Predictors of Loan Success and Default in Chinese Microfinance Portfolio

In the eastern Chinese city of Ningbo, a privately run microfinance lender spent five years quietly approving and rejecting thousands of loan applications by hand, one folder of paperwork at a time. That unglamorous process has now produced one of the most detailed portraits ever assembled of China’s offline informal lending market, and the picture it reveals is striking. Researchers who analyzed the lender’s full portfolio found evidence of discrimination against female and divorced borrowers, discovered that business owners borrow four times more money than other applicants yet default at half the rate, and showed that the rules lenders use to decide who gets money differ sharply depending on whether the loan is for a business or a household. The findings, drawn from more than 10,000 real applications, pull back the curtain on a financial world that most academic research has ignored because its records exist only in filing cabinets, not databases.

The study, published in Information Systems Frontiers, was conducted by Chang Chuan Goh, Wanning Li, Anthony Bellotti, and Xiuping Hua, who gained access to an anonymous lending institution founded in 2014 in Ningbo, a port city of 6.21 million people in Zhejiang province. Ningbo is an unusual and revealing setting. It has a GDP per capita of roughly 170,000 yuan, about 24,000 US dollars, ranking second in its province and twelfth nationally, and it hosts a centuries-old tradition of informal finance, including the historical qianzhuang, or native banks, that once dominated Chinese commerce. The borrowers in this dataset are not the destitute villagers who typically appear in microfinance studies. They are relatively wealthy residents of a prosperous industrial city who nonetheless cannot get credit from a mainstream bank, forced instead to pay interest rates that cluster around 21.6 percent per year.

Why such borrowers remain shut out of the formal system is one of the study’s central puzzles. The researchers document that most of the loans in the portfolio, roughly three quarters, went to small and medium enterprise owners, a group that Chinese commercial banks have historically avoided. Big banks, the literature suggests, are risk averse and tend to exclude SMEs because individual borrowers are informationally opaque and costly to assess. The result is a paradox: people with substantial incomes, valuable homes, and large credit card lines still turn to a high-interest informal lender, because their constraint is not poverty but access. Financial resources, in this context, mean the ability to obtain credit at all, and for Ningbo’s underbanked business owners that ability was severely limited.

The dataset itself required considerable cleaning before any conclusions could be drawn. The initial sample contained 11,079 applications filed between July 2014 and July 2018, covering payday loans, commercial loans, agricultural loans, and business loans. Every applicant had to submit application forms, identification cards, household registers, education certificates, marriage certificates, and credit reports, which lenders verified by hand. After removing loans with excessive missing data and 57 outliers whose interest rates sat more than three standard deviations from the mean, the researchers were left with 10,601 observations, of which 4,645 were funded. For default analysis, they applied a standard definition from the existing literature, treating borrowers with more than 90 days of delinquency as defaulted, and imposed a two-year observation window, yielding a funded sample of 3,533 loans.

The first major finding concerns demographic discrimination. Using chi-squared contingency tests across gender, age, education, and marital status, the researchers found that funding rates and interest rates were not distributed evenly across groups. Female applicants defaulted less often than male applicants, yet received no advantage in funding, a pattern the authors describe as taste-oriented discrimination against women. Young borrowers between 20 and 35 were funded less frequently than older applicants despite showing no significant difference in default behavior, and less-educated applicants faced the same combination of lower funding and no worse repayment. Divorced and single borrowers, by contrast, did default more often than married and widowed ones, so the unfavorable treatment they received the researchers classify as profit-oriented discrimination, a rational response to measured risk rather than pure bias. Because all applicants were locals of a single city, the design neatly sidesteps the confounding effects of regional economic differences that plague studies spanning multiple markets.

The comparison between business and non-business borrowers is where the dataset’s granularity pays off. Business loan applicants were, on average, four years older than other applicants, and their reported monthly income averaged around 147,000 yuan, more than seven times the figure for non-business applicants. Some 57.2 percent of business borrowers owned their homes, compared with a markedly smaller share of household borrowers, and their homes were worth more. Their credit card lines were more than twice as large, and their average credit scores were higher. Yet despite this affluence, they borrowed far more: the average business loan granted was approximately 168,000 yuan, more than four times the average for other loans. The researchers argue that this combination of wealth and informal borrowing is strong evidence that China’s formal banking system is failing to serve even reasonably well-off entrepreneurs.

Default statistics sharpen the contrast. Business loan borrowers defaulted at a rate of 15.0 percent, exactly half the 30.2 percent rate among non-business borrowers, even though they carried debts four times larger. Their funding success rate, 45.8 percent, was correspondingly higher than the 38.9 percent achieved by other applicants, and they paid lower average interest rates. These gaps led the researchers to suspect that lenders treat the two loan categories with genuinely different decision logic, and to design their analysis accordingly rather than pooling all applicants into a single model.

To capture that logic, the team reverse-engineered the lender’s actual two-stage decision process using logistic regression. In the first stage, loan officers screen applications using demographic information, social capital proxies such as family size and emergency contacts, assets and liabilities, and credit history, all without a formal credit score, because acquiring one is costly and slow. Applicants who pass this pre-screening progress to a second stage, where the lender pulls a credit score and makes the final funding decision. The researchers modeled each stage separately, then built a third model for default among funded loans. Crucially, they included interaction terms between a business-owner indicator and every other variable, allowing each characteristic to carry a different weight depending on loan type. One curious detail emerged during modeling: 15 funded applicants had no credit score at all, likely reflecting the influence of guanxi, the Chinese practice of cultivating personal relationships, which is well documented as a factor in informal lending. Because these exceptions represented only 0.15 percent of the sample and their missingness was not random, the researchers excluded them rather than imputing values.

The regression results confirmed that lenders apply different criteria to different loan types. Gender, marital status, and the number of credit inquiries all showed opposite or distinct effects for business versus non-business applications, visible in the direction of the interaction coefficients. Business ownership itself was significantly associated with funding success, confirming that lenders make systematically different decisions for entrepreneurs. But here the story takes an unexpected turn: when the researchers ran the equivalent model for default, neither the business-owner variable nor any of its interaction terms was statistically significant. The predictors of default, in other words, appear to be the same for business and household loans, even though the two groups default at very different rates and get judged by different rules. The divergence in funding decisions, the authors suggest, reflects what lenders believe rather than what actually predicts repayment, a subtle form of information asymmetry in a market where decisions are made by human officers reading paper files.

The study’s broader significance lies in what it says about a population that online lending research cannot reach. Most previous work on Chinese informal finance has drawn on peer-to-peer platforms, where borrowers must by definition have internet access and digital literacy, making them systematically better educated and better off than the truly underbanked. Studies of rural microcredit, meanwhile, focus on poverty alleviation rather than the credit dynamics of wealthy cities. By documenting an offline market serving affluent but excluded borrowers, the researchers fill a genuine gap and add offline evidence to a discrimination literature previously built almost entirely on online platforms, including documented gender and education gaps in Chinese P2P lending. They also uncovered age discrimination against young borrowers, a pattern not previously reported in online data. For anyone wondering who falls through the cracks of both the formal banking system and the fintech boom, the answer, it turns out, includes some of China’s most prosperous citizens, queuing at a microfinance office in Ningbo with their marriage certificates and household registers in hand, paying 21.6 percent for the privilege of being seen.