Credit Scoring Vs Personalised Lending

In India, all banks have internal credit scoring models and it has not helped them in lending to SMEs.

Two key reasons why banks (formal lending institutions) do not lend to SMEs in India are
a.) Lack of information on actual profitability/cash flows. No official documented evidence of income can be found. This is primarily in keeping with practice of dealing in cash, belief in oral contracts and mostly absence of contracts (sale/purchase) altogether. In addition, the myopic tendency to avoid taxes leads to minimal documentation or under-reporting of incomes. If no concrete data is found, what do the bank feed into their credit scoring models?

b.) Small ticket size and hence high operating expense per loan. Ideally a credit scoring model reduces the high opex of small loans by taking an automated call on whether to do a loan or not without having a credit person go through the case in detail as he is expected to do in larger loan ticket sizes. But, then, for these kind of small loans with no concrete data, what does one feed into the credit scoring model? Subjective evaluations of the credit appraising officer?

It is undisputed that the effectiveness of a credit scoring model is dependent upon the quality of data being fed. However, since the quality/reliability of data that is available to be fed into the credit scoring model is poor and subjective, especially in case of SMEs in India, how does one use a credit scoring model? In fact, due to high level of subjectivity, the data being fed into the credit scoring model can jeopardise the credit scoring techniques.

Credit scoring models can speed up process but they can not replace personalized diligence based lending till the time good data is available. There is barely any electronic footprint left by SMEs that can be dug up for credit analysis. The bank account (if available) details wouldn’t show enough balance, their transactions would be fairly thin. Understood that visiting the customers for a small ticket size loan results in high opex but in the case of SMEs where transactions are primarily cash based, expecting to lend to customers in the SME segment without meeting the customers, suppliers, buyers of the SME and without visiting the location is a recipe for disaster.

A few new age lenders are depending upon use of surrogates/proxies for assessment of actual cash flows, followed by close monitoring of loans. They depend exceedingly on customer visits. Their portfolios have performed well but yes, it is too early. While it is not free of subjectivity, this approach seems to be better than that of large banks who use an inflexible credit scoring model based on documented data. I agree that the rate of growth in such kind of specialized lending may not be as fast as mainstream financial institutions till the time sufficient electronic footprint is generated by the target SMEs. Meanwhile, using some form of credit scoring models in parallel can add a layer of check over the existing rigorous personalized appraisal procedure. This helps in reducing the impact of subjectivity in the personalized lending processes.

Credit scoring models are needed. The critical question that we need to answer today is, how do we improve the quality of data available to be fed into credit scoring models? If not immediately, how do we build the right data backend that provides high quality data in the future?

According to me, this kind of data backend will have two components, one that deals with general data points which can be used for bench marking and two that deals with individual specific data points that further become a part of the general database:

a.) Benchmarking: A good data backend can actually be a like a platform where location specific details on various businesses and margins are fed by staff of lenders. Over a period of time and volume, these numbers will give adequate guidance on the claims of margins/profitability made by the potential borrowers. Once the coverage of data collection efforts improve with time credit scoring models can play a better role.

b.) For individual evaluation: Individual credit/liability histories need to be pushed into this data back-end. Electricity bills, mobile phone bills, credit bureau details, need to be automatically fed or fed based on requests. The question is will the respective companies share data? Will a mobile company share prepaid recharge data of a customer?

Once this is done over a period of time, I believe a credit scoring model will start making more sense for SME lending in India. However, ditching personalized lending altogether, would continue being a distant dream for a fairly long time, if not for ever.

The approach can not be Credit Scoring VS personalized lending. The approach instead has to be Credit Scoring AND personalized lending.


Payments, Banking and Cost implications of cash – India

Electronic Payments have always intrigued me. I have written about this in the past. I was reading through a few more documents on electronic payments and read through the Reserve Bank of India Vision Document on Payments. Quite an insightful document in terms of statistics. However, my feeling was it doesn’t quite clearly layout the strategic framework to be adopted for payments in India. SOme statistics from the vision document and some other sources.

 Penetration of banking services
  • Of the six lakh villages in India, the total number of villages with banking services stands at less than one lakh villages as at end March 2011 and nearly 145 million households are excluded from banking.
Penetration of Electronic payment
  • Only 0.6 million of the 10 million plus retailers in India have card payment acceptance infrastructure.
  • Mid-2011, the number of non-cash transactions per person stands at just 6 per year.
  • 32% of e-commerce takes place through the system of “cash on delivery” (COD) NOT online payment.
 Other numbers:
  • The Indian bill payment market is a US$ 160 billion market. Indian households pay on an average 50 -55 bills a year. Among the electronic payments infrastructure, ECS occupies a 50% share followed by cards and bank account funding.
  • It is estimated that Government subsidies alone constitute more than Rs. 2.93 trillion and if these payments are effected electronically, it may translate to 4.13 billion electronic transactions in a year.
  •  The penetration of ATMs is 63 per million population and that of PoS terminals is 497 per million population
 Banking Infrastructure
  • Today, the banking infrastructure in the country consists of 80,000 bank branches, 1,50,000 post offices, 88,000 ATMs, and 500,000 POS machines. Of these, the rural banking infrastructure only consists of about 30,000 bank branches and 1,20,000 post offices. In comparison, there are more than 10 lakh telecom retailers that operate throughout the country.
  • 18 million outstanding credit cards and 228 million debit cards.
 How much cost does the economy bear to support a cash economy?
Cost of cash to the economy is 5-7% of GDP.
-costs for rbi  include printing currency, currency chest management, and wear and tear
-cost for bank include cash logistics, cash management, security, storage, and the opportunity cost of idle cash in branches and ATMs