Methods for Classifying Stocks by Return Volatility
Summary
The document considers ways to divide stocks into low, medium, and high volatility groups when annualized volatility estimates are available. It presents percentile cutoffs and thresholds based on the distribution’s standard deviation, then outlines alternatives such as equal-sized groups, choosing boundaries to balance differences between groups, nearest-neighbor clustering, and minimizing within-group variance.
The central guidance is that the classification should serve its intended use. Choices depend on constraints such as sensitivity to outliers and whether groups need similar membership counts. The discussion gives no empirical comparison, formal selection procedure, or evidence that one method performs best; it emphasizes that fixed cutoffs are arbitrary without a defined objective. The classification should be selected for that objective and applied consistently. The brief exchange does not specify a particular stock universe, volatility estimation window, or downstream trading application, so it cannot prescribe universal thresholds.
Key ideas
- Percentile thresholds can divide a stock universe into volatility groups.
- Standard-deviation cutoffs are another possible classification rule.
- Clustering methods can form groups by similarity or by minimizing within-group variance.
- Group sizes, outlier sensitivity, and the intended use affect which method is appropriate.
- Without an objective, chosen boundaries are arbitrary and no single partition is universally correct.
Tags
Full text
# How to classify stocks by their volatility? # How to classify stocks by their volatility? I would like to hear other possible ways of classifying Stocks by the Volatility of their returns. Assuming that I want to characterize each stock as Low, Medium or High Volatility Stock and assuming that I know the Annualized Volatility for each of the stocks in my sample, what ways are there to do such classification? I can think of two: - Below, say, 30th percentile (of the Annualized Volatilities) -> Low Volatility; Between 30th and 70th Percentile -> Medium Volatility; Top 30th percentile -> High Volatility - (-2)*Std.Dev (of the distribution of the Annualized Volatilities) -> Low Volatility; Between (+-2)*Std.Dev -> Medium Volatility; (+2)*Std.Dev -> High Volatility Feel free to point out papers where I can find my answer. ## Answer by Bram (score 1, accepted) https://quant.stackexchange.com/a/14473 There are a lot of ways of doing this and what a good way of doing this will be driven by your needs as well. Criteria such as whether the method needs to be (in)sensitive to outliers and whether or not your groups need to be of the same size will influence this. One way to do this would be sorting the volatilities and group them: - in groups of equal size - such that the mean differences between the two groups are equally large - as a nearest neighbor algorithm with 3 groups dictates - such that the sum of variance with groups is minimized (I think that's almost the same as nearest neighbors) - ... As long as you pick something and apply it consistently, it might not even matter that much which you pick. ## Answer by Shahar (score 0) https://quant.stackexchange.com/a/14519 It really depends what you're trying to achieve! What is the ultimate goal? What are your constraints? Which stocks are you looking at? Without the answers to the above, any partition is just arbitrary: why choose 30th and 70 percentile, vs 10th and 90th? Why choose (-2)*Std.Dev and (+2)*Std.Dev vs just -1 and +1? The selected (and perhaps only correct) way to classify volatilities is the one that satisfies (or at least optimizes) your ultimate goal.
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.