General Topics
Markets
Technical Topics
Brokerage Firms
Community Lounge
Site Support

# How much trial and error is involved in fitting to a curve?

Discussion in 'Trading' started by 76132, Jul 21, 2012.

1. ### 76132

I'm new to math, Matlab, stats, etc...

Anyways, I'm trying to fit some data points to a curve, and the equation Matlab spits out is quite inaccurate even though Rsquared is 0.97. Tweaking the degrees of the polynomial doesn't help and from looking at the points, a degree higher than 2 isn't representative of the data. The equation I have is so inaccurate that I don't trust it at all.

So, now I'm stuck and don't know what the next step is, except trial and error, i.e. adding and subtracting terms from the equation until I get a regression that I trust more.

Is this the right way to go about it? What are some obvious next steps?

I've thought about breaking the data points into different groups and finding a regression for each group, then linking the group of equations together. Is this wrong?

2. ### logic_man

Maybe try to recreate the model in a different program to see what the equation is there, for comparison. If it's the same, maybe you're just not understanding why the equation is what it is.

3. ### 76132

Well I have a pretty good understanding of the relationships between the variables by just watching the tape. And that's why I determined the equation Matlab gave me is inaccurate, because it doesn't agree with what I've observed. I can't think of any missing variables although a missing variable is a possibility.

I guess I'll try to learn R and see what that will give me. Thanks.

4. ### Random.Capital

What do you mean by "inaccurate"? With an R2 that high, it is extremely accurate, by defnition.

Are you using "accurate" and "I trust it" as synonyms?

5. ### DeeDeeTwo

Why not just use Excel...
Which has had full stats capabilities since Excel 2000...
And you can easily automate, say, Excel 2007...
With any .NET language.