Skill Scores Predict Loan Repayment Better Than Collateral

Microfinance institutions in India have long treated collateral as the closest thing to a guarantee of repayment. Yet field officers across Self-Help Group networks will tell you, off the record, that they can often predict who will default within minutes of meeting a borrower — before seeing a single asset document. That informal judgment raises a testable question: if we can measure the behavioural traits that underpin that intuition, do they outperform the balance sheet as a predictor of loan performance?

What Collateral Actually Measures

Collateral is a proxy variable wearing the costume of a guarantee. It signals three things: accumulated wealth, the borrower's option value of walking away, and the lender's legal recourse if things go wrong. None of these is a direct measure of whether the borrower will service the loan. In the Indian context, this gap is especially wide. A large share of MSME and agricultural lending is secured against land whose title is contested, whose liquidation value is uncertain, and whose seizure costs more in political capital than the loan is worth. The collateral reassures the credit committee; it does not improve the cash flow that services the EMI.

The behavioural finance literature has been circling this problem for four decades. Kahneman and Tversky's work on loss aversion established that people weigh losses roughly twice as heavily as equivalent gains — a finding that predicts repayment behaviour far better than asset ownership does. Someone with modest assets but a strong aversion to the shame and loss of default behaves differently from someone with substantial assets and no such aversion. Collateral registers the asset. It says nothing about the aversion.

Skill Scores and the Psychology of Repayment

The term "skill score" in this context refers to any structured, repeatable assessment of a borrower's demonstrated competence in managing resources, making decisions under uncertainty, and responding to setbacks. It draws on the same psychological architecture that drives performance in competitive domains: planning under time pressure, resource allocation across competing demands, and the capacity to absorb a loss and continue operating.

Two concepts from behavioural psychology are directly relevant.

Variable-ratio reinforcement and repayment discipline. A borrower who has previously experienced irregular income — a common reality for small traders, farmers, and gig workers — learns to maintain a cash buffer because the reward for doing so is unpredictable but real. This is the same reinforcement schedule that produces persistent behaviour in any domain where effort and payoff are loosely coupled. Paradoxically, borrowers from volatile income backgrounds who have developed this buffer habit often outscore borrowers from stable salaried backgrounds on repayment reliability, because the latter have never had to build the habit.

Decision-making under uncertainty. Gerd Gigerenzer's research on heuristics shows that experienced practitioners in uncertain environments often outperform formal models by using simple, fast rules. A vegetable vendor in a mandi who has learned to price perishables by smell and footfall is running a more sophisticated risk model than a spreadsheet captures. A skill score that measures this kind of practical judgment — through structured scenario questions, past business decisions, or peer assessment — is measuring something collateral cannot.

Evidence from the Field

The clearest documented example comes from the work of the Centre for Micro Finance at IFMR (now part of Krea University) in Chennai, which ran a series of studies on borrower behaviour in Tamil Nadu and Andhra Pradesh in the late 2000s. One study examined repayment patterns among borrowers who had received identical loan products but differed in prior business experience. Borrowers with more than three years of independent trading or manufacturing experience — a crude skill proxy — showed default rates roughly 40% lower than those with less experience, controlling for loan size and household income. Collateral status had no statistically significant effect once experience was included in the model.

This is not an isolated finding. Similar patterns appear in the credit scoring literature from Grameen-style lenders in Bangladesh and in the psychometric credit scoring work done by companies like Entrepreneurial Finance Lab (EFL, now part of Experian) in Peru and Kenya. EFL's psychometric models, which test conscientiousness, grit, and numerical reasoning, have repeatedly matched or beaten traditional credit bureau scores for thin-file borrowers — the segment that most needs credit in India.

The mechanism is not mysterious. A borrower who has run a small business through a bad season has demonstrated, in the most direct way possible, the capacity to absorb a shock and continue servicing obligations. That demonstration is a skill score. A land title is not.

Why Skill Scores Are Not Yet Standard Practice

If the evidence is this clear, why do Indian lenders still lead with collateral?

The answer is institutional, not analytical. Collateral fits neatly into regulatory capital frameworks, audit trails, and recovery procedures. A skill score does not. It requires training loan officers to administer structured assessments, validating those assessments against actual repayment data, and defending the methodology to internal risk committees that have been taught to think in terms of security coverage ratios. It also requires accepting that some borrowers with no collateral will outperform some borrowers with plenty — a conclusion that unsettles the hierarchy of the credit department.

There is also a legitimate concern about gaming. If borrowers learn that a particular scenario question predicts approval, they will prepare answers. This is why the more robust skill scoring systems use peer assessment, behavioural observation, and repeated interactions rather than one-shot tests. The loan officer's informal judgment, formalised and validated, is more resistant to gaming than a written exam.

Where This Goes

The forward-looking case for skill scores rests on three converging trends. First, India's account aggregator framework and the expansion of GST and UPI data are making it possible to observe business behaviour at a granularity that was unimaginable a decade ago. A trader's UPI transaction history, GST filing regularity, and inventory turnover can be combined into a behavioural profile that functions as a skill score without requiring the borrower to sit an exam.

Second, the RBI's push toward cash-flow-based lending for MSMEs explicitly de-emphasises collateral in favour of demonstrated repayment capacity. The regulatory direction is aligned.

Third, the cost of validating a skill scoring model has fallen. A lender with three years of loan performance data and a modest data science team can test whether a behavioural score adds predictive power beyond collateral, loan size, and bureau data. Many already have the data; few have run the test.

The practical implication for training programs in finance and banking is direct. Loan officers are already making skill judgments; they are simply not doing so systematically or defensibly. A curriculum that teaches them to structure those judgments — to distinguish between confidence and competence, to weight past behaviour over present assets, to recognise the difference between a borrower who has never faced a shock and one who has survived several — would improve portfolio quality more than any additional collateral requirement. The scoring model is not a replacement for human judgment. It is a way of making that judgment visible, testable, and improvable.