Any thoughts on whether it makes more sense to use the standard deviation or rather the absolute deviation to measure the variability of fundamental factors (e.g. sales)?
If you want to give more weight on larger deviations, use standard deviation.
In which component do you plan to use this measure?
In ranking, only the order matters. So, if SD or AD always return identical order, it shouldn't matter.
1 Like
You'll want to use relative standard deviation. If you simply use standard deviation or absolute deviation and you're looking for companies with steady sales, you'll be favoring companies with sales close to zero. Relative standard deviation takes the standard deviation and divides by the mean. SalesRSD%Ann and SalesRSD%TTM are good factors to use here: the first measures the relative standard deviation of annual sales over the last ten years, and the second measures the relative standard deviation of TTM sales over the last ten semiannual periods. The same functions are available for most fundamentals (e.g. DbtSTRSD%Ann).
Thank you all for your help. A quick follow-up question for @yuvaltaylor. Jvj pointed out that standard deviation gives more weight on large deviations. This means that ratios that use the standard deviation divided by the mean are much more prone to extreme outliers than metrics using the absolute deviation would be. Doesn't this result in an overestimation of the variability of fundamentals?
I see that p123 uses abs(mean) in RSD. A web search shows RSD defined with either mean or abs(mean). Is there a reason to prefer abs(mean)?
Not necessarily. It depends on what you're looking for. Let's say you're examining a dozen quarterly sales reports and one of them is ten times higher than the others. Is that something to ignore (it's clearly an outlier) or something to be extremely concerned about? It's really your decision here.
1 Like
Yes. Standard deviation is always positive, and dividing by a negative would mess up ranking no end.
1 Like