Scaling and Memory in Return Loss Intervals: Application to Risk Estimation