Risk: More Transparency with Less Regulation

By SAM SAVAGE

The recent economic crisis has pitted the finance industry against regulators in what looks like a classic zero sum game. But a new paradigm might simultaneously increase transparency while reducing the amount of regulation.

An Historical Parallel

What did Fibonacci bring back to medieval Italy from his schooling in Algeria? Absolutely nothing, … nada, … zilch, … zero. And along with zero he brought the rest of the Arabic numerals; 1,2,3,4,5,6,7,8 and, my favorite, 9. Up to that point the Italian bankers, business men, and financial engineers had been struggling to keep their books straight and their hedge funds hedged with Roman Numerals, which are algorithmically challenged. Can you imagine the medieval scams and screw-ups that resulted from this opaque number system? The bankers and regulators were no doubt at each other’s throats. Then in 1202, Fibonacci published his book, Liber Abaci, which showed how decimal based calculations could be applied to problems in commerce. The book, in which his famous sequence appears, was a best seller, and once the medieval business men had tasted arithmetic, they never looked back. In fact, practically the only places Roman numerals are used today is to intentionally obfuscate film copyright notices, so you won’t know how dated that crummy movie is, or in legal documents, to intentionally make the numbers harder to alter.

Classical Statistics, the Roman Numerals of Risk

The classical statistical techniques used in most risk models are today’s Roman Numerals. They don’t even allow for simple addition. For example, to get the total risk across multiple elements of a portfolio, or divisions of an organization, you can’t just add up the risks of the individual parts. Instead you must perform a convoluted calculation, known, appropriately, as a “convolution.” The same applies to adding up the risks of a bunch of banks to get a handle on systemic banking risk; a subject getting a good deal of air play these days.

Actually, the Arabic numerals of risk have been under our noses for decades. In 1947, while convalescing from an illness, Manhattan Project mathematician, Stanislaw Ulam invented Monte Carlo simulation for estimating the probability of winning at solitaire. This is a brute force computerized technique to simulate uncertainty by firing random numbers at a model like machine gun bullets, and monitoring what comes out the other end. Eventually people began to use sets of pre-generated scenarios to model such things as interest rates. By blasting the same set of potential rates through models of various business units of a bank, the results could be added up, scenario by scenario, to create a set of total risk scenarios across all business units.

Distribution Strings

In 2005, I helped Royal Dutch Shell apply this approach to its portfolio of exploration projects. By 2008, the size of the scenario data base was getting out of hand with tens of millions of numbers representing oil price and discovery volumes at sites around the world. We then came up with the idea of compressing thousands of scenarios into a single XML string in Excel, and the Distribution String, or DIST was born. In the following year, the idea was improved and turned into an open standard in collaboration with Merck & Co., Oracle Corp., SAS Institute and Frontline Systems. We were even fortunate enough to get input from Harry Markowitz, who besides being the father of modern portfolio theory, has played an equally prominent role in the area of computer simulation.

Connecting the Seat of the Intellect to the Seat of the Pants

Technological breakthroughs rarely occur in a vacuum. The light bulb and electricity, the automobile and the gas station, and the chicken and the egg, are some famous examples. In the case of Distribution Strings, the complimentary technology is Interactive Simulation. In traditional Monte Carlo simulation, the user executes a command, and the software goes chug, chug, chug, and a report is delivered. Compare that to an electronic spreadsheet, in which you change a number in one cell, and all related cells change instantly. If you don’t think interactivity is important, imagine a bicycle with a command line interface, on which you type “Lean Left,” then hit a fire hydrant. In our early models at Shell, we were able to make our simulation models interactive by copying spreadsheet formulas a thousand times, so that each set of formulas looked at a different scenario. It was cumbersome, to say the least, but a user could make a change to the portfolio, and immediately see the impact on the distribution of results, providing powerful intuition.

Size Does Matter, and Smaller is Better

Recently, Frontline Systems developed technology for making Monte Carlo simulation in spreadsheets interactive. That is, a thousand trials are run before your finger leaves the key. This freed us from the drudgery of making multiple copies of all the moving parts in our model, and reduced the number of formulas by a factor of 1,000. Then the following year, the Distribution String reduced the number of data elements by another factor of 1,000. To paraphrase the late Illinois Senator Everett Dirksen, a factor of a thousand here, and a factor of a thousand there, and pretty soon you have a simpler paradigm.

Today, there are several off the shelf programs supporting Distribution Strings in both the spreadsheet and other environments. To learn more, visit ProbabilityManagement.org. There, among other things, you can download an interactive demo version of the Shell model, and a prototype of a portfolio model developed with Paul Kaplan, VP of Quantitative Research at Morningstar, which swaps in different DIST libraries for historical data, lognormal, or fat-tail assumptions (see the illustration below).

The Emperor’s New Clothes

In the December 2009 issue of Risk Professional Magazine, I co-authored an article with Aaron Brown, of AQR Capital Management, entitled The March Toward a Consolidated Risk statement, in which we suggest that this compact new data type for storing scenarios could lead to a more disaggregated approach to risk modeling. One can, for example, imagine regulatory agencies that publish benchmark DIST libraries for the entire world to see, covering GDP, Housing Prices, Unemployment, Interest rates, etc. Regulated institutions could readily run the libraries through their business models and return the resulting DISTs to the regulators, who could simply add them together to gauge systemic risk.

The key here is, that unlike the current Roman Numeral based risk models, anyone with a spreadsheet could do a sanity check on the input assumptions, and a DIST of housing prices, for example, which showed no possibility of decline, would be met by howls of derisive laughter reminiscent of the Emperor’s new clothes.

About the Author:

Dr. Sam L. Savage is author of “The Flaw of Averages: Why We Underestimate Risk in the Face of Uncertainty” and Consulting Professor of of Management Science and Engineering at Stanford University, and Fellow of the Judge Business School at the University of Cambridge.

Tags: DIST models, risk paradigm, sam savage, The Flaw of Averages