MIDDLEWARE
Digital Dividends
McCombs School of Business researchers use TACC’s Lonestar system to model market behavior
Economics has often been called “the dismal science,” and with the economy staggering that may seem to be the case.
However, for Lorenzo Garlappi and Stathis Tompaidis, researchers from The University of Texas’ McCombs School of Business who are at the forefront of financial modeling, it is an exciting time to be studying markets.
“Economic modeling is not any different from any other field of science,” said Garlappi, associate professor in the McCombs finance division. “It has the ambition of trying to describe phenomena through a stylized model designed to capture some of the features that you believe are important for explaining the particular events that you observe in the real world.”
Since the 1700s, economists have taken an increasingly scientific approach to the study of financial markets and asset values. With the advent of powerful high-performance computing (HPC) systems, it has become possible to create virtual markets with a complexity and realism approaching actual economies, and to produce tools for predicting how financial markets will act and understanding their underlying logic.
Why do firms who face high risk of financial distress promise lower returns? Why do many investments tend to be profitable in their first twelve months (momentum) and then lose money afterward (reversal)?
There are a great many empirical facts about markets that are widely accepted but are not well understood by investors, managers, or economists.
An economy, after all, is a very complex apparatus, with thousands of traded companies, hundreds of millions of investors, and a near-infinite number of possibilities. At the same time, certain aspects of markets are known. Investors act rationally (for the most part), and when two sides are satisfied with their positions — when everyone feels they’ve made a good deal — an equilibrium emerges.
To tame the chaos of market data, financial analysts and money managers make use of models that help elucidate the market.
“Sometimes, it’s hard to make sense of the data without the discipline that the model can give you,” Garlappi explained. “The model is a pair of glasses that allows you to see more clearly through the forest of the data. After you have the model in place, you validate it through empirical tests. The purpose of the model is to deliver predictions that can be directly tested and falsified in real world data.”
Both Garlappi and Tompaidis, as modelers, had originally developed their virtual economies on personal workstations. But when the problems they wanted to solve grew too complex and data-intensive, they turned to the Texas Advanced Computing Center (TACC), where the parallel processing power of supercomputers allows for more intricate models, while dramatically reducing the time to solution.
Garlappi, working with Professor Hong Yan at the University of South Carolina, uses TACC’s Lonestar system to develop and test a financial model that explains the dynamics of equity returns for firms in financial distress.
Simulating the investment decisions of idealized firms that make optimal decisions at each point in time, reveals to Garlappi how the market operates, and how changing parameters impact a firm's investing and financing decision as well as the risk and return of the securities it issues.
“I create a population of firms, each behaving optimally,” Garlappi explained. “I ask my laboratory economy to produce data about the return that a given firm will experience if it makes optimal decisions. Then I perform empirical tests on the simulated data exactly as it is done in a laboratory experiment.” The goal is to calibrate the model so that the parameters deliver data similar to what is observed in the real world.
It is possible to solve this type of problem on an individual computer, but using Lonestar reduced Garlappi and Yan’s time to solution from about three weeks to a day and a half, allowing the researchers to calibrate their model and derive insights blindingly fast.
Lorenzo Garlappi (above) and Stathis Tompaidis (bottom) use Lonestar to solve financial problems that cannot be answered without numerical methods.
Apropos of the current financial situation, the researchers found that in times of distress the renegotiations between firms and their creditors, where debt obligations are reassessed to keep a business afloat, play a large role in the risk-structure and the profitability of those firms.
Stathis Tompaidis, associate professor in the McCombs Department of Information, Risk and Operations Management, also uses TACC’s HPC resources to solve problems that cannot be answered without numerical methods.
Applying HPC techniques was second nature for Tompaidis, who came from a nonlinear dynamics background, where numerical methods are commonplace. However, he found that in finance, advanced numerical methods were scarcely used.
“Scientific computing in finance has not been widely adopted,” Tompaidis said. “Mostly, people with more of a technical background have started using it.”
His work applies computational algorithms based on Monte Carlo methods to the choice of optimal asset allocations: how investors decide where to put their money in light of different factors. “It comes down to a problem where you have to evaluate your allocation based on what could potentially happen — what could the returns be, given what choices you make,” he said, “and that’s something where you have to project far into the future to take into account the possible needs that you may face.”
One example of Tompaidis’ research involved creating models to understand how investors choose what amount to invest in risky assets at different stages of their lives. To his surprise, he found that some optimal decisions run counter to the advice typically given by money managers.
For example, a young person with many earning years in front of them will often be found holding a portfolio that is concentrated on high-risk stocks, rather than being diversified, which has typically been attributed to youthful irrationalism.
“The advice that everyone receives is to hold a widely diversified portfolio,” he said. However, according to Tompaidis’ model, risk-taking youthful investing turns out to be an optimal approach, leading to the highest long-term earnings. “Investment strategies really depend on a person’s stage of life,” he said.
- Many experts have blamed quantitative, computational models — widely used by large investment banks — for failing to anticipate the current economic downturn.
Garlappi and Tompaidis have their own perspective on the models, their overuse by analysts, and the public’s expectations of the markets as a whole: - “People are looking for some kind of scapegoat. It’s not the model’s fault. For people to make money, they have to take risks, and when people take risks, there’s always the possibility of things going bad. And then when things go bad, they say, we were taking risks and we did not realize it.” — Tompaidis
- “In August 2007, when the so-called quant funds started to melt down, the problem was that everyone was using the same framework and placing the same bets on the same assets. Then if all of a sudden you need to sell some of your assets, there’s nobody to buy. This is a consequence of using a model that everyone else is using. It exacerbates any price fluctuation.” — Garlappi
- “We’re financing the economy to a greater extent than we’ve done in the past and we’re doing that through leverage. So the choice is, you reduce the leverage, but in doing that you provide less financing for people, you slow down the growth of the economy, but you don’t have collapses like this one. Or you don’t reduce the leverage. You let the economy grow and then once in a while you have a big collapse.” — Tompaidis
- “This was not a problem of computational methods or models, it’s the fact that we think what we know that history is going to repeat, and sometimes it doesn’t. I used to teach a course on risk management and I used to make that point all the time.” — Garlappi
- “I think the situation is really bad. I don’t think financing is coming back any time soon; it’s going to be a couple of years.” — Tompaidis
While dynamic, non-linear computational models can help unravel hard-to-understand aspects of financial markets, they also add a layer of convolution that makes it difficult to assess the impact of a single factor, and should be used judiciously.
“The whole point of having fast computers available is to be able to study complicated models not for the numerical complexity of the algorithm, but for the model itself, to better represent reality,” Tompaidis said. “It cannot just be a model that is complicated. It has to be a model that is relevant.”
In the coming years, Garlappi intends to couple equity and bond interactions into a more comprehensive model of market behavior, while Tompaidis explores what happens when a fund changes managers. Their hope is that the computational models they create, and the insights they deduce, trickle down to bankers and money managers and inform the way investment decisions are made.
Though the use of simulated markets has faced some criticism of late, both by social scientists who believe investors are less logical that the models imply, and by pundits who blame the over-reliance on quantitative models for the current meltdown (see sidebar for a discussion of this fact), it is clear the models offer major benefits, both to the purely academic understanding of finance and to the real-world business of investing money and issuing debt and equity.
For that reason, the use of advanced computing to develop ever more complex models will continue to grow, with scientists like Garlappi and Tompaidis, and supercomputing centers like TACC, at the vanguard.
******************************************************************************************************************************
Aaron Dubrow
Texas Advanced Computing Center
Science and Technology Writer