Fly-by-Wire Finance

Blake LeBaron

International Business School, Brandeis University

February, 2009

"Fly-by-Wire" refers to the type of interface between the human (pilot) and the avionics of a modern aircraft. In most of aviation history this connection was relatively simple, and the pilot received lots of physical feedback from how the airplane felt. A good pilot can probably sense that something doesn't feel right about the plane and take corrective actions. Modern aircraft put the computer between the pilot and the aircraft flight controls. The level of computer control can vary from minimal assistance, to complete autopilot systems, and even to aircraft which are dynamically unstable, and only the machine can control. Over the years finance has evolved similar flight control systems.

The origins of fly-by-wire in finance go back to the mid 1950's when Harry Markowitz clearly demonstrated that the computer could play a crucial role in basic portfolio risk management calculation. Curiously, even though computing power increased at an exponential rate this technology did not catch on all that fast. Running a portfolio "optimizer" in the mid 1970's was still considered radical technology by the finance community. Investment professionals understood diversification, and portfolio issues, but preferred to stick with their gut intuitions on prudent risk exposure.

I believe this state of risk management began to change in the 1980's because of two things. First, investment portfolios were becoming more global. Portfolio managers who would have previously turned off the computer, and "used the force" to determine their portfolio exposures across sectors of the U.S. economy were now being forced to make these decisions across countries. This was an area where their Bayesian priors were much weaker, and they looked to the computer for advice. Both practitioners and academics began seriously poking into the black box of computer based risk management. The second key change was the onset of derivative markets. As we all know, the use of derivatives means that one must delve deeper than simply looking at the names (or the ratings) of instruments in the portfolio. Only the computer could handle some of the more complicated hedge positions which now could be implemented. The move to take the machine front and center, and let it just fly the plane was now well underway.

The advent of VaR gave modern risk management a number which could easily explain a dimension of risk in a large and complex portfolio. Now risk management had the equivalent of the indicator gauges on old style autopilots that helped guide an airplane in for a landing. The final step was to tie this value to reserve capital, and risk exposure decisions, and take the pilot out of the equation. The pilot no longer could get any feeling in his or her gut since the problem was now too complex for human understanding. Both institutions and policy makers (through the internal models rule book of Basel II) steadily signed onto the idea of fly-by-wire.

The fly-by-wire world was not without some interesting crashes. The most important was the failure of Long Term Capital Management in the late 1990's. They were clearly at the cutting edge of modern financial avionics. Their technology was designed to monitor and control risk of a large and complicated portfolio. Quantitative risk models ruled the day, and controlled the fund's overall exposure. When it came to risk they had specific VaR targets in mind. Most importantly, through leverage they pushed the risk numbers to their limit. They borrowed extensively to squeeze the last amount of return from their strategies, while keeping the VaR autopilot dead on in the flight approach indicators. The collapse of this fund indicated problems with risk management systems, and their ability to guide capital reserves and leverage levels. (Obviously, the autopilot could only be as good as the modeling information fed to it.) It was clear that the systems were not working as planned, but somehow financial institutions ignored this rather large warning signal. Looking back, a kind of much bigger NTSB-like report guiding policies for the future is what was greatly lacking at that time.

The current crises shares some similarities to the LTCM crisis. The computer was given full control of risk management operations. Fly-by-wire was now no longer just part of the some boutique hedge funds, and currency trading desks, but had become the industry standard technology for risk management. Also, it was pushed into new components of risk like credit risk, and operational risk where its capabilities had not been tested extensively. Since both these forms of risk involve small probabilities of left tail events, formal testing in the statistical sense might require a very long time. A new twist on risk management also began taking shape as securitization moved forward. New classes of securities were designed which the human operator had absolutely no hope of understanding. Now the computer was not just flying the plane, but also building it as well. The old "fly by the seat of your pants" pilots always could feel something about the risk of a stock or corporate bond, and probably a layer of derivatives on top of that, but the mezzanine tranche of a CDO is something few of us would have any intuition about without the computer.

Up to this point this essay probably makes me sound like a Luddite. This is actually not the case. The computer is here to stay, and some form of fly-by-wire finance will always be with us. We cannot go backwards to a world of securities which are so simple anyone can interpret their risk just by looking at them. Derivatives and securitization are both here to stay, and if correctly used can still generate great benefits in terms of risk sharing. So what is the future of risk management going to look like? Quantitative risk management is not about measuring the "risk level" of an entire enterprise. We should realize that this is a hopeless task. It is about providing tools to the more intuitive (seat of your pants) realm of risk management, so they can do a good job of merging their scenarios and beliefs with the available data. (Technically, this would be a form of empirical Bayesian decision making.) Much of this may involve technologies that allow CFO's and other high level types to turn the dials themselves to see what will happen to their airplanes in the future. It will also depend on getting them the information in a form or visualization that will allow them to reasonably combine their experience and judgement with the empirical evidence. One thing behavioral economics knows is that how information is framed can greatly impact the ability of decision makers to make the right choice. The future of fly-by-wire finance is therefore not to pull the plug on the machine, but to put the human back into the pilot's seat with the computer as trusted advisor. Its mission is to present the pilot with the best available information, but in the end, human intuition will have to fly the plane.