# Small question about normalization

Lets assume I want to normalize some stock data ( prices or log prices) to compare for different types of correlation for example.

And here is the question how should I normalize:

a) by subtracting common mean of prices of interest and divide it by common sd?

b) or should I use individual means and sd's in this procedure?

• Why do you want to normalize before calculating correlations? – Jean-Paul Apr 15 '15 at 20:03
• I don't want to. I was just comparing before and after etc. – Bobby Digital Apr 15 '15 at 20:09
• I was reading some article where method a is mentioned, also two answers here favor method a... but why? – Bobby Digital Apr 15 '15 at 20:12

## 3 Answers

The point of normalization is to put everything on the same level (i dont mean price level.) Prices are usually nonstationary, so CLT doesnt apply, while returns arent. So @siegel 's answer is correct in saying use a) with return data.

• Yes returns are great, we all know it, but let's not subset just to it alone. Also method b for sure will put everything on the same scale with mean=0 and sd=1 for all observations, which may not be the case for method a at all. Or maybe you mean something different by same level? However both methods will produce equal correlation... – Bobby Digital Apr 15 '15 at 19:16
• So far I don't see the point of method A ... Also what is the advantage of A over B in case of returns?? – Bobby Digital Apr 15 '15 at 19:43

I would prefer choice a), however, I'd work with returns, not prices.

You normalize for example by having a mean of 0 and a standard deviation of 1 for the data Use in R the scale function.