Stats Geeks – what method would you use?
Six Sigma – iSixSigma › Forums › Old Forums › General › Stats Geeks – what method would you use?
 This topic has 12 replies, 10 voices, and was last updated 12 years, 3 months ago by Jonathon Andell.

AuthorPosts

August 17, 2009 at 12:58 am #184880
Eugene ByelyakovParticipant@EugeneByelyakov Include @EugeneByelyakov in your post and this person will
be notified via email.Firstly, you might want to ensure your data is normally distributed and in control — otherwise predictions would be useless. Depending on the type of data you could be looking in using Xbar, NP, P and C charts.Secondly, predicting, to me, sounds as determining an equation that will show correlation between response variable and predictor(s). This is achieved by using regression analysis.
0August 17, 2009 at 2:15 am #184882
SeverinoParticipant@Jsev607 Include @Jsev607 in your post and this person will
be notified via email.Ummm… did you not read that he had “binary” data?
0August 17, 2009 at 3:50 am #52539I’ve got a set of timeseries binary data I’d like to discover patterns in, and hopefully be able to develop a predictive model for – not sure what’s the best approach to analyzing, looking for some suggestions.
0August 17, 2009 at 5:18 pm #184889
TaylorParticipant@ChadVader Include @ChadVader in your post and this person will
be notified via email.Siggysig
I found some interesting stuff on Google by doing a quick search of “Predicting Binary Data” may be of some help. Way over my head, but the premise seemed to be on the same line of what you are looking for0August 17, 2009 at 6:27 pm #184890I’m not exactly sure what you’re trying to model, but binary logistic regression may have some potential use. See the following example with binary output:
http://europe.isixsigma.com/library/content/c070418b.asp
JB0August 20, 2009 at 4:07 am #184928
Jonathon AndellParticipant@JonathonAndell Include @JonathonAndell in your post and this person will
be notified via email.Please tell us a little more. What is the process, and what outcomes do the data represent?
0August 20, 2009 at 1:06 pm #184940Not a process per se – the binary is based on spot market electricity prices.
Basically I have the forward price, established the day before, and the realtime price, established in the trading day. I’m trying to determine how best to predict whether the forward price will be higher or lower than the real time price.0August 21, 2009 at 1:28 pm #184953Why don’t you see, first, what you have when you align the forward price and the realtime price on the projected/real date? Then you could use a 2nd yaxis to run the difference. Might want to include some consideration for factors that might have affected both the projected and the actual prices. Sounds like a fun analysis! (Why would the type of data affect the analysis? I’m not getting that part.)
0August 21, 2009 at 1:42 pm #184954You do not have to use binary data. If you have the before and after values, then do all your work on the difference. This should be continuous and if done correctly will be a more powerful analysis. Good luck
0August 21, 2009 at 3:51 pm #184960SiggySig:Your data is not binary(high/low) it is continuous ($/KWhr)General advice? Logarithmically transform your price data Include daily average system wide load for California prices Ignore ambient temperature Capture weekly variation using the autoregressive structure of your models Use dummy variables for holidays and long weekends Consider using a regime switching or mean reverting jump diffusion models to handle spikes Read this paper: http://mpra.ub.unimuenchen.de/10428/Cheers, Alastair
0August 22, 2009 at 12:40 am #184964
SeverinoParticipant@Jsev607 Include @Jsev607 in your post and this person will
be notified via email.So the lesson learned here is when you don’t present enough information about your goal and what data you have at your disposal you will wander about with obscure answers for days, but when you give the right info you’ll get a meaningful response within about a day.
Great information Alastair! 5 stars.0August 24, 2009 at 2:43 am #184974
KluttzMember@UnionofConjoinedScientists Include @UnionofConjoinedScientists in your post and this person will
be notified via email.The answer you’re looking for isnt a traditional Six Sigma tool. Stochastic modeling / Monte Carlo simulation is the correct method for your problem. Any other tool mentioned here will be static as opposed to probabilistic.
That will be twelve thousand dollars please.0August 24, 2009 at 4:44 am #184978
Jonathon AndellParticipant@JonathonAndell Include @JonathonAndell in your post and this person will
be notified via email.The data is only binary if you convert perfectly good continuous data into vastly inferior discrete (“binary”) data. You could consider some time series models, or perhaps some regression analysis if you also have potential Xvariables (don’t forget higher0order terms and “lag” functions). Either way, never make the mistake of degrading continuous data into discrete.
0 
AuthorPosts
The forum ‘General’ is closed to new topics and replies.