One could argue that in the history of the world, it has never before depended on models as much for life and death decision-making as it has during the last couple of months as a result of COVID-19.
Pandemic or epidemiological models have heavily informed governments and driven their decisions with regard to their response to and preparation for COVID-19. These models are used in multiple areas, for example to decide on intervention measures (e.g. length of and easing of lockdown), budgeting for National Treasury, surge / peak planning to determine resource shortfalls and tactical resource allocation. Model risk events in any of these areas can have massive implications for society as a whole.
COVID-19 forecasts are subject to significant uncertainty, given how new this disease is, how different the disease has played out in the various countries and how sensitive the model outputs are to these highly debatable assumptions. As a result, these models have been widely criticized.
Given the above, model risk is very much present in the new reality that we are living in today. What if the models are wrong? How wrong might they be? What if the resulting decisions taken by governments are not optimal? Were the models being used for the various decisions fit for purpose? Did the users really understand the limitations of the underlying model outputs?
This presentation will focus, with reference to and using examples of various COVID-19 models, on why it is important to manage model risk as well as the key principles that need to be adhered to in order to manage model risk, especially where the consequences of not doing so and the uncertainties in the models used are so significant. The proposal is for this to be an interactive session, using Menti to engage the audience.
Key practical outcomes:
Why do we need to manage model risk?
How can we manage model risk when dealing with significant uncertainty in assumptions?
What are the key model risks that need to be taken into account in decision making?
How managing model risk change in the future as models become more complex?