im taking a time series analysis course right now and have to come up w/ a data set for the semester. i was interested in toying w/ nfl/ncaaf stats for the course. so far the professor has related the material towards how the brain responds to visual and auditory stimuli. a lot of stuff on dynamic systems so far. his examples are more focused on univariate data and i was wondering if you know much about time series analysis and how it may apply to forecasting sports on a multivariate level. if you have any input or opinions on this id appreciate it!
-
-
That's kind of an open-ended question. What did you have in mind? Time series certainly appear everywhere.
If you're looking to uncover a profitable model using time series analysis, one trick to try is to come up with a simple GLS specification based on your sport-specific knowleddge, treating the residual as an AR(1) process. This forms the basis of many a statistical arbitrage strategy on Wall Street.Comment -
well my objectives arent really set in stone but i was looking into toying w/ some football data over say the last 10-15 years but was wondering how time series analysis is on a multivariate level since im assuming my data matrix will consist of nested matrices for each team? now im not really sure how well time series analysis applies to forecasting yet (noticed there's a chapter on forecasting my book) but i am either planning on observing how teams change, grow in success, etc etc through time and possibly predicting their success in the future (whether it be game by game, year by year, or change in tendencies through time). is time series analysis primarily focused on regression analysis?Comment -
Within the context of forecasting you can think of time series analysis as a subset of regression analysis.Comment -
would a simple GLS be the right idea in this case? i would assume using this procedure in a sports situation that could generate many outliers would not be the right procedure. i was thinking about running a robust regression and generate a rrline through iterations to calculate the most convenient line through the plotted data. i guess i could transform the data if i felt certain variables were non-normal which is somewhat the same thing. as for treating the residuals of the regression as an AR(1) process, i am not too familiar with the process. are arma models generally determined by ls or could i create a model based on a more robust procedure?Originally posted by GanchrowComment -
What do you mean by "in this case"?Originally posted by Quebb Diesel
Exactly what underlying process are you seeking to model?Comment -
using LS in the presence of outliers. i think im just having a hard time conceptualizing how to perform statistical procedures to calculate predictions? is that what an AR process does? it appears to estimate some alpha for lags of your X to generate a predictive X...Originally posted by GanchrowComment -
I['m not sure I'm understanding ... are you asking for a general refresher course on forecasting or ARMA? Either way I can recommend Econometric Analysis by Greene. If you're particularly interested in time series I can also recommend Time Series Analysis by Hamilton.Originally posted by Quebb Diesel
And my previous post should have read "Exactly what underlying sports process are you attempting to model?"Comment