I touched on this previously, but I want to address it again.
There is widespread belief in what is sometimes called “time diversification.”
The question is really: is there mean reversion in the equity risk premium (ERP) over time? If so, if you have a bad ERP experience early it reverses later so that if you have enough time you capture that and get back to (better-than) even over a more conservative allocation. There is some evidence of that but it’s pretty small. Your risk does not decrease with time. Samuelson made this point 60 years ago, but for some reason most people still think it works.
That paper was the first time (I think) the “myth of time diversification” was debunked. I.e. stocks do not get less risky with more time. The probability of loss decreases, but the magnitude of shortfall grows proportionally. (Again, there is an argument that longer-term stock returns are somewhat mean reverting so that the volatility/risk does not increase quite as much as theory would predict – i.e. with the square root of time. I would actually tend to agree with that.) Most people misunderstand the concept because they confuse “very unlikely to happen” with “can’t (or won’t) happen.” Mark Kritzman wrote a clearer explanation on this topic here, but you may not have access. It’s also in his wonderful book which I highly recommend.
The point Samuelson and Kritzman are both making is that diversification only works cross-sectionally, not serially. If you invest all of your funds simultaneously in 100 risky ventures (1/100th in each) of which most have decent returns, a few have spectacular returns, but a few also become worthless then you win. However, if you invest all your funds in 100 risky ventures one-at-a-time, rolling gains into the next venture, if any one has a negative 100% return then you are wiped out.