Data Forecasting - Granularity vs Time

Ralphr

Any statistics gurus out there ?

Looking at historical link utilization data and trying to forecast out 90 days from now.   Trying to decide what would be a more accurate predictor.  Would more granular data such as daily averages over the last 40 days  where I can exclude Saturdays and Sundays and calculate for core  business hours  be more accurate than lets say daily averages over the last 120 or 240 days where at best I could only exclude weekends ? 

mcbridea

Not speaking as a statistician but from experience, I would go for the 40 days of more accurate information. It better models the behavour you are forecasting. You could do the longer time frame that includes non-business hours then factor out the difference in business hour usage to non-business hour usage. This factor will usually be between .4 and .6 and varies with length of business hours and non-business hour traffic rates.