Frank Knight and Donald Rumsfeld had some thought-provoking ideas about the differences between risk and uncertainty, as well as their impact on decision-making. What can we learn from the philosophies of the revered economist and the former US Secretary of Defense?
For some time, I have reflected on the dynamics of the firm decision-making process under imperfect information, contemplating how we, as risk managers, can improve our communication with boards and senior leaders — as well as identify the boundaries of our increasingly analytical-driven risk management processes.
There is much we can learn from the seminal work of economist Frank Knight (who clarified the differences between risk and uncertainty) and former US Defense Secretary Donald Rumsfeld’s comments on the decision-making process. A fine line exists between risk and uncertainty that, when coupled with cognitive bias of risk and business management, can lead to faulty and potentially catastrophic decisions.
Frank Knight is well-known among economists for having delved into the differences between risk and uncertainty in 1921 in his book, Risk, Uncertainty and Profit. For Knight, risk management is not knowing the outcome from an action in advance; rather it’s having the ability to reasonably measure its likelihood.
More formally, according to Knight, there are both a priori and statistical probabilities that provide a means of assessing risk. An example of an a priori probability would be the mathematical representation of a credit loss distribution.
At the other extreme, Knight also refers to estimates used to ascertain the likelihood of an outcome — but these reflect a lack of or poor information from which to draw reliable inferences. This lack of information noted by estimates gives rise to “Knightian Uncertainty.”
Nassim Taleb’s Black Swan events align well with Knight’s definition of uncertainty. Risk management’s heavy dependency on data and analytics has transformed the way firms strategically position and manage their businesses. Still, nearly any industry can point to spectacular failures that posed catastrophic risk to the firm, customers, markets or the overall environment due to a misunderstanding between risk and uncertainty (amplified by inherent management biases).
Rumsfeld, on the other hand, famously adapted a framework leveraged in project management and strategic decision-making circles that relates the awareness of a risky outcome to knowledge obtained about it. To describe the world in which decisions are made, he colloquially distinguished between “known knowns, known unknowns, and unknown unknowns.” A summary of Rumsfeld’s remarks against Knight’s depiction of risk and uncertainty can be found in the table below. Risk management operates largely within Quadrants 3 and 4 of the matrix.
Lessons from the Crisis
To bring this matrix to life, consider how decisions about one of the riskier mortgage products originated during the housing boom — the pay option adjustable rate mortgage — were influenced by a combination of misunderstanding regarding risk and uncertain outcomes and inherent management bias. By the time of the financial crisis, the pay option ARM had been radically transformed from a product with reasonably well-known credit, interest rate and operational risk characteristics to one with a number of features that severely reduced the statistical reliability of forward-looking estimates of risk for these loans.
The underwriting standards were relaxed on several dimensions, leading to significant risk layering of these products. What’s more, there was a significant shift in borrower attitudes toward these mortgages that would amplify risk in ways that were virtually unknown (even by lenders) at the time, essentially becoming an “unknown unknown.”
The “known knowns” of the original pay option ARM were that it posed credit risk that reflected a lognormally-shaped loss distribution. Decisions to originate and hold in the portfolio the original version of the pay option ARM were based on solid historical data and well-controlled processes to manage these loans. In that regard, these decisions were based on “known unknowns,” since management was aware of the risk these products posed and had extensive historical data from which to model risk outcomes.
As these loans morphed into a much riskier version of that original product, subtle but critical changes in the dynamics of decisions about the pay option ARM product occurred. Performance experience on these new pay option ARMs was limited, due to significant changes undergoing the product and processes used to manufacture these loans. The lines would ultimately become blurred between “known unknowns” and “unknown unknowns.”
Borrower selection, payment shock and eventual contagion effects would come to vastly understate credit and operational losses over time. More concerning was that boards and senior business management made decisions to expand pay option ARM lending significantly, partly based on firmly-held views of the market and competition.
Eventually, the acceleration of pay option ARMs and continued relaxation of product attributes in the years preceding the crisis resulted in massive losses for a number of large originators, in some cases contributing to their demise.
Known vs. Unknown
When risks can be modeled reliably, it reduces the potential for cognitive bias to influence decision-making. Risks and profits can be measured with greater confidence, making decisions much easier.
In contrast, when awareness of risk is relatively unknown, the potential for bias to cloud decision-making increases as personal views and anecdotal stories that support closely-held views or biases creep into strategic discussions, filling the gap that risk analytics would otherwise provide.
When economic conditions are favorable, inherent human tendencies to more heavily weight recent experience occurs, consistent with findings in the area of behavioral economics. Coupled with herd mentality, recency bias becomes a powerful force in driving decisions in the absence of good information.
The risk and uncertainty decision matrix described in this article provides risk and business managers with several important lessons.
For risk managers, it is critical to understand how product shifts influence the quality of data and models — and to what extent these changes move the analysis from the “known unknown” to “unknown known” quadrant or even to the Knightian Uncertainty of “unknown unknowns.” For business management, the matrix demonstrates why it’s important for both the board and C-suite managers to resist the temptation to allow inherent human biases to fill the void left by analytical results, thereby influencing strategic decisions.