03-20-2015 12:17 AM
Below a data mining approach that you might want to try. If you want to go into time series analysis, you probably more about it than I do! It's been a while...
Data mining approach
Back in the day in grad school at the Institute for Advanced Analytics one of our teachers described a really neat idea:
Use a linear regression procedure to output the parameter estimates (betas) of a trend for a 1-day, 2-day, 3-day, and nth-day prediction (or hour for that matter). Then use those betas as inputs for a predictive model that predicts the nth+1 beta. If the nth+1 beta is positive then you have a positive run-up that is quantifiable in terms of both the magnitude and also the accuracy coming out of your best predictive model.
[If someone knows the formal name of this technique, please comment about it! really interested!!!]
Lenon, give it a try with this approach. If you run into trouble, happy to dig up that code or try to reproduce it.