▪️If something involves ruin, probabilistic estimates are worthless. 100 people playing Russian roulette exhibit different properties from one man playing 100 times. After 7 trials, the man might face the ultimate death risk.
2/7 ▪️Only one of the potential paths will be brought into actuality. If unsure what to do next, you should ask yourself: 'Do you want this again and innumerable times again? ' Would you accept this 'eternal return' for your portfolio?
3/7 ▪️People love to make money, and hate to lose money. To survive financial storms, one must overcome this humanness. Avoid immediate gratification. Sometimes, this is more difficult than trading itself.
▪️Risk management is far more important than revenue management.
4/7 ▪️Tail hedging means losing money most of the time. Sacrificing arithmetic return for the geometric return. Most investors don't understand systemic risks.
▪️Forecasting is borderline impossible. You should abstain from forecasting, unless you are Nancy Pelosi's husband.
5/7 ▪️Sharpe Ratio has created the false dichotomy: either low risk, low returns or high risk, high returns. With tail hedging, it is possible to increase your returns by lowering risk.
6/7 ▪️The principal-agent problem will prevail as long as profits are shared between the financial manager and principal while losses fall on taxpayers and clients.
▪️Losses from a series of mediocristan risk events are highly predictable, given that they are IID (car accidents). The extremistan risks are catastrophes that cascade into even more risky events. Their aggregate losses are difficult to evaluate.
2/7 ▪️Catastrophic events violate LLN by nonlinearity and dependence, therefore losses cannot be calculated through CLT.
▪️1+a & 1-a. Perturbate your current position by +-a. If you are much better off in the +a position than the loss of -a, you are anti-fragile.
3/7 ▪️If the average of your payoffs in the 1+a and 1-a exceeds your current payoff, you are anti-fragile.
▪️antifragility is convexity.
▪️Most papers bullshit with correlation and p-value.