# will this methodology end up giving me a nonsense regression equation.

I'm wondering if this is a valid methodology to find the best regression equation for a given data set.

User provides a rang of estimated value for some set of variables. Th algorithm uses the given range to randomly generate a large number of regression equations like the one bellow.
Price = actualValue1* estimatedValue1 + actualValue2* estimatedValue2;

We solve them all for price and pick the best one as regression equation. Will this methodology give us a valid regression equation or will it just find some random set of estimated values That happens to give a price that is close to the actual price?

PS. I've taken stats but I'm not a statistician So I'm in a little over my head here. If I need to clarify something please let me know.

• Can you give a precise mathematical description of your question, specifying exactly what is being done? The trouble with plain English text descriptions is that they leave quite a bit of room for ambiguity and guesswork, making questions harder to answer. – Kirill Sep 15 '17 at 14:12

## 2 Answers

That depends on (i) how many random samples you draw, and (ii) how good a regression you want to have. The point of finding the best fit regression is that you want the best model out of all possible models; you only have finitely many, and so in general you will not get the best possible model, just the best out of the random samples you've generated. Obviously, if you draw more samples (assuming they are truly random), then the best among them will be at least as good as before, and maybe better, but in general not "best".

No, this is not a valid way to proceed. There is no guarantee of finding the best regression coefficients. Use multiple linear regression: https://en.wikipedia.org/wiki/Linear_regression