The credit crisis: Analytics' most dangerous game

What a difference a year makes. Back in August of 2007 I spoke with Dennis Santiago, CEO, founder and Chief Executive Officer of California-based Institutional Risk Analytics, a professional services firm for the financial services industry. While most of his comments didn't make it into my subsequent column, Greed, Analytics and the Mortgage Lending Crisis, much of what he said at the time turned out to be quite prescient, given what has transpired since.

Santiago knows his way around the complex financial instruments that brought down Wall Street: He got his start in the business putting together a collateralized mortgage obligation (CMO) system for Countrywide. He also has performed operations research for the military, where he focused on global stability analysis. Santiago and I talked about credit crisis and how it was transforming the discipline of risk analytics and its use in financial services. His advice and perspective remains as timely as ever. In fact, the market appears to be turning back to risk analytics in a big way as it begins the slow climb out of the credit market crater.

Here's Santiago's take on what happened - and where we're going from here. His most interesting comments are underlined.

What does your company do?

It develops advanced analytics which are mathematical algorithms for assessing risk.

How do you do that?

We have bank analytic tools used by [banks] that want to assess the condition of a financial institution as to safety and soundness. [It is] for people interested in whether a bank is a party they want to do business with on bank-to-bank basis.

We [also] look at risk analysis for companies that are not banks. Banks have an interest in tracking the condition of obligors. Our techniques are used to gauge the condition of a company. We help to do statistics around these measurements.

What can those tools do - and not do?

These tools, they're based on a school of thought that looks at the empirical information about the company. There are two schools of thought in risk analysis. One is empirical. The other is statistical. Empirical assumes that market prices are efficient. This is called efficient market theory. That way of looking at risk assumes that markets are so efficient that you can't hide anything from them.

A parallel theory says that one looks at the fundamentals, the fundamental theory of valuation. It looks at the books and ignores the stock price. The two models are both quite good and at certain periods in the business cycle one tends to be more favored than the other.

Fundamentals [are favored] in times of more uncertainty. Market pricing is more favored when everything is going well. We're in a period of uncertainty.

What went wrong with the statistical models?

Statistics and theory-based models have driven risk analysis for so long that they've reached a point where they've flattened out on their effectiveness in seeing what's going on. That relates to the credit crunch. Part of what allowed the crunch to happen is that some of these models came to be relied upon so much that people began to work the model as the equivalent of value as opposed to maintaining a healthy desire to always make sure you look at more than one model at a given time and make an assessment of which you believe more.

So because the models were too popular that they failed?

You're seeing the classic effects of something akin to a biological population explosion scenario where a set of models became so used that they came to wash out everything else and the balance was lost. Now it's being caught.

So are the models changing?

Yes. That's the third line of analysis that seems to be beginning to evolve. Heuristic analysis, where we're trying to infer things from things that are not structured or numeric. Things like trying to make sense of the pattern of a conversation. We have tools that look at SEC documents and read through them. We are not analyzing them like credit risk but we use keywording techniques to assist in speeding up cataloging of file documents.

What you're seeing now is overall the technologies are becoming much faster at rumbling through larger and larger libraries and returning the most relevant information to the user at speeds that were unimaginable even a few years ago.

Multicore processors are making it possible [to do] pretty much all the stuff that has been algorithmically thought about for many years. Now you can do it at thousands of times the speed.

What roles did such tools play - or not play - in creating the credit crunch? Was it the tools or the people who used them that were the problem?

They saw the risks. What they did was take another mathematical concept called a derivative, to essentially write an insurance policy to that risk. You know things are degrading but the theory is that the system is so large that no single debacle can sink the system.

People who figure they can get a high return on the hedge bought them. As risks go up you expect to see hedge funds rise in number and business volume.

One of the problems was as the systems became more and more accelerated the ability to see the underlying collateral became more and more opaque. As that happened people relied on what they had left, which were the mathematical models.

The problem is that all math has a limited lifetime. Nothing is applicable forever, particularly when it comes to financial risk. You can't correctly predict risk if you can't see the collateral. If you can't see the goods, you're guessing. While no single one of these things is enough to sink the world, if you get enough of them you get a systemic risk. You have a blockage in the ability to flow the information you need to see what's going on.

How did the problems first manifest themselves?

There were two manifestations: On the loan origination side and on the finance side. The finance side has to do with how this process has shifted its financing schema to rely more and more on this type of statistical, mathematics-driven structure for risk management.

In the old days it didn't used to be that way. The mortgage bank borrowed money and used that to fund loans. Most [loans] were conforming and it would securitize those loans and deliver those to GNMA and FNMA and they would generate bonds with them. These days that situation changed because of the upsurge in subprime [loans].

On the loan origination end a parallel scenario was playing out. The free availability of lower and lower quality loan qualifications caused the housing market itself to artificially rise in price. Most of the price appreciation was due to fact that price appreciation made for easy money. A lot of people went up on their mortgages with housing prices that were inflated and relied on the fact that housing prices had continuously gone up for 10 years. The deal could only go to the next step if housing prices continued to go up. On a 2-step [loan they] could then refinance.

[The problem] has to do with the operational business policy of the participants in this market.

What we try to do is create analytics techniques that make those business policies more visible for people to see for people who operate those banks or do business in those banks. These are tools to help the bank to its job better.

How quickly can the markets adjust?

The thing about financial institutions is that, unlike stock prices, it takes a lot of time for the loan exposure and financial position exposure of an institution to adjust. They're slow moving.

For example, a bank is a mixture of loans pooled in different areas. It may have real estate and commercial portfolios, cash, and they borrow money. They have a soup of options they can play around with to manage the bank. Banks are set up to be as stable as possible. When they set up things like return on equity or assets they look for a target rate and try to manage the exposure of the bank to get to that stable point that fits with the risk taking policy of the institution. It can manage what the overall default rate should be.

You want to see how things are going. The tools tell that [with] forward-looking scenarios. Defaults are where banks try to manage that rate. You'll always have defaults. They look at current default rates and make changes in loan issuance policy as they move forward.

You have institutions out there that are beginning to follow that policy fairly hard. They are trying to change the nature of lending portfolio to increase the population of truly safe loans to counter balance the ones that they know are not so good. They're trying to reach a balance point.

How will the tools be used in the future versus today?

They will do things differently because economic conditions change. What one learns from analysis is that the world is not a static place, it's forever evolving. If one is to survive one has to be able to tweak one's tools. That's really the quest from the analysis side of the shop.

What's coolest stuff you're working on today?

From a technology standpoint, we're at the beginnings of seeing more and more thin portlet technology coming around. It's components that can appear on very thin terminals. That allows people to further centralize where algorithms and databases are managed and allows the tools and information to be delivered globally at practically light speed. What used to require a dedicated terminal can now happen from any Internet connection. The technology is evolving to the point where you can build something and it can be deployed on multiple users or outlets and essentially appear for any user anywhere equally fast.

What portlets do is allow ever more sophisticated things to be able to flow more transparently to people. That provides a technology basis for winnowing down the opacity of some of the unknowns in the system.

If you want the entire U.S. economic system to work better you need to streamline the blockages. Looking for technologies that can streamline those blockages is a very good thing. The availability of portlet technology on the front end forces the heuristic interpretation technology to line up behind it. It creates a demand point.

Had they used the right risk analysis tools, would the crisis have come out any differently?

I don't think [there was] anyone that knew what they were doing who did not see the increased risk they were taking on. It was obvious, whether you used a model or not.

The real thread that keeps recurring to me is that people looked at that and said "God, we've got to do it because if we don't some one else will and they'll take the money off the table."