Sector/Market charts and outliers

[KnowThePast]

A question for anyone that has previous calculated average PE, PB, etc for the entire sector, or entire market.

What is/are the accepted ways to deal with outliers?

To give an extreme example, a company with a market cap of $100m reports a $1k profit, putting it on a PE of 100,000. One such occurrence is enough to completely throw off the numbers for the whole group.

So, what to do?
If exclude, at what point?
If include, what kind of adjustment to make?

While I am at it - the average PE's normally quoted for markets - are these weight adjusted by market cap, or are these truly an average?

 

[DeepState]

A question for anyone that has previous calculated average PE, PB, etc for the entire sector, or entire market.

What is/are the accepted ways to deal with outliers?

To give an extreme example, a company with a market cap of $100m reports a $1k profit, putting it on a PE of 100,000. One such occurrence is enough to completely throw off the numbers for the whole group.

So, what to do?
If exclude, at what point?
If include, what kind of adjustment to make?

While I am at it - the average PE's normally quoted for markets - are these weight adjusted by market cap, or are these truly an average?

Typically eliminate negative P/E from the universe.

Calculate not with weighted market cap, but with the 'one-stock' principal. This basically treats the universe as if it were a single stock. It is not just a simple average unless badly calculated. One-stock methods, which are the true representation for a sector/market P/E etc., usually wipes out the outliers which tend to be very small and contribute little relative earnings/book-value etc to an overall market. Unless you are talking about a tech/IP-style boom where earnings are negative and book value is negligible (even then, you can use revenue or gross profit or EBITDA), this works without any adjustment or censorship to the universe.

Other methods include windsorisation to rein in the extremes. Alternatively truncate or eliminate to/outside of discretionary bounds which may be nominal or ordinal.

 

[KnowThePast]

Typically eliminate negative P/E from the universe.

Calculate not with weighted market cap, but with the 'one-stock' principal. This basically treats the universe as if it were a single stock. It is not just a simple average unless badly calculated. One-stock methods, which are the true representation for a sector/market P/E etc., usually wipes out the outliers which tend to be very small and contribute little relative earnings/book-value etc to an overall market. Unless you are talking about a tech/IP-style boom where earnings are negative and book value is negligible (even then, you can use revenue or gross profit or EBITDA), this works without any adjustment or censorship to the universe.

Other methods include windsorisation to rein in the extremes. Alternatively truncate or eliminate to/outside of discretionary bounds which may be nominal or ordinal.

Thanks RY, exactly what I was looking for.

Changed my calcs to work on one-stock principal, eliminated negatives, looks a lot more reasonable now.

Interesting to see that this method is, in effect, weighted by market cap. So depending on what you are looking for, the data could hold some interesting misrepresentations. Different filters for the universe are needed in order to get a "truerer" number.

Attached is a nice chart my software now generates, thanks to you. PE, PB and PS over the last 10 years on my universe. Looking at it, and knowing some of the pecularities of those periods and my universe, I think there's very limited use to an all stock average.

Thanks heaps for your help again, RY.

 

[skc]

Interesting to see that this method is, in effect, weighted by market cap.

I am not sure if that is the case?

Company 1, market cap $100m, NPAT $10m, PE = 10
Company 2, market cap $500m, NPAT $25m, PE = 20

Cap weighted PE = 18.3333 [(10x$100+20x$500)/$600].
One stock method PE = 17.14 ($600m/$35m).

Attached is a nice chart my software now generates, thanks to you. PE, PB and PS over the last 10 years on my universe.

Impressive

 

[KnowThePast]

I am not sure if that is the case?

Company 1, market cap $100m, NPAT $10m, PE = 10
Company 2, market cap $500m, NPAT $25m, PE = 20

Cap weighted PE = 18.3333 [(10x$100+20x$500)/$600].
One stock method PE = 17.14 ($600m/$35m).



Impressive

Thanks skc.

And yes, not strictly speaking market cap weighted, but skewed towards larger cap stocks. A simple average would be (10+20)/2=15. That's the figure I initially wanted - the average valuation of a stock in the universe, but outliers proved it difficult. The solution solves the problem, but doesn't answer my original question

But, by filtering universe on market size, industry, etc. I can drill into it to get a better picture now.

I think the best use for this is when comparing a few companies in the same line of business.

 

[DeepState]

View attachment 60221

Amazing data management KTP. I hope to be able to produce something in the vicinity of what you are doing before too long. I really look forward to comparing notes with you (and others, too).

 

[KnowThePast]

Amazing data management KTP. I hope to be able to produce something in the vicinity of what you are doing before too long. I really look forward to comparing notes with you (and others, too).

Thanks RY,

Can't wait to see what you've been up to.