Rolling Averages, exponential smoothing, even regression are largely designed to generate forecasts which "wring out" large variations in actual values from forecasts, trying to get to the underlying trend. That makes sense in many ways, most importantly because random variation is random, and we won't be able to predict it (at least not with any methods we'll cover in this class). Some large swings in actual values are not random (changes in underlying operations, seasonality, etc.) and can be anticipated and incorporated into forecasts. But sometimes large, non-random variations are difficult to anticipate and we might only realize they are not random in hindsight. The better you are able to recognize situations where "un-anticipable" variations might impact your forecast, the better you will be able to create more accurate forecasts. That ability to recognize a changing situation that could lead to changes in the variation of actual values comes largely from experience and observation. So, describe a time in your life when you experienced a big variation and whether or not, in hindsight, you believe there was an indicator of the potential for that variation prior to it occurring.