Buffet famously said loss avoidance is rule number one, and rule two is to remember rule one.
You buy flood insurance every year, even if it only floods once every 15 years on average, and even when it hasn’t flooded in 25 years.
If you make 10% for 9 years and then lose 20% on year 10 (1,886.36 from 1,000), you’d be better off making 8% for ten years (2,158.92 from 1,000).
If the 20-something losses all their savings, that sucks. If a 45 year old losses all their savings they have people to provide for. Which doesn't just suck, it's detrimental to his life and hapiness.
A 20 something has 40 years to bounce back, a 45 year old has 15. Time horizions dictate risks that can be taken.
A constant-fraction rebalanced portfolio has nothing to do with avoiding loss. It's purely about maximising growth. Such a portfolio, in the long run, outperforms all individual assets it is constructed from.
I agree with your general sentiment. My reasoning to get there is different:
I don't, for example, think anyone should invest all their capital into a risky asset. Not because it might crash, but because it performs poorly compared to the best investment (which is a balance weighted toward safe assets.)
Logarithmic utility corresponds to maximum growth of wealth (Kelly criterion), so insurance is actually often compatible with maximum growth of wealth.
How can something be negative EV yet maximise growth? Compound returns.
Insurance is only negative EV when considering a single period. The ongoing act of having insurance is positive EV in terms of growth. The way to get to that is to count EV as the geometric mean instead of arithmetic mean.
This is a very common mistake still, even though it was discovered by Bernoulli in 1734. I strongly recommend reading that paper. It is very easy to read.
High competition low profits would be well functioning for a consumer, low competition high profits would be well functioning for an investor
For the subset of people that would need an insurance without knowledge that could prevent that, the consequences should be distributed among all people. (Sure, there are exceptions if taking too big a risk.)
And personally I feel most medical issues and school should be paid by the state as it would be too unfortunate if an individual should face alone the consequences -- and possibly couldn't afford for an insurance, or is likely not to buy it because has other monetary issues.
As you point out, insurance is a good example of a <1 EV; its purpose is to reduce volatility.
Another example: lottery tickets in the occasional case where the EV>1. This is supposed to, for instance, lead a rational economist to buy a lottery ticket (or many lottery tickets!) when the Powerball jackpot hits some particular threshold, say 500 million. However, money isn't linear in terms of value to individuals, and for most people the difference in life impact between winning a billion dollars vs 500 million is not anywhere close to 2x - indeed the two outcomes are more or less effectively identical.
tl;dr: because money is not linear in the value it adds, EV is not a good optimization metric for highly skewed outcomes.
This is the right way to think of repeated bets (rather than in isolation) and Bernoulli's 1734 paper on it is a very readable intro to thinking in terms of the Kelly criterion.
Would I bet $20/day that something will happen to this car that would make the rental car company want to be reimbursed for?
Depending on rental car company the limits of scratches, dents etc .. can be very low.
So effectively it's a bet against you, other people and more generally the world.
For interest, there's a very common negative expected value bet that almost everyone is required to make: insurance.
We don't consider that gambling, in fact we often tell our parents to buy some when they fly on holiday.
Why? The answer touches on the lottery.
We care about not just the average case, we care about what might happen.
Regarding Kelly criterion, there's a good reason why people don't used exactly the amount it says. If you look at the risk, ie the chance your probability is wrong, there's a chance you are overbetting.