Introduction to Prediction Luke Gotszling Co-founder & CEO at fina"y.io luke@fina"y.io @lmgtwit September 11, 2013 | SFPython | San Francisco 1 

Shark meets cable http://www.#.com/cms/s/0/4557b69c-c745-11de-bb6f-00144feab49a.html http://www.youtube.com/watch?v=1ex7uTQf4bQ 2 

CPU graph 3 

Linear regression y = α+βx 4 

Linear regression Benefits: We" supported and straightforward calculation Built-in estimate of the degree of fit: R2 (“coefficient of determination”) Problems: Doesn’t handle cycles Questions about parameters (e.g. amount of entries used for regression and steps of extrapolation) 5 

EMA (exponential moving average / exponential smoothing / Holt-Winters) Image citation: http://lorien.ncl.ac.uk/ming/filter/filewma.htm 6 

EMA yt = αxt+(1-α)yt-1 y1=x0 7 

EMA Benefits: More recent data weighed more heavily Seasonality can be taken into account Problems: Relies on reversion to mean Divergence and multiple seasons in data Weighting options 8 

Other approaches Higher dimensional polynomial fits (and exponential) Fourier transforms Machine learning: neural networks... Bayesian RSI (relative strength index) and other methods used in technical analysis in finance 9 

Data bit.ly/sfpython_prediction_slides bit.ly/sfpython_prediction_notebook 10 

Thank you! luke@finally.io @lmgtwit 11