Use SSO as an example (a 2x leveraged index). It has a high expense ratio and it's susceptible to leveraged decay. But the long-term numbers speak for themselves: https://stockcharts.com/freecharts/perf.php?SSO,SPY
Let's suppose you you invest at the worst possible time in the past 20 years, September of 2007 (that's actually worse than summer of 08), and that you held until the last couple years.
If you look at the charts (link above) which take into account fees and leveraged decay, you'll see that after 10 years it outperforms the index. More specifically, you'll see that after the crash it took the normal S&P index (SPY) 4 years to get back to 0% return, whereas it took SSO an extra 1-1.5 years (about 5-6 years from the crash), about another year (about 7 years from the crash) to catch back up to the the SPY yield, and another 4-6 more years to get close to 2x-ing the SPY (got close in 2017 and currently it's close again). Keep in mind those comparisons take as a given that you invested in the worst possible time (right before the 2008 crash) and also that the market goes up (as it did) 50% over the course of the 7 years after the crash (that's about a 6% annual average return).
The argument, then, is that this bet makes sense given time horizons of 20-30 years. I'm willing to make the bet that on average, over the course of say 25 years, the stock market will (1) probably have a 2008 style crash (2) and then, if that happens, it would go up at least 6% annually for 7 years after the crash (if it takes more years that's fine too).
And of course that's all absolutely worse case scenario. I expect it to do significantly better. If you'd have bought in, say, 2010, it was doubling the S&P pretty consistently over time and after 2 years (2010 onward) it was significantly above 2x maintained until the current day (because of the compounding effect).
So IT SEEMS TO ME the downsides of a 2x leveraged investment are offset over long periods of time even when those long periods of time include huge stock market crashes (i.e. 2008).