risk appetite seeking stable income, bank shares are a good choice, with the additional prospect of capital gains if the share price rises.

6. True reading of the NPL ratio: the “after write-offs and disposals” definitional trap — denominator inflation and numerator reduction acting together to push the ratio down.

Background of the skepticism: China’s economic growth rate fell continuously from 14.2% in 2007 to less than half that by the lecture point, during which time enterprises went through zombie-clearing, capacity reduction, and waves of bankruptcies, yet the rise in banks’ NPL ratios appeared limited, prompting suspicion of “fabricated accounts.” The key lies in the definition: NPL ratio = balance of non-performing loans after write-offs and disposals / total loans; write-offs mean using provisions to absorb NPLs; disposals mean selling to bad-asset management companies; the NPL balance on the books is a number that has already been cleaned up. Two compressing factors: first, the denominator keeps expanding — loan growth of 13%+ per year mechanically lowers the NPL ratio even if the numerator stays flat; second, the numerator is actively reduced — in 2018 commercial banks disposed of nearly CNY 2 trillion in NPLs, in 2017 CNY 1.4 trillion, and since 2006 cumulative write-offs have totaled trillions, while the current book NPL balance still stands at over CNY 2 trillion. The current commercial bank NPL ratio is approximately 1.7%–1.8%; there was one notable jump when the banking regulator required all loans overdue more than 90 days to be classified as non-performing, with rural commercial banks showing the biggest change. The low NPL ratio is the combined result of denominator inflation and numerator reduction; if all disposed-of and written-off NPLs were fully restored, the actual NPL ratio would be significantly higher than the current level.

Key Data Anchors (Data point circa 2018/2019)

Indicator	Value
Lending–deposit spread example	Lending 4%, deposits 2% → spread 200bp
Commercial bank NIM	Peak 2.7–2.8% → lecture point 2.2–2.3% (joint-stock banks ~2%, some periods only slightly above 1%)
China sector-wide NIM ceiling	Has never exceeded 3% since data recorded
U.S. banking sector NIM	Above 3% (after full competition)
Number of licensed banking entities	~4,000
Bank dividend payout ratio	At least 30%+
Overall A-share dividend yield	~2%
Big Five bank dividend return	2011 onward minimum 4%+, 2013 approaching 6%, recent ~5%
A-share-listed bank dividend yield	~4% (3% in low years), ≈ bank wealth-management products at the time
Economic growth rate	2007: 14.2% → lecture point: less than half
NPL disposal / write-off	2018 disposal near CNY 2 trillion, 2017 CNY 1.4 trillion; book balance still over CNY 2 trillion
Current NPL ratio	1.7%–1.8%
Annual loan growth	13%+

Reasoning Structure

flowchart TD
  Q["Title question: Do banks have room/incentive to finance small and micro enterprises?<br/>Profitability-side refutation of 'windfall profits / monopoly-driven high financing costs'"]
  Q --> A["Evidence chain 1: Net Interest Margin (NIM)"]
  Q --> B["Evidence chain 2: Bank shares"]
  Q --> C["Evidence chain 3: NPL ratio"]
  A --> A1["Convention: NIM ≠ lending–deposit spread; most accurate indicator"]
  A --> A2["Trend: Post-liberalization → continuous decline from 2013 to 2.2–2.3%"]
  A --> A3["International: at or below midpoint / below U.S. / never exceeded 3%"]
  A3 --> A4["~4,000 licensed banking entities → 'monopoly-driven high financing costs' hard to sustain"]
  B --> B1["Paradox: media cry windfall profits yet nobody buys bank stocks"]
  B1 --> B2["Selection bias: buyers say expensive / sellers say margins thin"]
  B --> B3["Dividend yield ~4% ≈ bank wealth-management products at the time; stable-income vehicle"]
  C --> C1["Definitional trap: 'after write-offs and disposals' balance / total loans"]
  C1 --> C2["Denominator inflation (loan growth 13%+)"]
  C1 --> C3["Numerator reduction (disposals and write-offs of trillions)"]
  C2 --> C4["Compressed to 1.7–1.8%; actual rate significantly higher when restored"]
  A4 --> Z["Meta-framework: identify conventions → restore to international/full-sample → remove selection bias → ground in data"]
  B3 --> Z
  C4 --> Z

Transferable Rules of Judgment

Align definitions before comparing magnitudes: NIM ≠ lending–deposit spread; NPL ratio contains “after write-offs and disposals”; nominal NIM contains an inflation component — any cross-entity / cross-country comparison must first unify definitions and units of measurement.
Restore to international and full-sample benchmarks: Looking at a single domestic indicator as “high/low” is meaningless; position it within major global economies and the full-sector sample (at or below midpoint / never exceeded 3%).
Remove selection bias: Buyers vs. sellers, listed vs. unlisted, disposed-of vs. remaining on books — the sample itself has systematic biases; restore before drawing conclusions.
Ground in data details: Replace the emotional judgment of “windfall profits / monopoly” with verifiable numbers such as dividend payout ratios, dividend yields, and disposal/write-off volumes.

Compiler’s Perspective

Coordinates: Category = Banking and Real Estate / axis_h = Fa / axis_v = Why It Is So

Connecting-layer: The sharpest edge of this framework is not any individual number but the introduction of the “selection bias” tool. Those who use old definitional conventions to judge banks typically make the following move: see media reports of “banks earning tens of billions per day” → conclude “windfall profits” → infer “banks have room to concede profits.” This chain of inference goes wrong in three places: ① looking at absolute profit rather than NIM on an international coordinate (units are not comparable); ② “those who cry expensive” are all buyers; sellers (bank management, investors) and their judgment of bank-share value are systematically excluded (selection bias); ③ the NPL ratio of 1.7%–1.8% looks at the residual after the denominator has been diluted by 13% growth and the numerator reduced by disposals and write-offs; treating it as “actual credit health” is a misreading. Three definitional errors stacked together produce the “windfall profit” sensation — the unique contribution of this framework is to string the three errors together simultaneously with one “definitional disambiguation” methodological chain, rather than correcting each in isolation.

The NIM of 2.2–2.3% at the 2018/2019 data point is consistent with the direction of overall change in bank assets and liabilities described in The Structure of China’s Total Social Financing: interest-rate marketization has continuously compressed the lending–deposit spread, and banks are not passive beneficiaries of windfall profits but are instead under pressure from the progressive disappearance of spread protection.

Sources

Compiled draft z-0117 · collected 2026-07 Data point approximately 2018/2019

Z-Finance

Explorer