In generating expected returns for asset classes using historical data, investors should incorporate the longest period possible. However, investors should also examine whether the period includes a variety of market and economic environments and should test multiple sub-periods to mitigate the bias that may result from arbitrary starting and ending points (known as endpoint bias).
Even historical returns for periods as long as twenty years may not prove a useful guide for generating expected returns. To complement a historical analysis of asset class returns, investors may benefit from forward-looking scenario analysis, based on an understanding of the fundamental drivers of historical returns.