He did say he used the Monte Carlo method as well.

You can use cFIREsim to run Monte Carlo simulations. Results are usually worse than historical data. You portfolio has to survive random movements in the variables you set that might not happen in practice in the historical data set.

So there is pure monte carlo which is what cFIREsim uses, and then there are the professional level monte carlo analysis tools that look at more variables and the interconnectedness of those variables. Straight monte carlo just looks at mean returns and standard deviation. It isn't smart enough to look at how bonds and stocks normally perform against each other. It isn't smart enough to know that if stocks go up 50% 5 years in a row you are probably in a bubble so the odds of lower returns in the future increases. It isn't smart enough to look at inflation's effects on stocks/bonds.

Example: pure monte carlo could run 10,000 scenarios and one of them shows the potential for +30% returns every year for 40 years. From a math perspective it is possible. However we know it probably shouldn't actually be included in the possible results of the future. If you run 10,000 scenarios you could have one scenario where you have -30% per year for 40 years, and one with positive 30% returns for 40 years. Your $100,000 could be worth $10 or $10 billion... Some of the results look realistic, but your extremes get very very extreme.

Professional investment firm level monte carlo simulations might start with stocks average 10% returns and have a 15% standard deviation, bonds are X and X%, cash is X and X%, etc. But then they will say that if stocks are exceeding that average by a significant margin the probability that stocks could have lower returns in the future increases. Stocks could keep going up, but the 'odds' start to shift the other way after an extended period of really high returns. They are smart enough to say that if stocks are down 20% the odds that bonds will go up in value increases. So instead of very very extreme results you might see 10,000 possible solutions where the bottom one includes a few really bad bear markets or even a scenario that looks worse than the great depression and then some amazing scenarios with very good but still realistic future returns.

So the worst case might be -50% returns for stocks over 20 years, and the best might be stocks averaging 15% returns over 20 years but plenty of ups and downs to get that 15% average. So your 100k could be 50k or in the low millions. You get results that are actually probable, useful, and at least the possibilities near the mean look similar to what we have seen in history.

Short answer: I believe any monte carlo analysis used by Vanguard(or any other firm that size) is going to be a lot more advanced than what cFIREsim is using.