Generally while calculating forecasts the formula used is
% error = ((Actual/Forecast) -1) * 100
However while forecasting trends this could be misleading. In case a forecast is made for an increasing trend and the actual values start decreasing, even though the % error is small, the prediction is completely wrong.
Is there any other method of calculating error or accuracy where an under-forecast or over-forecast could be suitably accounted for?
P.S. the error should be calculated for each forecast/actual value and not for the entire data set.