R^2 vs. significance of the the variables

[wrangjangler]

Hello community

I am currently working with paneldata to see if there is correlation between sustainability and performancce in the energy and materials sector of the S&P500.
I ran the regression twice, one with the logarithm of MarketValue (=MarketValue.WINS.LOG) and one without (=MarketValue.WINS).
The outcome is different and I really wonder if I should go for the log MarketValue or not?
The regression with Log MArketValue has a 0.1 -0.2 higher R^2, but the other variables are not really significant anymore.
The "normal" regression has more significant variables but the R^2 is not performing as well, as the one in the log-MarketValue-regression.

I don't know which model is the right one and thought, may you could help me out. I attached the 2 regressions, so you can take a closer look for yourself.

Thanks in advance and have a nice weekend.

 

[Miner]

Are you using R^2 or R^2(adjusted) for your comparisons? Do not rely on R^2 alone. R^2 will increase as you add terms to the model, even when they add no value. Use R^2 (adjusted). It adjusts the score for that effect.

 

[wrangjangler]

I see, thank you very much
I think I will leave MarketValue out completely, because it is StockPrice x Number of Stocks and isn't that a no go to use a variable that includes the variable you want to explain?

 

[Miner]

It depends on how you are using it. You would not want to use a variable that is part of an equation used to calculate another variable because it would result in multicollinearity. For example, you have one variable that is a linear calculation using another variable. You would not want to include the other variable. However, if you want to investigate a possible interaction between stock price and number of stocks, you can do that.

 

[hlsmith]

If you were debating between two versions of a variable (variable vs ln variable) you would base its selection on content knowledge and distribution of residuals. Not significance or R^2.