My planning software uses historical averages to run simulations. This is supposed to be more robust than modeling with "random" returns. But I've been thinking about this a bit more. While historical data do reflect things like trends, long periods of downturns, long periods of inflation, etc., the problem is they're actually very poor at future forecast. For example, if we use 1900-1950 data, it would be very bad at forecasting 1950-2000 data. Economies go through major changes. World events are unpredictable.
It just feels that continuing to use historical data to forecast retirement is setting myself up for failure. At the same time, history is all we have. It really does seem like a situation without a good solution.