I'm not aware of any algorithm, but the methodology would be to make a determination as to the type of and amount of traffic from each site. For example, will there be large file transfers, backups, imaging and at what times of day. That and taking into account email, web traffic, voice?, etc. Not necessarily an easy task and it may just come down to best guess.
You also need to identify the traffic flows. If all traffic is going to be from hubs to the spoke, that's one thing. However, if there will be traffic from hub to hub, that essentially doubles the bandwidth at the hub as it will have to receive from the source and then send it back out to the destination on the same interface.
My opinion is that if you are going to have 20 sites ranging from 20 - 50 Mbps you'll need a bigger hub circuit. Even if all sites were 20 Mbps, that would be an aggregate of 400 Mbps. So if all 20 were at 1/2 utilization you'll still be at 200Mbps at the hub.
Not sure what type circuit your getting, but if you could get a rate limited Gig connection at the HUB, you could then find out the costs for different rates and it may be easier to adjust to higher rates as utilization dictates.
Hope this helps.
