Models of Economic Success: Can HPC Help Turn Things Around?
Global economic turmoil affects us at both organizational and individual levels. As companies cut staff to reduce expenses, investors anxiously review
their portfolios. If we lose our jobs, we want savings to fall back on. Let’s face it: a lot of us are in a bad mood. We’re frustrated, nervous,
and even angry.
Can we look for hope in the financial sector investing more money? Geez, that’s a tough sell. That’s who a lot of people are mad at in the first place.
We watch the news. Reckless investments are what got us all into this mess in the first place. But before you disparage the idea, answer this.
Would you like financial institutions to do everything possible to make investments safer in the future? That’s where high-performance computing
(HPC) comes in.
The notion of using HPC to model portfolio performance against macroeconomic trends is not new, nor is it worth spending a whole column talking about
what a great idea it is. Rather, it is important to emphasize why HPC may still be a good investment, with an understanding of what the biggest
challenges are going forward.
HPC: Now More than Ever
Tabor Research has conducted numerous surveys of the HPC user community since the federal takeover of Fannie Mae and Freddie Mac last September. In
each survey, the overall outlook is consistent; the majority of users are planning to increase their total HPC investment.
If this seems counterintuitive at first, consider the plight of, for example, a U.S. automotive manufacturer. Surely, the company needs to cut costs
dramatically. However, it is also interested in increasing revenue, and it might do everything it can to bring its 2010 minivan to market sooner.
That takes engineering, and engineering takes HPC.
The link between HPC and top-line revenue is not unique to manufacturing. In most application areas, HPC usage is tied directly to revenue generation
(or, in academic and government sectors, the advancement of research or fulfillment of some other top-level goal).
The same logic applies to financial services. An HPC user at one financial institution told me, “In banking, everything is about risk analysis and
regulation.” The path forward to make investments safer and rebuilding consumer confidence is through more modeling, not less. We just need to
be realistic about what the challenges are and how to manage them.
More than Just the CPUs
The biggest challenge to HPC modeling in capital banking isn’t computation; it’s data management. In a collection of interviews with users in the industry,
that point got hammered home again and again. Users said: “We’re talking about hundreds of terabytes of data that’s all over the place.” “Finding
and validating data is the biggest time sink by far.” “How you label columns is critical to allowing others to find the data.” Conversations ranged
more toward metadata than megaflops.
The same challenge exists for insurance companies. Basic fields such as date of birth are well-defined for life or auto insurance policies, but not
for homeowners’ insurance. One user in that industry told me, “In reality, we grew up as a set of silos. No one thought we would eventually integrate
them. The customer data is represented differently in each silo.”
More to the current point, once modelers acquire the appropriate data, they need to do the right things with it. As a market research professional
myself, I can confidently attest: Analysis is only as good as the analyst. One user told me very candidly, “You need quantitative analysts who
know how to look at data and predict a trend. Some banks made some very bad decisions around subprime loans. What kinds of analysts made those
Models of Success
The path forward in financial services certainly requires more careful risk management. HPC will play a big role there, and it makes sense for users
to increase their investments. But to do so successfully, companies must consider the end-to-end workflow and analysis, from data gathering to
decision making. Consider it an investment for all our futures.