year 1989. We roll over the forecasting period of an equal 16-year length until we use the
estimating period of 1989–2004 for the predicted value of 2005. The Theil’s U statistic
captures the average forecast accuracy for each firm over the 17-year period of 1989–
2005. A low U statistic indicates high predictability or high relevance, and vice versa.
Therefore, Theil’s U should be negatively correlated with FCFO.
For the full sample, the Theil’s UFCFO statistic is calculated for each industryyear
from 1974 to 2005 (32 years). For example, the Theil’s UFCFO statistic for