When player ratings are inaccurate for a historical season, this is probably the reason: ratings are based on the player's overall career, not their stats in any single season. This is an intentional design choice. Football Mogul and Baseball Mogul are "general manager simulations". If player performance matched historical performance on a season-by-season basis, it would be too easy to build a dominant team every year.
For example, Terdell Middletown was a mediocre running back for Green Bay in 1977, coming off the bench and rushing for only 2.8 yards per carry. In 1978, he rushed for 1,116 yards and 11 TDs and made the Pro Bowl. Knowing this, it would be easy to trade for him at the end of the 1977 season — and just as easy to trade him away after 1978, before his performance falls off a cliff.
This graph shows the difference. In the default "simulation mode", Tom Brady has a great career, but it's impossible to predict his best season. In Replay Mode, his 2007 campaign will be godlike. But in 2008, New England has to build their offense around Matt Cassel.
*Because of the lack of historical stats for offensive linemen, several years around the current season are used to calculate ratings for guards, tackles and centers.