This post is based on things I learned from, or began researching due to, a talk by Troy Magennis at LKNA ’14.
I had never thought deeply enough about risk to really differentiate between types. Since LKNA ’14, I have learned about aleatory and epistemic risk.
- Epistemic: uncertainty due to gaps in knowledge
- Aleatory: uncertainty due to variability or randomness [like throwing dice or flipping a coin]
Differentiating between the type of risk is important because they are mitigated in completely different ways. Epistemic risk is the easier type to deal with because it is something that can be overcome. We can endeavor to fill our gap in knowledge that is causing the uncertainty which creates the risk. However, we can’t remove randomness from a process like flipping a coin. We can, however, learn to handle aleatory risk better using past historical data with probabilistic models. The more we understand the probabilities, the better we can assess the risk we are facing. This is where Troy’s teachings on subjects such as Monte Carlo simulations are very helpful.
This is a topic I want to dive into at a much deeper level. A couple of pages I have found helpful:
- Epistemic Uncertainty (netjeff.com)
- Herding Cats: Both Aleatory and Epistemic Uncertainty Create Risk
What have you read that has taught you well in this area of uncertainty, risk and probabilistic forecasting?
August 20, 2014 at 12:46 am
IMO these two categories of risk are extreme ends of a spectrum. There grades between.
Flipping a coin follows the laws of physics. If we would know ALL involved forces in SUFFICIENT detail, we could predict the outcome. So it is as well an epistemic problem of insufficient knowledge. IMO there just different grades of lacking knowledge or different grades of effort to cover the knowledge gap. So it is about finding the knowledge gaps, which i can fill economically (profitable cost/benefit) versus gaps, where the costs of filling are higher than the benefits of closure.
Second, a gap doesn´t have only to two states: open/closed: A gap can be partially closed. There is a whole spectrum between knowing nothing and knowing everything. The real world scenario is about finding the sweet spot: where is the tipping point, where my knowledge acquisition costs are not anymore justified by the knowledge benefits ?