Arnold Kling  

A Theory of Financial Intermediation

Nobel Prize in Abstraction... Clark gives Greif Grief...

I write,

Fundamentally, financial intermediation is about enticing investors to buy securities backed by investments whose risks the investors cannot fully evaluate. The intermediary, such as a bank, hedge fund, or ordinary corporation, specializes in evaluating risk. The investor who buys securities from the intermediary looks to the past performance of the intermediary as well as to concise summaries of the risk of those securities. The ratings of "AA" or "A+" by bond rating agencies are just one example of these concise risk summaries.

...At each step in the layering process, some of the detailed information about the underlying risk is ignored. Instead, investors rely on summary information. It is this use of summary information that makes these investments liquid -- that is, it enables them to be bought and sold by many investors. As an intermediary layer is added, while the amount of detailed risk information is going down, liquidity is going up. The result of this process is that the ultimate borrower -- in this instance, the home buyer -- pays a much lower risk premium than would be the case in the absence of liquidity.

This is a theory of financial intermediation that I do not believe you will find in the economics literature. I think it is an important theory with interesting implications. It

explains the booms and busts to which the economy is subject, as exemplified by the bank runs of the Great Depression, the 1996-2000 euphoria/crash of dotcom stocks, and the recent boom and bust in subprime mortgage lending.

Financial innovations, such as credit scoring for mortgage loans, allow intermediaries to make better risk decisions. At first, when an innovation is unproven, few investors are comfortable with it, and the reduction in risk premium is slight. Over time, investors gain confidence in the track records and disclosure methods of the intermediaries, and this lowers the risk premium. Occasionally, a combination of investor overconfidence and poor disclosure practices causes this process to overshoot. Risk premiums get too low, somebody gets burned, and the market corrects. At the point of correction, the flaws in disclosure practices become evident, and the market shuts down until new methods are developed and investors recover their confidence in intermediaries.

If I wanted to publish this theory in a professional journal, I would first have to dress it up in mathematics. This would mean coming up with a way to describe the difference between complete information and a summary of information. That is not so easy. It would involve coming up with a way to describe the advantages of specialization, something that was easy for Adam Smith to describe in words but which is not so easy to describe in math (I think of Paul Krugman as being one of the few to try). For my theory of financial intermediation, the specialization has to be in information acquisition and summarization.

So, putting this model into formal terms would be a challenge. The reward for meeting this challenge would be publication. Maybe that is worth it. But my conjecture is that formalizing the model would not add to its value, and conceivably it could send the interpretation of the theory in the wrong direction. I would argue that the formalization of Keynes' theories served to emasculate it of the interesting notion of a conflict between the desire to hoard and the desire to create.

Also, apropos recent discussions on our blog of the villain/victim paradigm, my essay challenges the media characterization of a financial executive as a villain.

UPDATE: more on risk, transparency and the mortgage market, along with the macroeconomic impact and policy responses, from Fed Chairman Ben Bernanke.

Comments and Sharing

COMMENTS (19 to date)
Josh H writes:

My question isn't whether economic theorems can be formalized in mathematics, but whether they can be tested in reproducible experiments and are falsifiable. If so, what does it matter if they are set in mathematics? If not, can this be called science, and how do we know which theorems are bunk?

Nathan Smith writes:

I'll defend the mathematical modeling approach. It is sometimes, not always, the case that a mathematical model can yield non-obvious true insights. Larry Iannaccone's paper "A Formal Model of Church and Sect," which I've just been reading, seems like an example. In other cases, the math may not add any insight, but it can make it clearer to people within the profession just what it is that you've said, and how it resembles or differs from what others have said. Also, even if *you* don't get any new insights from the mathematics, you may enable others to extend the model more easily at the mathematical level, and *they* may get new insights from the math.

An explanation of business cycles in terms of institutional development-- with markets and/or regulators solving ever more complex problems of institutional design as the economy's complexity increases-- would be, I think, an interesting addition to the literature.

Nathan Smith writes:

By the way, for the beginnings of a formal model of specialization, see the sketch of an Agent-Based, Spatial, Endogenous Growth Model which I just posted on my blog. The seek is to use computational modeling.

Ajay writes:

I was curious what this financial executive actually did that you believe he's being unfairly crucified for so I actually read the WaPo article you link to. This is a guy who's on trial for accounting irregularities that cropped up long before the recent subprime mess. Do you have inside information that would show that he's innocent? Because his long legal struggle is very different from the recent subprime-related media and popular critiques of the financial industry. It doesn't fit in with your argument at all. It would be like writing a blog post about how the justice system is unfair to black people, undercuts their community in various ways, and harasses them constantly, and then saying "Just like they did to OJ." Really? You think OJ was unfairly crucified by the justice system?

Karl Smith writes:


I tend to think you are essentially correct about the role of risk pricing in the business cycle. I would add that I think there is also an externality effect that I am wrestling with modeling.

Essentially the fact that one business defaults on its general (not simply financial) obligations increases the probability of further defaults. This can cause a rapid acceleration in the risk premium. Something like a risk cascade.

This acceleration has the effect of an increasing tax rate, driving a wedge between market participants and destroying transactions.

I would think advantage of modeling it is that we can test it. We can test it statistically but we can also do more general back of the envelope type stuff to see if the effect is big enough to explain the cycle and if it has the correct dynamics.

Ultimately if we want to move beyond philosophy to practical economic advice then we don't we have to be able to generate some type of quantitative predictions?

Matt writes:

This will be a fun project.

Let me talk about our local farm banker, an independent with an office right in the farm community. This guy actually does improve the efficiency of the local ag business, his risk intermediation is a real energy efficiency game.

Setting aside money, what does this guy do? He drives his pick-up to all the local farmers, meets the local truckers, meets the equipment salesmen, and gets thease guys to meet, contractually, on delivering a value range over a transaction range. In the purest form, acording to my theory. In fact, this guy really does that, he will pester the tractor salesman on behalf of the orange farmer.

He removes cyclic volatility that he estimates by organizing the timing of the market, he is an linear estimating auctioneer. He recovers the power factor losses from turbulence, actual power factor losses, like tractors siting in the show room floor too long and become outdated.

By greasing the skids, he keeps the collection of local ag industries operating as close to a real valued single eigen vector as he can. He is a matrix diagonalizer. He does the same things economists do, decomposes an aggregate economy into a different spectral estimation that is better diagonalized.

He is a player, don't foget to include his wealth function and his competitors.

Matt writes:

He goal of the mathematical model, if I worked the problem, would be to perform convolvation operations on actual wealth distributions and calculate the retained wealth by the banker community for an expected gain in efficiency.

This is similiar Mankiw approach on taxing height, he decomposed labor into height and residual labor, then height, social progressivity, and residual labor. He estimated the error bound between decompositions, then convolved actual height and labor distributions to predict the cost in progressive redistribution.

We figure out how much the banker made, we hang the banker.

aaron writes:

This is the same theory behind money in general.

aaron writes:

Matt, sounds like a project for a climate modeler. ;)

General Specific writes:

Without quantification, arguments are prone to hand waving. They're also prone to reinterpretation as the facts come to light. And more prone to political and ideological manipulation.

Newton's level of mathematical sophistication was not sufficient for a modern jet figher. Yet you propose that economics as a science step back to the level Adam Smith's rather wordy tome.

There should be room to explore and play with ideas. But let's not stop quantifying them.

MT57 writes:

1) I think most financial intermediaries seeking to attract investments emphasize their ability to identify opportunities for profit/reward, as opposed to evaluating risk. I know that "opportunities for profit/reward" can be re-expressed as "identifying the mispricing of risk" but I thought it might be helpful to mention this in case you choose to distribute it more widely.

2) I hope you will elect to maintain the discipline of mathematical modeling. I distrust a merely normative approach to economic questions and, in addition to the virtues already mentioned, modeling rigorously enables one to maintain the intellectual high ground vis a vis lazier adversaries.

Arnold Kling writes:

I did not say that Leland was connected to the subprime mess. In fact, my guess is that if he had remained in charge he would have not only have stayed out of that mess but chastised the lenders who were getting into it.

My point is that (a) he realized that it is important to understand how you make your money and (b) he tried to make the corporate disclosures reflect reality (we published the NPV curve along with our financial statements). The "accounting irregularities" were an attempt to make the accounting earnings track the true value of the company. The idea was to treat hedge instruments symmetrically with the assets being hedged. Making the accounting "regular" made the earnings statements more misleading, not less.

Ajay writes:

Hmmm, after actually reading the linked column, I see that Arnold tries to justify his support for Brendsel there. I assumed from the excerpts here that Brendsel's situation was not explained in the column but I see I was wrong about that. However, I'm skeptical that Brendsel is being persecuted simply for what one accounting firm allows and another doesn't. I would think the law wouldn't be vague enough to allow prosecution in such a case but I don't know much about accounting law.

General Specific calls for quantification in economics but does not realize that mathematization does not equal quantification. Most mathematization is simply a restatement of words with symbols and almost never leads to quantification. To quantify, one would have to actually insert data into a model and that almost never happens with most mathematical models. He could argue that mathematization is often an intermediate step before a model is programmed into a computer and then fed data but the truth is that there is even more work that needs to be done to turn a mathematical model into a computer program. It is much easier to go from the idea directly to the computer program, specifying it precisely in computational terms, than to go from idea to formal math to computer. The formal mathematic step is a useless step that is currently highly valued only because of tradition and because the economics profession is mostly fairly mathematically and computationally illiterate.

This status renders MT57's comments laughable. Kling is not being merely normative but verbally descriptive. The "intellectual high ground" you claim is actually a swamp that has no use. The mathematization of economics is comparable to how every technical discipline produces tons of jargon. It is done largely to shut people who do not want to devote all their time to learning this useless jargon out of the discussion. The claim from those using this tactic is that the jargon facilitates discussion by providing linguistic shortcuts but one only has to look at the broad variety and worthless complexity of the jargon to realize that is not what it is for. All technical disciplines constrain the supply of those in their profession and the amount of criticism they take from outsiders by hiding behind this jargon. The long-standing kings of such verbal obfuscation are those in the medical profession (did you know medical students were required to learn latin as recently as a couple decades ago?). Mathematical modeling, as separate from computational modeling, is simply another type of jargon that is much more difficult for most people to comprehend.

General Specific writes:

Ajay: Please note that by quantification I do not mean closed form mathematical solution. I mean quantification with the use of real world data. Hence my example of the fighter jet, which is not in all cases described by nor controlled by closed form solution, but has been quantified in such a way that the on-board processors (as well as the computers used to design it) can map input data into output data and then compare that output data with inputs (feedback).

That's all I meant. Verbal description is good and is all that can be done in many cases. And that is better than nothing. But I do have concerns that the argument for verbal description can be used as an excuse to either (a) not attempt the hard work necessary to quantify or (b) wave hands.

Galbraith argued for qualitative economics, not quantitative (if I remember correctly). I'm good with qualititative political economy, with sufficient emphasis on political. In other words, try something based on a good argument, see if it works. But still have measurements (quantification) of some sort to ensure that one is actually achieving what one intends.

Ajay writes:

General Specific, it sounds like you're in favor of computational modeling, which I'm sure Arnold is, and certainly I am, in favor of. However, you took Arnold to task for quantification when all he was doing was decrying the knee-jerk requirement of formal mathematical models by many economics journals. He was not arguing against quantification and neither am I. We're merely arguing that formal mathematical models are not very useful. btw, I think your final description of qualitative political economy is a good description of what can actually be achieved in any social science.

General Specific writes:

I may have read Kling's suggestion wrong. Wouldn't be the first time, contrarian that I am. But I hear him talking about conceptual models without any suggestion that certain aspects have to be mapped into numbers, nor a discussion of the relationships between the different concepts, which also would be numerical if programmable.

A business philosophy, or a financial theory, can certainly be described conceptually. It may even make someone a lot of money--e.g. creating a company based upon these concepts. Or help the economy. In other words, it may prove very useful.

I just argue that one needs to be cautious about what Kling proposes. His previous ideas on the future of economics sounded, in some ways, like economic journalism to me, not economic science. E.g. his proposal for more blogging, etc. I just don't see much closure in blogs or even progress. Sometimes. Not often.

That said, fields like cognitive science are filled with conceptual models (e.g. pinker's books) that are only partially mapped into the real world. They're still interesting, but pinker's models won't blow up an economy. Conceptual models, put into practice, might (e.g. if risks are not properly understood).

A Rating Guy writes:

Hi -

Oddly enough, I work for a rating agency (which will remain nameless) somewhere in Europe, and am one of the boffins who do the ratings. Let me make a couple of comments.

Financial intermediaries do indeed live on the ability to quantify risk. Unfortunately, there is virtually no way to make an academically meaningful risk analysis based on the danger of a particular loan non-performing, as a) the data will either be unavailable due to privacy laws or b) the data is limited to the customers of a single bank, which is not representative of the universe of borrowing clients.

Hence within the literature there are generally no good case studies or meaningful empirical studies available, as the data is not available, nor will it ever be.

Further, this is an area where businesses can be made or broken and the fate of extremely large amounts of money is at stake: companies that provide rating services do not talk about their methodologies, except to clients, because these are trade secrets.

I know that I do extensive back-testing and that I have a fairly impressive track record: when one client took the ratings we did 10 years ago and took a close, hard look at what actually happened (and to decide whether to continue to retain us...), we were the only significant indicator, including their own risk analysis. They calculated that if they had listened more closely to us, they would have saved several hundred millions over the last 10 years in non-performing loan. And we're talking real-world losses that actually had to be paid.

Like I said, I will not talk methodology, other than to say that it is fiercely quantitative. There is no other way to go: rating things qualitatively simply depends too much on a subjective judgement of factors to be meaningful.

Getting back to the subject itself: financial intermediaries in the real world are intensely interested in ratings, but know how difficult it is and how dangerous it is to rely on them, as many do nonetheless, without understanding what stands behind them. Some banks have entire departments which do nothing but these kinds of analyses: their track records are more often than not less than stellar, as they lack an objective set of standards to work off of.

This is why Basel II was so important: it made banks and other financial intermediaries take a long, cold, hard look at risk, as their ability to operate depends on their reserve requirements. A properly run bank can exploit these requirements to be very successful: a competitor who lacks the risk analysis skills will not be nearly as competitive in terms of return to capital.

This is a great area to be in, but it is almost completely empirical work done under nondisclosure rules that preclude discussing it even within an academic environment, let alone a public environment.

Brad Hutchings writes:

President Buh just made this exact point about liquidity being a good thing in the market. And then followed up with the point that there's a disconnect between public understanding and how this market works. If the dumbest guy in America gets this...

Matt writes:

Arnold wants an economic model based on the value of information, or the value of predictability. Arnold needs a specific cost mechanism related to the acquisition of producer predictability. He needs to relate information management to producer inventory efficiency.

The value measurement he could use would be the capital reserves a producer needs to insure he does at least as well as the market average in the probability of his own market failure.

That is, capital reserves/market share, this ratio, relative to other market participants determines the effective use of capital in making accurate predictions of individual market performance. The ratio ignores the mechanism by which capital investment produces less prediction noise, but it does demand that we can write the force equations betweens inventory control and market capacity.

By managing market inventory better than the average, the producer causes the market to become more valuable and his share to increase.

Comments for this entry have been closed
Return to top