Anyone working with spreadsheets knows how useful and indispensable they can be. At the same time, it is also well known how easy it is to make an error, perhaps in an embedded formula or a cell reference. A colleague of mine who works in the area of financial model risk has spent a good deal of time and money fixing spreadsheets that have errors in them—errors that can have significant impact on financial results and public reporting of financial results. This chapter looks at this risk and what can be done to mitigate it.

In 2013, we learned that spreadsheet errors may also have been indirectly responsible for extending the financial crisis in Europe a year or two longer than it might have otherwise been. It was discovered that a spreadsheet error was behind the results of a study that was influential in the setting of austerity measures in Europe post financial crisis.¹ These measures, albeit indirectly, led to job losses across Europe and the United States. If these losses occurred as a result of an influential piece of research that was based on a spreadsheet error, this surely qualifies as a very significant operational risk event.

So what happened exactly? Two Harvard professors, Carmen Reinhart and Kenneth Rogoff, wrote a research paper in 2010 called “Growth in a Time of Debt.” The paper by two very well‐known and respected economists claimed that there was a close correlation between a country's growth rates and a country's debt level where it exceeded 90 percent of gross domestic product (GDP). Though the professors say they never made the claim that the relationship was a directly causal one, it did not stop others from drawing that conclusion. Indeed, policy makers in Europe and the United States widely cited the study in prescribing austerity measures aimed at bringing down debt levels.

Other economists, however, in reviewing the research results were unable to recreate the results. Finally, after a research team at the University of Massachusetts revisited the data and recently published its results, it became clear why that is.² The key finding, that countries with over 90 percent debt have negative growth rates, is only obtained when certain important data points are excluded. Furthermore, it seems that some of these data points were excluded because a formula in the spreadsheet did not include certain rows of data. The new study specifically obtained the correlation claimed by Rogoff and Reinhart by excluding Australia, Austria, Canada, and Denmark. Once these data were included, the authors of the new study showed the average growth rate for countries with over 90 percent debt ratio is actually 2.2 percent.

So can we say that this spreadsheet error led mistakenly to austerity measures that, in turn, have needlessly led to millions of job losses in Europe and elsewhere? Of course, there were many economists and policy makers in favor of austerity measures and the importance of maintaining low debt levels well before this paper was published. No lesser a figure than Margaret Thatcher would have been an honorary member of such a group, for example. There were also other well‐known flaws in the Rogoff‐Reinhart arguments that policy makers chose to ignore when citing Rogoff‐Reinhart's study in support of austerity measures—for instance, that the low growth rate preceded the high debt level rather than vice‐versa. At the same time, the episode is another reminder of the operational risks inherent in Excel and other spreadsheet programs. Firms that spend hundreds of millions of dollars in new technology annually have pockets of the firm reliant on formulae hidden in obscure, nestled spreadsheets. Yes, it would be nice to ban Excel from such uses, but unfortunately, ease of use and flexibility make it strangely addictive. In this context, firms should continue to apply resources to identify and ensure close oversight of any spreadsheets that inform the firms' books and records or any other key production data. Furthermore, continued migration from such dependencies should be part of this continuing effort. Economists should do the same.

Mitigating the Risk

Unlike a small team of research economists, large investment banks deploy spreadsheets in very large numbers across a number of functions. Though it is well known that Excel is highly error prone and is not suitable for use as part of a mission‐critical information system, it very often is. While information officers and business partners yearly repeat their mantra that they will replace spreadsheets with appropriate enterprise platforms, spreadsheets have a habit of wiggling their way back in. There have been many examples in the last few years of functions having a critical dependency on a spreadsheet that was implicated in an operational risk mishap of one sort or another. Such incidents have included: misreporting of the value of a financial asset,³ misreporting of a valuation at risk (VaR) model,⁴ and misreporting of various credit models used to calculate credit capital calculations for the federal banks stress tests.⁵ Behind all these incidents is the continued use of Excel for present‐value calculations, profit and loss analysis, VaR modeling, complex security and positions valuations, and many other uses. The proliferation demonstrates that it is impossible to contain the spread or use of spreadsheets—they are simply too useful, too flexible to ignore. Thousands and thousands of spreadsheets can be found on the networks and hard drives of analysts and traders, some of them developed on the fly, others incorporating sophisticated programming and code. Some of the spreadsheets contain significant outright errors; others may contain small errors that can change in significance, depending on the data. Many investment banks alerted by the significant impact of errors in spreadsheets impacting regulatory submissions, including quarterly financial statements released to the public, have started to put in place programs to mitigate the damage. The first step in such programs is to identify the purpose of each spreadsheet that is in use. It is important to distinguish between spreadsheets that impact or help to calculate the books and records of the firm or contribute toward the calculation of the market risk that the firm carries versus spreadsheets that help bankers make investment decisions. The second step is to identify spreadsheets that include complex formulae and code. Once you have isolated these “high‐risk” user tools, you can put in place a remediation program aimed at identifying errors in the code or in the formulae. This is no simple task, but putting in place such steps is an important part of any operational risk remediation program. There are tools available today that can help to automate this prioritization and error identification process and should be utilized wherever possible.