We don't know the exact way they calculate the numbers, and what the difference is between hourly and monthly for example.
Most likely the current hour/month's sales are weighted high, and past hours/months sales weighted progressively lower. This means short term fluctuations can affect the rank significantly, and make it not represent the average sales, but if it consistently stays in a range for a long enough time then the rank should reflect the average sales.
If the monthly ranks feature data from past months, not only from the current month, then it's not 100% representative of who won the month on Amazon, and therefore of NPD. Again, if it stays in a consistent range month-to-month then it should be representative, but if there is fluctuation in a month then it may not.
Still the Amazon ranks have a fairly high rate of prediction. I don't think we should dismiss them as a data point just because once in a while it's not indicative. VGC is often not predicting on target either.
