Thread: Automated curve fitting for regression

Hi,

Would greatly appreciate help on the following:

1. Is there a way of automating the identification of the best fitting curve in a (auto)regressive model?
I understand I can layout in the different curves (linear, exp, ln, log, sqr, poly), calculate the adjusted R2 and then select the curve with the highest R2 as the one. Is there a simpler way?

2. Am also building a multiple regression model and looking to automate the identification of the best set of data to use.
Say I have 10 possible independent variables (not all need to be included in the regression), all of which could be lagged or not lagged. This gives a great deal of possible combinations and takes time to find the best combination.
Is there an add-on I can use to do this?

Am new to the forum and have looked into previous posts but could not find an answer to the following (might not have looked long enough!)

Anyway, greatly appreciate your help.

G.

Duplicate post will likely be locked or merged, so be aware of that.

1) R2 is certainly one (probably most common) way to measure "goodness of fit" though it certainly isn't the only way. I can't readily think of a "simpler" way to determine best fit among different equations. Only different ways.

2) I can't think of any add-on that would do this, especially without stating some very specific requirements. I would expect that any code/spreadsheet set-up for this sort of thing is going to be pretty specific to the type of data being collected/regressed and other things specific to the problem at hand. I know I wouldn't trust some random add-on off the internet to do this for me until I had thoroughly tested it on my data with my requirements.

Many thanks for this

Will take good note on duplicating posts.

Re 2) am looking at something to automate the production of Anova tables not the selection of the best set of independent variables. Suppose am going to have to write a macro (not my specialty) which runs different combinations and output a table for each run. Any better idea?