Businesses routinely make operational or strategic decisions in uncertain environments.  When faced with the economic crisis stemming from COVID-19, such decisions may carry greater consequences than in the past.  Moreover, identification of the causes of various financial outcomes associated with COVID-19, both retrospective and prospective, matters a great deal for such business decisions and may also influence decisions around financial reporting, labor, and insurance claims.  Causal attribution of financial performance may also be important for valuing businesses or sectors as the basis for post-COVID-19 lending or equity investments.  

Separating causal influences from others in the present case, however, may be more complex now than ever before.  For instance, a recent class action lawsuit in Canada is seeking insurance payouts relating to three categories of harm.

• Loss of revenue caused by a decrease or elimination of customers after social distancing advisories
• Loss of revenue caused by federal, provincial and municipal orders that restrict operation or entirely close businesses
• Loss of revenue caused by “the costs of addressing physical damage to business premises due to the presence, release, discharge or contamination of COVID-19 at the business premises…”¹

Practically classifying losses into these or other categories can be difficult, however, because causation is often subtle.  While it is a simple concept to think of a cause as something without which some effect would not have occurred, determinations of causality are beset with potential pitfalls and must be carefully modeled.  Fortunately, practitioners have a wide set of tools available, which, if applied appropriately, can yield insights around attribution of financial losses or risks.

Standard accounting provides a backward-looking snapshot of performance.  Accounting measures can be used to measure how much a business was impacted ex-post by a certain event.  But they will not necessarily provide insight into what caused the loss. 

Economic loss attribution follows economic principles.  That is, the cause and effect relationship should follow a logical theoretical path that can trace the cause to the effect.  Such models often seem quite complex, but sometimes they are, in fact, quite simple.  Economists routinely work with a variety of widely accepted models in order to make sense of cause and effect relationships.    

Econometric models, including regressions and other forms of statistical analysis, are often used to evaluate relationships between variables while controlling for confounding factors.  Key variables influencing the process are included and critical elements of timing are considered.  Appropriate modeling techniques and functional forms are also weighed, along with potential biases within the underlying data.  Comparing the results of such models with underlying hypotheses and economic theories often allows practitioners to draw inferences of causality.

Even after taking such factors into account, the primary inferential criterion for causality can be, in fact, quite subtle.  Take, for instance, the classic example of treating a headache with aspirin.  The aspirin may provide relief of a headache within an hour only 30% of the time.  But if headaches only go away within an hour 5% of the time without aspirin and the difference is statistically significant at conventional levels, we can conclude aspirin causes the headache to go away (assuming the model controls for all other factors).  So even if aspirin doesn’t work 70% of the time, a causal relationship can still be inferred consistent with the treatment.  While it may seem at first glance easier to conclude causal effects using the treatment approach over a simpler comparison, sometimes it is harder to do so.  In the above example, for instance, it may be harder to distinguish 30% from 5% (depending on the distributional properties) using the treatment approach than it is to separate some higher percentage from zero using a simple comparison.

Machine learning and Monte Carlo simulations may be used to apply the causal insights gleaned from economic modeling and provide robustness to forecasts of future risks or performance.  The results from regression frameworks or statistical models based on historical data, including accounting data, can often be used as inputs to such analyses in order to gauge the effects across thousands, or even millions of complex path-dependent scenarios, yielding insights that may not initially be visible to the practitioner.

Together, accounting records, economic logic, and statistical analyses that point in the same direction can support an inference of causation, which is valuable in a wide array of applied settings.  Given an expectation of a great deal of causal analyses arguing in support of (or contrary to) a wide variety of economic outcomes from the COVID-19 pandemic and policy responses, it will be more important than ever to bear in mind the logical principles and available tools in evaluating arguments in the press, among investors, and in the courts.