I don't think this is sound advice. The Kelly criterion has been thoroughly explored in the finance world and pretty much everyone agrees it's way too optimistic in situations where it's hard to determine accurate probabilities, such as real life. It's given rise to many alternatives such as Half-Kelly, Kelly minus constant, etc, but that just goes to show you how inaccurate of a proxy it is. Insurance companies have a reasonably good grasp of probabilities because they have actual data. They can see how many of their clients actually use the insurance and when and this gives you real inputs to go from. It's complete nonsense to make up a number like oh my car has a 33% chance of developing a big fault this year out of thin air, and to then make financial decisions based on that.
This was my exact thought when I hit that point in the article. How am I supposed to know the probability to make these calculations? And if I make up numbers that "seem" reasonable, how is that better than buying insurance to sleep better?
Many of these probabilities can be looked up in official statistics. Then you can adjust a tiny bit depending on whether you think you have more or fewer risk factors than the average.
When I feasibility tested by asking my wife to guesstimate a couple of these she ended up very close to official statistics. The first-level trick she used is Fermi estimation.
You can totally do it and it's worth practicing. There are even competitions in it open for anyone! I like the Quarterly Cup at Metaculus. The next iteration starts early January.
I don’t believe I can come up with accurate numbers like that. I have no chance against a big company with a bunch of actuaries who do that for a living.
I think I really can sometimes tell if I’m riskier or safer than the average person. I can observe other people’s behavior and compare it to mine. But even then, my impressions will be subject to a lot of biases (eg. everyone thinks they are an above-average driver).
If I assume the insurance company has done the math so they make a decent margin, and if I’m confident I’m a better or worse risk, it’s possible I can use that reasoning to guess whether insurance is a good value for me.
It seems like you're assuming a zero-sum game here where you are hoping to earn more from insurance claims than you paid in premium. This is not how insurance is supposed to be evaluated, as TFA points out.
You're not against the insurance company. Your decision is relatively separate from theirs.
If I believe insurance is well-priced, the cost ought to be pretty close to the value it’s providing me, maybe cheaper if the insurance industry has some advantage like good rate of return or higher if they are uncompetitive. That makes me indifferent to whether I have insurance in a lot of areas. Therefore, I’m not spending a lot of time purchasing insurance. Nothing against the industry.
Given I’m unable to produce the probabilities necessary to compute the Kelly criterion, I can’t use that to make the decision. Instead, I have to use dumb heuristics like “I need liability insurance to get rid of my risk of ruin” or “I can easily afford to replace my car, but my passion is parking underneath dead trees during storms, so I should pay for comprehensive coverage.”
In either case I have no idea how to compute the probabilities, but I’m still pretty sure I made the right decision.
> If I believe insurance is well-priced, the cost ought to be pretty close to the value it’s providing me
The value of insurance to the buyer depends on the buyer's wealth. The insurance the electronics store offers on the tablet I buy is well-priced for their target audience but not for someone with my emergency expense buffer.
But how do I evaluate how the magnitude of effect of my income on the value I get from the insurance (relative to the average buyer who sets the market price) if I’m lacking the figures to compute the Kelly criterion? Is it useful to me at all as a consumer? Is there version of the formula I can use if I want to implicitly accept the work of the insurance company’s actuaries, but skew the result to account for how I differ from the average customer?
That's an interesting question! The actual risk assessments thr insurer does are obviously confidential, but you're asking whether we can reverse engineer their premiums into underlying probability estimations? So we'd still be estimating probabilities, but based on information leaked from insurers instead of first principles.
I'd consider fractional Kelly (whether half, quarter, two-thirds, etc.) still a member of the Kelly family. After all, it arises when one performs a full Kelly allocation on some fraction of one's wealth, keeping the rest out of the market. It's not that Kelly is an "inaccurate proxy" -- it is provably the allocation that maximises growth -- but that people for various reasons don't want to maximise growth. They prefer slower growth in exchange for other properties, such as more cash being available for emergency needs.
> It's complete nonsense to make up a number like oh my car has a 33% chance of developing a big fault this year out of thin air
Are you claiming regular people cannot learn to make calibrated probability judgments and/or look up car failure rates? Maybe that is the problem with the Kelky-based framework: it requires forecasting a probability distribution and many people are not great at that, although they could learn it.
Since it wasn't clear from your comment, what alternative framework are you proposing for the insurance decision?
Insurance isn’t about maximising overall outcome (we know it doesn’t do that!) It’s about managing critical (often existential) risks. You buy insurance when you need to protect against an unrecoverable loss (house burned down, car was stolen, permanently injured and unable to work, etc.) that you can’t just “ride out” and self-finance.
...but it does do that under the assumption of compounding, and that's part of why it exists. I recommend learning about the Kelly criterion and E log X strategies to see why.
The key insight is we shouldn't look at the arithmetic expectation of profits of isolated bets beccause that causes us to overinvest in uncertain profits and underinvest in insurance.
