Talk:Herfindahl–Hirschman index

Hefindahl-Index or Herfindahl-Hirschmann Index (HHI)? - is there any difference? which term is used more often?

/krolik

There is no difference, and the former term, just plain Herfindahl Index, is used more often. -M

CFA study material uses decimals
Decimals are also used in Bruno Solnik & Dennis McLeavey's book "International Investments". see p. 265 Cheers, FSR

Normalized vs. Non-normalized.
I don't know what standard practice in economics is. In math/physics expressing the s\sub{i}'s in the interval [0,1] would yield an in index in that interval i.e. it would already be normalized. If you express the s_i's in percent, as noted above, monopoly = 10,000. Wouldn't that be the unnormalized HI? Then dividing by 10000 would yield the normalized HI?

Then again, I'm just a failed physicist. If normalization goes wrong, I just renormalize :) BTW I did google the net to no avail. I emailed the one hit I was allowed to see for free :)

-r

The non-normalized version is by far the more widely used, at least traditionally. -M

So, the question remains. Since the non-normalized version (o to 10,000) is the most-commonly-used, why on Earth is Wikipedia defaulting to the normalized version? If nothing else, it's just confusing. On top of that, it requires all of the research on HHI, and the antitrust-related work from the Department of Justice, to be normalized before it can be posted onto Wikipedia, causing excess work in order to keep Wikipedia consistent with itself. Can we just change it all to non-normalized, and include the option to normalization as a note? Ben (talk) 21:02, 5 May 2012 (UTC)

- The practice I have seen is to use whole numbers. When people use decimals the result is a decimal and it leads some to believe the result is somehow a percentage. The article has a mistake in converting the result of the calculation into a percentage. The index does not have an intuitive meaning, the units are squared percentages. — Preceding unsigned comment added by 75.149.136.114 (talk) 02:19, 12 June 2013 (UTC)

squared ² / L_2 norm
Why is the L2 norm (squaring each s_i, ²) considered sufficient "stretching" of the importance of concentrated market share? For example you would get even more weight attributed to having a few big firms by cubing ³ the data and less weight attributed to having a few big firms by putting the data to the one-and-a-half power. Why is 2 the right number?

Perhaps there is a proof showing that squaring the data has some nice property. Is this the case?

Also is there any way to test the goodness of this index? E.g., are here correlations between decreasing HHI and decreasing price over time in various markets? Is there a way to test if the index is inappropriate because certain market players fill qualitative niches or for another reason "should" be distributed as e.g. 10-10-80?

Crasshopper 16:43, 18 May 2006 (UTC)

There really isn't a good way to test this for niches, as you describe. The "should" that you are talking about is too conceptual to be born out in the data. -M
 * If I have to speculate, there is no "right" power. Different markets will have a different minimum number of firms that guarantees perfect-competition-like behavior. It would probably make sense to come up with a general HHI metric, which gives the power as a parameter. Then different markets could have different values for the parameter derived from some price elasticity standard, or simply experts' consensus.

Anyway, American regulators have decided to allow different HHIs for different markets and that seems to be sufficient for practical purposes so far. --Cryout 18:50, 10 August 2007 (UTC)

Origin of the name
(I am usually posting on wikipedia NL, but this remark seemed more appropriate with an American audience.) This article is missing a reference to the scientists Herfindahl and Hirschman (Hirschmann ?), either in the form of an internal wikipedia link or bibliographically near the bottom.--158.169.131.14 08:18, 12 September 2007 (UTC) (Lieven Smits)

Tresholds
where is come form those tresholds? (1000, 1800, ...) why 1000? —Preceding unsigned comment added by 80.191.12.33 (talk) 04:54, 20 August 2008 (UTC)

And shouldn't those treshold attain to the adjusted/normalized index? Herfindahl index ranges from 1/n to 1, so if n<10 then no index could be considered as revealing no concentration.

Math Errors in the Examples?
Sorry, but shouldn't the math formulas in the examples-as-stated actually be:

Case 1: Herfindahl index = 6 * 152 + 10 * 12 = 1360

Case 2: Herfindahl index = 1 * 802 + 5 * 22 + 10 * 12 = 6430

Jimad (talk) 19:48, 22 March 2009 (UTC)

No, percents are decimals, but the .1 should be .01 in both cases. I'm gonna fix it. elvo86 (talk) 20:17, 21 December 2010 (UTC)

Herfindahl index of Fractionalization - disambiguation page needed?
There is a variation on the Herfindahl index that is commonly used in political economy analysis of ethnic/linguistic/religious fractionalization in societies (in fact, the Herfindahl ethnic linguistic fractionalization index, or ELF, is considered the most important measure of ethnic fractionalization in the field today).

The computation is similar, substituting ethnic groups for firms, and (attempting) to include all ethnic groups (instead of the 50 largest, as with the firm calculation). To turn this into an index, the sum is subtracted from 1.

My question: does this warrant a separate wikipedia page (and a disambiguation page), since the subject matter is different, or a simply a new section on this page, since it is essentially the same formula?

Zjfstout (talk) 00:04, 10 February 2009 (UTC)Zjfstout

Threshold
Why only use the 50 largest numbers in the definition? This only relates to a practical implementation of the index, as opposed to its definition (because using the 50 largest is simpler and this does not make a big difference to the final number). Lachambre (talk) 14:34, 17 April 2009 (UTC)

I've made a change in the Decomposition paragraph that specifies the need of having all the market divided by the N firms. Otherwise the correct formula would be $$H =\frac1N\left(2\sum_{i=1}^Ns_i-1\right)+N V$$ instead of just $$H =\frac1N+N V$$. I suggest to not restrict the definition to 50 firms but write in the header that in practice only the first 50 firms are examined. There isn't such a restriction in the definition in Wikipedia in other languages as well. Cheater no1 (talk) 22:05, 9 November 2011 (UTC)