Thread by @WittyUsername30

Some highlights / takeaways / thoughts / comments from #RWRI 14, Day 5:

1. Optionality provides you with convexity: you can throw out the bad option and choose a better option.

Optionality is worth something, so the whole idea is to avoid paying too much for it. Many times, it isn't priced in at all (e.g. a restaurant reservation that you can cancel for free if you find a better one).

2. In any field, what you actually need to know is not what outsiders think you should know. As a practitioner, you gain a deep understanding of the properties that outsiders (e.g. academics) simply can't obtain.

3. The best indicator of future economic success is a lack of resources. You don't have much to lose, so you are convex.

4. The problem with most models is that they are static and don't factor in the 2nd+ order effects. They don't factor in the complexity that occurs when randomness / stress is injected into the system.

Most models plug the average value into a formula. Instead, they should be averaging their formulas across a variation of values, and using the results to determine if they are in Mediocristan or Extremistan.

5. "There exists no estimation without estimation error."

If someone tells you there is a 0 probability, and you ask them why, and they tell you "I estimated it" - run away.

There is *always* an estimation error. So an estimate of zero means the true probability must be larger than zero.

6. When you have maximum uncertainty about a binary event (e.g. who will win the election), the probability will be 50%. So the more uncertainty there is, the closer the probability is to 50%.

Example: Trump entering the Presidential race introduced a lot of uncertainty. Some election forecasters put his odds of winning at < 1%, <5%, etc., when they should have had it near 50% because of the high level of uncertainty.

7. Just because a data point is highly unexpected doesn't mean you should assume it is wrong / irrelevant.

Your simplified model of the world is more likely to be wrong than your unexpected data point.

8. In Mediocristan, you can run use traditional stats (mean, standard deviation, bell curve, etc.).

In Extremistan, these more traditional tools are unreliable. Instead, power laws and Extreme Value Theory (EVT) should be used.

9. "Scientists are not scientific. Science is scientific."

Just because someone says they are a scientist doesn't mean they follow the rules of science or understand convexity / fat tails.

10. "You're not trying to get a quantitative result. You're trying to get a property in the most robust way that you can."

In other words, power laws / EVT isn't about precision. It's about gaining qualitative insights about a distribution.

11. With power laws, the lower the alpha, the fatter the tail. The mothers of all fat tails are pandemics and wars (alpha < 1). These present the most existential risks to humanity, so any risk in these fields should be taken extremely seriously.

12. The main causes of fragility in the markets are too much debt and globalization (interconnectivity).

13. Disallowing companies / industries from going bust prevents corrections / improvements to the system.

When a firm goes bust, the system is purging something that wasn't designed to survive.

While the firm goes away, the assets (people, inventory, etc.) get reallocated.