A recent study by The Intertek Group tried to assess how the use of financial modeling in asset management had changed over the highly volatile period from 2000 to 2002. Participants in the study included 44 heads of asset management firms in Europe and North America; more than half were from the biggest firms in their home markets.
The study found that the role of quantitative methods in the investment decision-making process had increased at almost 75% of the firms while it had remained stable at about 15% of the firms; five reported that their process was already essentially quantitative. Demand pull and management push were among the reasons cited for the growing role of models. The head of risk management and product control at an international firm said, “There is genuinely a portfolio manager demand pull plus a top-down management push for a more systematic, robust process.” Many reported that fund managers have become more eager consumers of modeling. “Fund managers now perceive that they gain increased insights from the models,” the head of quantitative research at a large northern European firm commented.
In another finding, over one half of the participants evaluated that models had performed better in 2002 than two years ago; some 20% evaluated 2002 model performance to be stable with respect to the previous two years while another 20% considered that performance worsened. Performance was widely considered to be model-dependent. Among those that believed that model performance had improved, many attributed better performance to a better understanding of models and the modeling process at asset management firms. Some firms reported having in place a formal process in which management was systematically trained in modeling and mathematical methods.
The search for a silver bullet typical of the early days of “rocket science” in finance has passed; modeling is now widely perceived as an approximation, with the various models shedding different light on the same phenomena. Just under 60% of the participants in the 2002 study indicated having made significant changes to their modeling approach from 2000 to 2002; for many others, it was a question of continuously recalibrating and adapting the models to the changing environment.
Much of the recent attention on quantitative methods has been focused on risk management—a relatively new function at asset management firms. More than 80% of the firms participating in the Intertek study reported a significant evolution of the role of risk management from 2000 to 2002. Some of the trends revealed by the study included daily or real-time risk measurement and the splitting of the role of risk management into two separate functions, one a support function to the fund managers, the other a central control function reporting to top management.
In another area which is a measure of an increasingly systematic process, more than 60% of the firms in the 2002 study reported having formalized procedures for integrating quantitative and qualitative input, though half mentioned that the process had not gone very far and 30% reported no formalization at all. One way the integration is being handled is through management structures for decision-making. A source at a large player in the bond market said, “We have regularly scheduled meetings where views are expressed. There is a good combination of views and numbers crunched. The mix between quantitative and qualitative input will depend on the particular situation. For example, if models are showing a 4 or 5 standard deviation event, fundamental analysis would have to be very strong before overriding the models.”
Many firms have cast integration in a quantitative framework. The head of research at a large European firm said, “One year ago, the integration was totally fuzzy, but during the past year we have made the integration extremely rigorous. All managers now need to justify their statements and methods in a quantitative sense.” Some firms are prioritizing the inputs from various sources. A business manager at a Swiss firm said, “We have recently put in place a scoring framework which pulls together the gut feeling of the fund manager and the quantitative models. We will be taking this further. The objective is to more tightly link the various inputs, be they judgmental or model results.”
Some firms see the problem as one of model performance evaluation. “The integration process is becoming more and more institutionalized,” said the head of quantitative research at a big northern European firm. “Models are weighted in terms of their performance: if a model has not performed so well, its output is less influential than that of mod- els which have performed better.”
In some cases, it is the portfolio manager himself who assigns weights to the various inputs. A source at a large firm active in the bond markets said, “Portfolio managers weight the relative importance of quantitative and qualitative input in function of the security. The more complex the security, the greater the quantitative weighting; the more macro, long-term, the less the quantitative input counts: Models don’t really help here.” Other firms have a fixed percentage, such as 50/50, as corporate policy. Outside of quantitatively run funds, the feeling is that there is a weight limit in the range of 60–80% for quantitative input. “There will always be a technical and a tactical element,” said one source.
Virtually all firms reported a partial automation in the handling of qualitative information, with some 30% planning to add functionality over and above the filtering and search functionality now typically provided by the suppliers of analyst research, consensus data and news. About 25% of the participants said that they would further automate the handling of information in 2003. The automatic summarization and analysis of news and other information available electronically was the next step for several firms that had already largely automated the investment process.
Advances in information technology are behind the widespread adoption of modeling in finance. The most important advance has been the enormous increase in the amount of computing power, concurrent with a steep fall in prices. Government agencies have long been using computers for economic modeling, but private firms found it economically justifiable only as of the 1980s. Back then, economic modeling was considered one of the “Grand Challenges” of computational science.
In the late 1980s, firms such as Merrill Lynch began to acquire super- computers to perform derivative pricing computations. The overall cost of these supercomputing facilities, in the range of several million dollars, limited their diffusion to the largest firms. Today, computational facilities ten times more powerful cost only of a few thousand dollars.
To place today’s computing power in perspective, consider that a 1990 run-of-the-mill Cray supercomputer cost several million U.S. dollars and had a clock cycle of 4 nanoseconds (i.e., 4 billionths of a second or 250 million cycles per second, notated as 250 MHz). Today’s fast laptop computers are 10 times faster with a clock cycle of 2.5 GHz and, at a few thousand dollars, cost only a fraction of the price. Supercomputer performance has itself improved significantly, with top computing speed in the range of several teraflops7 compared to the several mega- flops of a Cray supercomputer in the 1990s. In the space of 15 years, sheer performance has increased 1,000 times while the price-performance ratio has decreased by a factor of 10,000. Storage capacity has followed similar dynamics.
The diffusion of low-cost high-performance computers has allowed the broad use of numerical methods. Computations that were once per- formed by supercomputers in air-conditioned rooms are now routinely performed on desk-top machines. This has changed the landscape of financial modeling. The importance of finding closed-form solutions and the consequent search for simple models has been dramatically reduced. Computationally-intensive methods such as Monte Carlo simulations and the numerical solution of differential equations are now widely used. As a consequence, it has become feasible to represent prices and returns with relatively complex models. Nonnormal probability distributions have become commonplace in many sectors of financial modeling. It is fair to say that the key limitation of financial econometrics is now the size of available data samples or training sets, not the computations; it is the data that limits the complexity of estimates.
Mathematical modeling has also undergone major changes. Techniques such as equivalent martingale methods are being used in derivative pricing and cointegration , the theory of fat-tailed processes, and state-space modeling (including ARCH/GARCH and stochastic volatility models) are being used in econometrics.
Powerful specialized mathematical languages and vast statistical software libraries have been developed. The ability to program sequences of statistical operations within a single programming language has been a big step forward. Software firms such as Mathematica and Math- works, and major suppliers of statistical tools such as SAS, have created simple computer languages for the programming of complex sequences of statistical operations. This ability is key to financial econometrics which entails the analysis of large portfolios.8
Presently only large or specialized firms write complex applications from scratch; this is typically done to solve specific problems, often in the derivatives area. The majority of financial modelers make use of high-level software programming tools and statistical libraries. It is difficult to overestimate the advantage brought by these software tools; they cut development time and costs by orders of magnitude.
In addition, there is a wide range of off-the-shelf financial applications that can be used directly by operators who have a general under- standing of the problem but no advanced statistical or mathematical training. For example, powerful complete applications from firms such as Barra and component applications from firms such as FEA make sophisticated analytical methods available to a large number of professionals.
Data have, however, remained a significant expense. The diffusion of electronic transactions has made available large amounts of data, including high-frequency data (HFD) which gives us information at the transaction level. As a result, in budgeting for financial modeling, data have become an important factor in deciding whether or not to under- take a new modeling effort.
A lot of data are now available free on the Internet. If the required granularity of data is not high, these data allow one to study the viability of models and to perform rough tuning. However, real-life applications, especially applications based on finely grained data, require data streams of a higher quality than those typically available free on the Internet.
In its modern sense, financial engineering is the design (or engineering) of contracts and portfolios of contracts that result in predetermined cash flows contingent to different events. Broadly speaking, financial engineering is used to manage investments and risk. The objective is the transfer of risk from one entity to another via appropriate contracts. Though the aggregate risk is a quantity that cannot be altered, risk can be transferred if there is a willing counterparty.
Financial engineering came to the forefront of finance in the 1980s, with the broad diffusion of derivative instruments. However the concept and practice of financial engineering are quite old. Evidence of the use of sophisticated cross-border instruments of credit and payment dating from the time of the First Crusade (1095–1099) has come down to us from the letters of Jewish merchants in Cairo. The notion of the diversification of risk (central to modern risk management) and the quantification of insurance risk (a requisite for pricing insurance policies) were already understood, at least in practical terms, in the 14th century. The rich epistolary of Francesco Datini, a 14th century merchant, banker and insurer from Prato (Tuscany, Italy), contains detailed instructions to his agents on how to diversify risk and insure cargo.5 It also gives us an idea of insurance costs: Datini charged 3.5% to insure a cargo of wool from Malaga to Pisa and 8% to insure a cargo of malmsey (sweet wine) from Genoa to Southampton, England. These, according to one of Datini’s agents, were low rates: He considered 12–15% a fair insurance premium for similar cargo.
What is specific to modern financial engineering is the quantitative management of uncertainty. Both the pricing of contracts and the optimization of investments require some basic capabilities of statistical modeling of financial contingencies. It is the size, diversity, and efficiency of modern competitive markets that makes the use of modeling imperative.