Measured as autocorrelation of absolute or squared returns, clustering is strongly positive and decays slowly, often still visible at lags of a month or more. It is one of the most reliable regularities in all of finance, present in every liquid market that has been checked.
This is what makes garch and ewma work, and what makes volatility-targeting more than a gimmick: if today's volatility forecasts tomorrow's, you can scale exposure ahead of trouble rather than after it.
It also means independent-draw simulations understate drawdowns. Shuffling historical returns randomly destroys the clustering, producing tidier equity curves than reality allows, which is the argument for a block-bootstrap.
Related: garch, ewma, market-regime, block-bootstrap