Dealing with NPAs

Ask any banker how he/she evaluates companies to whom loans are being given, and you’ll be told they look carefully at liquidity ratios like the quick ratio (current assets less inventories as a per cent of current liabilities), or the liabilities to assets ratio, and so on.

In addition, they look at the growth prospects of the industry the firm is in and the quality of the audits, among other parameters — apart from the fact that firms with qualified audits tend to default more, research at ratings firm Moody’s found that firms who change their auditors had a 50 per cent higher default rate than those that did not change auditors the previous year!

Seasoned bankers then top this analysis off (sometimes it’s the other way around, sadly) with their own personal knowledge/contacts with the firms’ promoters.

Based on this rudimentary analysis, usually, a two- page note on the loan proposal is then sent from the branch office to the zonal/head office, where it gets cleared, either after some more analysis or on the basis of loan targets set for each branch office.

Such analysis, however, is often fraught with risk, as Moody’s showed when the firm was working on developing a credit risk model. I’d strongly recommend that readers download the paper (http://riskcalc.moodysrms.com/us/research/crm/56402.pdf) as it provides a lot more detail than I can possibly hope to give here.

Generally speaking, for instance, it seems logical that if a firm has a higher quick ratio, it has a lower chance of defaulting. Yet, as Moody’s found when it studied around 1,400 companies that had defaulted since 1980, lower-rated and unrated firms have significantly higher quick ratios in comparison to investment-grade firms.

Similarly, while a firm with a lower liabilities/assets ratio should be less likely to default on loans, Moody’s found that in general private firms have much lower liabilities/assets ratios than investment-grade firms, making a mockery of any exercise based on this ratio.

So, what’s a banker to do? Fortunately, help is at hand. At a recent conference on banking and financial services by consulting firm SKOCH, some bankers, like those from the State Bank of India and Punjab National Bank, talked of how they’d developed credit risk models using internal databases.

While no details are available of how these models work, a popular default-risk model used globally is the one developed by KMV, a firm taken over by Moody’s a couple of years ago (see this column of January 26, 2004).

While this model argues (establishes?) that stock price movements are the best predictor of potential default risk, another model (RiskCalc) developed by Moody’s is based on financial information on companies and claims good results as well. Both the Moody’s as well as the Moody’s-KMV models are used by some private banks in India.

Mahesh Vyas who is the managing director of the Centre for Monitoring Indian Economy, India’s largest private data collection agency, made a presentation of a model developed by CMIE for such default-risk, a model he claims shows marginally better results than the Moody’s one, and which runs on financial data that CMIE already provides in its Prowess database of over 8,000 companies in India.

According to Vyas, one of the biggest problems with risk analysis that banks do (he says few banks really use the risk models they have properly) is that they usually get data on the borrowing firm from the firm itself — while there is clearly some marketing spiel here since Vyas stands to benefit if more banks buy his Prowess financial database, the need for using independent data sources is an important one.

CMIE developed the model using data for 5,814 companies for the period 1995-2001 and there are 584 defaulting firms in the dataset. It found that around 3.5-4 per cent of firms defaulted each year, a number that, interestingly, is not too different from that in the US.

I got Vyas to run his model on defaulting companies I chose from the top of my head, and the results were chillingly accurate. Morepen Laboratories, I said, and after a minute or less, the model gave the company’s scores from 1995 onwards.

While the company got 10 out of 15 in 1995, it got 11 in 2001, 12 the next year and 14 in 2003 — according to Vyas, lending to firms with a score beyond 8 is a bad idea unless the interest premiums are sufficiently high. So, according to the model, lending to Morepen was always fraught with risk, even at the time when few, bankers as well as private citizens, thought twice about lending to the company.

Arvind Mills, similarly, scored 10 in 1995 (risky), and this increased to 15 in 2001 (very dangerous), and then fell to 12 in 2003 after it renegotiated loans (still very risky).

The model, however, breaks down when it comes to big firms like Reliance and Tisco (it shows them as risky propositions too) — Vyas says this is because the defaults sample he’s got is too small and since it does not include data on defaults by large firms, it is difficult to build in default-risk scenarios for them into the model.

Vyas has now got a bigger sample from 1993 onwards for over 1,500 companies and so will also be able to capture at least one boom and bust cycle as well.

Sadly, the RBI which has defaults data on all firms, large and small, for over five decades, refuses to part with it, though this would make default-risk modelling a far more robust exercise. So, in a sense, if more loans turn into NPAs, you have only the RBI to blame for it!