They could easily create a better simulation with a moving block bootstrap method. That is by sampling consecutive returns (for example for a whole week) and create a new time series with them.
Yep, I have a tool that is similar to this and this is what I did.
Unfortunately, if you are working with daily returns you need consecutive returns for way more than a week. You have correlated returns, in particular there are Friday-next Monday correlations that are important in the tails. You also have volatility clustering/asymmetries over daily periods (i.e. high volatility tends to be followed by higher volatility and the volatility responds differently to up vs down moves) and this tends to last way longer than a week.
It is very tricky stuff. In the real world, you will often find managers grouping based on their knowledge (i.e. X-Y was the 2008 crisis) and testing their portfolios against that. It is rather unscientific but it works (i.e. in this case, you might do something like a Markov model with a transition matrix of the daily probability of moving between volatility states, and then sample longer blocks from groups based on the state).
A simpler option (what I did) is to just look at yearly returns, and sample across countries (just using the US is horrible cherrypicking).
