The Bank Calculator
By Stephen W. Hiemstra
My transfer in 1995 to the economics department in the Comptroller of the Currency (OCC) initiated a period of intense learning, productivity, and networking. My particular unit, Global Banking and Financial Analysis (GBFA) developed a relational database in SAS to store and more easily retrieve call report data on commercial banks and thrifts and more generally tasked with financial analysis of community banks. Because most economists have some facility with using SAS, this database enabled our unit to undertake routine oversight of the financial condition of the banks.
My initial work in GBFA involved supporting advanced risk analysis research being undertaken by my colleagues, Tom and Swamy. Swamy developed an econometric approach at the Federal Reserve which was implemented in FORTRAN and ran on the UNIX operating system called the Stochastic Coefficients Estimation Program (SCEP). My task was to migrate this software over to run on Microsoft Windows and to make it accessible to researchers not familiar with FORTRAN. The migration to Windows was fairly straightforward. I then wrote a Windows program in C++ which prompted the user with menus to fill in the required variables and run the FORTRAN as a background process. The harder part was writing the hypertext documentation to explain to users how to understand and user the procedure.
This new program cut the time required to compete an estimation from weeks to a couple hours, making research substantially easier. Tom Lutton and Swamy Paravastu completed a number of journal-quality research papers over the next couple years. Meanwhile, I became acquainted with the procedure and became good friends with mathematician and programmer at American University, Ilok Chang, who had developed the FORTRAN program. Over nights and weekends during the next several years, we collaborated on development of an assembly language implementation of a matrix class for interval mathematics (1996-97). As part of my validation work on this matrix class, I developed a small calculator program to speed up the computations.
In addition to our research work in GBFA, we were encouraged to support the work of the bank examination staff with both macro-economic reporting and financial analysis. For the most part, our macro-economic reporting was ignored, which was a source of much consternation because a monetary crisis was developing in Asia. At one point, we were called upon by a large New York City bank to assist with working on the Asia issue and traveled to New York to brief and be briefed on the issue, but there was little appetite in our office to follow up. When several weeks later the Thai Baht crashed, the attitude about macro-economics did not change and we gave up on our reporting.
The support for financial analysis was different, in part, because everyone was convinced that they could do themselves, even if inadequately. The breakout project in financial analysis came when we were asked a second time to assist in reviewing liquidity risk. Liquidity risk kept coming up because there was an examiner with a special interest in liquidity issues who would periodically worry people enough to have management request an assessment. The first time this happened, we undertook a lengthy literature review and attempted to measure liquidity risk with a research effort—no one understood our work and it was dropped. The second time we received a request, I proposed a brief study of the liquidity ratings given by examiners to each and every national bank. With our new data system, this study was easily undertaken, briefly summarized, and widely cited. This liquidity study was ignored by the economics staff, but was loved by management so we found ourselves fielding more questions about the financial condition of national banks.
Because of my background in agricultural economics and bank examination, I found myself undertaking quarterly studies of the condition of agricultural banks. In response to this requests, I developed a databook of all the agricultural indicators found in the call reports. This databook was warmly received and I was invited to participate in an agricultural oversight committee which met from time to time with examiners from across the OCC. Eventually, I was able to convince OCC managers that, unlike in the 1980s, agricultural banking no longer posed a systemic threat to the national banking system and routine reporting on agricultural banks went from quarterly to annually and then to being dropped. The largest agricultural portfolio in the nation was held by the largest bank, but for that bank it was less than one percent of assets—in other words, agricultural credit risks were being adequately managed.
Between our research work on bank risk taking and our reporting on the financial condition of banks, it became obvious that economic research seldom had an impact on the culture of regulation while financial analysis, even if indefensible in an empirical sense, was routinely influencing administrative decisions. This problem caused our team great consternation, because we believed that our work was both theoretically and empirically sound. In the midst of this frustration, I began to see a disconnect between the contractual risks (credit and interest rate risk, in particular) which regulators followed with great interest and threats to the firms’ survival (liquidity and failure risk) which were often neglected. I coined the term, whole bank risk, to highlight this disconnect. Tom, Swamy, and I began to call ourselves the whole bank risk team.
The whole bank risk project had two primary components, one headed by Swamy and other I headed. My project involved improvements to a bank failure model which we developed in cooperation with the Federal Deposit Insurance Corporation (FDIC) and the Office of Management and Budget (OMB) using Swamy’s econometric approach. This initial failure model yielded a probability of failure which tracked the historical performance of commercial bank failures reasonably well, but the number was lower than FDIC was accustomed to seeing and they rejected the model. The model was accordingly written up and then ignored.
Later on, GFBA was tasked with developing new models for the OCC. A week-long series of meetings were planned in which staff unfamiliar with modeling sat and talked for days about how to develop models. This was extremely frustrating for those of us accustomed to modeling and being ignored. At that point, I had an idea—why not take our existing bank failure model and develop it into a Windows program which allowed the user to simulate bank failure probabilities in a calculator format, as I had done earlier with my interval mathematics validation? I skipped out of the meeting on Monday and returned on Wednesday to demonstrate my “Bank Failure Calculator” program.
The name changed to “Bank Calculator” to placate sensitivities, but the program itself was a great hit. Over the next 7 years, I spent about half my time estimating new failure models to add additional explanatory variables, validating the results, doing supporting studies, and porting the program to other computing environments, like SAS, Excel, and hypertext. I also gave demonstrations to numerous agencies across government interested in our approach. In its hay-day, the bank failure probabilities were updated for every bank in the national system and available along with supporting analysis to examiners across the OCC through the agency intranet.
 We envisioned this project accelerating the Human Genome project computations (https://www.genome.gov/12011239). However, in my validation work I discovered a weakness in the Pentium processor which would have required years of effort to resolve. At that point, we abandoned the project.
 Frustrated by the response but intrigued by the developing storm, I opened a commodity account and taught myself options trading and technical analysis nights and weekends.
 Bank examiners rate banks with 5 indicators each year: capital, assets, management, equity, and liquidity (CAMEL).
 Other people talk about enterprise risk.