Analysis of Domain Names Registered Across Multiple Existing TLDs and Implications for New gTLDs
Posted by Dooot Domain News, Feb 02 2009
DNS News - >>

Source: www.circleid.com

Summary

•  Hypothesis – Trademark holders are not, in general, registering their trademarks as domain names across the existing top-level domain (TLD) namespace.

•  Methodology – Examine the domain names registered in the popular generic top-level domains (\"gTLDs\" such as .com, .net and .info) while utilizing other publicly available information such as the USPTO database of trademarks, the English dictionary, DNS entries, UDRP records and whois records, to determine if the hypothesis is true.

•  Conclusions – 1) The vast majority of trademark holders are not registering their trademarks in all the current generic TLDs, let alone all the TLDs. Therefore, we do not expect them, in general, to register their trademarks in new gTLDs and 2) due to the expected costs to run a registry and the expected low number of defensive domain name registrations, there is no economic incentive for an applicant to obtain a TLD for the sole purpose of making money from defensive trademark registrations.

Introduction

If the hypothesis is true, then trademark holders, in general, do not appear to be \"forced\" to register their marks as domain names, otherwise you\'d see more marks registered across the current TLD space.

Costs that exceed benefits arise when domain name registrants register their marks for the sole purpose of preventing another party from registering that mark as a domain name. If trademark holders are not registering their trademarks in existing TLDs, especially the open TLDs where registering trademarks is easy and inexpensive, then they probably would choose not to register them in new TLDs. At least they do not appear to be forced (to protect their trademark, or for any other reason) to be registering their marks, otherwise there would be many more marks registered across the TLDs (since costs and ease of registration is low for the set of seven TLDs studied).

Names registered in common across gTLDs

As of January 19 2009, by analyzing the zone files for a number of TLDs, we found there were 194,325 second level names registered across all of these popular seven TLDs: COM, NET, ORG, INFO, BIZ, US, and MOBI (\"CNOIBUM\"). Each of these TLDs has more than one million domain names. The smallest of which is .mobi, which has about one million names, therefore, about 20% of its zone file is registered across all of the other six TLDs. The number of common domain names registered across various TLDs decreases as more top level zone files are examined. We do not have access to all 300+ TLD zone files (such as .ca, .tv, .fr etc). If we did, we would be able to tell which names are registered in all 300+ top level zones. We can only guess it is less than 194,325.

If all of these 194,325 names were trademarks that would mean the trademark community incurs $1.5M per year per TLD in costs for registrations, if each name in each TLD cost $8 and there were no offsetting benefits to any of these registrants. But the cost number is difficult to calculate, because a) it is certain that not all of these registered names are trademarks, b) not all of them are registered to the same registrant (and one word mark, such as \"delta\" could have more than one trademark holder), c) the registrants enjoy offsetting benefits such as traffic, which wouldn\'t otherwise be brought to their content (website) if not for the registration of that domain name, and d) not all names cost the same for each TLD. While the task of estimating the per TLD costs is not easy, it\'s also not impossible.

Names not registered across gTLDs

We also want to get an idea of which names are NOT registered across these seven TLDs. Thus, we also looked at various misspellings of \"verizon\" (as one example), in an attempt to better understand if misspellings and typos of trademarks were registered across these TLDs. We picked Verizon as one example to examine because they are known to vigorously defend their marks in the domain namespace.

We examined 263 strings that we believe may be considered close typo variations of the string \"verizon\" or more distant variations, or contain the exact string \"verizon\" (154 contain \"verizon\" exactly and 109 do not) and which are registered in .com. We then determined if this list was registered in any of the other six TLDs (note though Verizon.mobi, Veriz0n.mobi, and Ver1zon.mobi appear to be registered/reserved, they do not appear in the .mobi zone file, nor do they resolve.)

Since there is no definition of what a \"misspelling\" is, we do not know all the misspellings of \"verizon\", so we couldn\'t check all of them, nor are we sure that all the strings on our list are \"misspellings\" (for example \"horizon\" is on the list). But we did find that (not counting \"version\", \"horizon\", and \"verisign\") only three of what may be misspellings of \"Verizon\" (\"veriz0n\", \"ver1zon\" and \"verison\") which are registered in .com are also registered across five or more of the other six gTLDs. We also found eight of the strings that contain \"verizon\" (exactly) are registered in .com and are also registered across five or more of the other six gTLDs. The majority of them were solely registered in .com and not in any other TLD that we examined.

As another example, we looked at \"amazon\". To determine its list, we scanned the whois and noticed that every domain name that had \"amazon.com\" as well as \"8102\" or \"81266\" (being Amazon\'s PO box numbers) in the whois output also had the same name servers in common. So we made a list of every .com name that has at least one of those name servers. We believe this is a nearly complete list of .com names where Amazon.com, Inc. is the registrant. We found 7,078 .com names (including 19 IDN names), 4,573 of which have the exact string \"amazon\" within the domain (for example \"AMAZONKID.COM\" and \"EAMAZONIA.COM\"—generally not typos), and 2,505 of which do not (for example \"AMAZKN.COM\", \"AKAZONA.COM\", \"CICKRIVER.COM\" and \"EYEMDB.COM\"—more likely typos than not). Then we determined which of these names were registered across the other six TLDs. The results are as follows:

The vast majority (80%) of Amazon\'s names are registered solely in .com.

These are but two examples, but we can infer by them, and by the number of domain names registered in each TLD (there are many times fewer names registered in say .mobi and .biz compared to .com and .net) that misspelled/typos of trademarks are NOT generally registered across these seven TLDs, let alone across all of the more than 300 existing TLDs.

English dictionary words registered across gTLDs.

In an attempt to further characterize the set of 194,325 names, we looked at an English language dictionary. We compared the words in the dictionary to this list of domains which are in common across the set of TLDs.

We used the Princeton Dictionary (WordNet 2.1) which has names listed in four categories: nouns, verbs, adjectives, and adverbs. We removed the following characters from the words: \"_\", \"-\", \"\'\" (a quote), \".\" This resulted in 145,384 distinct words, though some words occur in multiple groups.

This results in 154,007 non-distinct English words in the dictionary database.

After analyzing the set of 194,325 \"in-common\" domains, we found that there are 173,175 domains in the set of 194,325 that are not words in the dictionary, and 21,150 that are words. So 89% are not single dictionary English words—they could be multiple-word strings, or non-English strings, or made up words as stings, or not words at all (such as numbers).

Do the registrations have the same registrant across these gTLDs?

For the 194,325 names, we then focused our attention on the whois for each name in each TLD to try to understand if the domains registered across this set of seven TLDs were registered by the same registrant or by different registrants.

It is difficult to parse out the registrant information for each name from each registrar\'s whois information for each TLD. Therefore, we concentrated on the email addresses found in the whois, as they are easier to parse due to the fact that they have an \"@\" and other characteristics which easily distinguish them from other text in the whois output. We looked at every email address for every contact but we cannot tell which email is related to which contact. For example, we don\'t know that email address \'A\' for domain example.com is the email address for the technical contact or the registrant contact or any other contact—but we do know the frequency with which it appeared in the record. Additionally, we do not know the underlying whois for names which are on ID Protect.

We assumed that names registered across all TLDs and have a common email address could either be registered to the same registrant or could, for example, have the same technical contact (say they were all registered at the same ISP domain name reseller) but have a different registrant contact. Given this, we looked at a two types of information, the email addresses in the whois and the name server names, to try to determine if the domain name had the same registrant in each TLD or a different registrant. For example, if the name has the same name servers and the same one email address in the whois across all TLDs, we figure it is very likely that the same registrant is the registrant for all the names across all the TLDs.

The effort to identify unique registrants resulted in the following:

We feel the number of names with the same registrant across the board is bounded by the following: it\'s no more than 20,715 and no less than 1,096. Due to the fact that many registrars use the same name server names across millions of registrants, we believe the 20,715 is high because different registrants could be using the same name servers. Of course a name could have the same registrant and also have different name servers across the TLDs. On the other hand if one email address is in common across the TLDs, we can probably assume the true number of names which have the same registrant contact to be higher than 4,698. One also has to consider ID Protect type services which do not disclose the underlying whois information but a) those services do not hide the name server names, and b) .US does not allow such services. So, based on all this, we estimate about 6,000 names (3%) are registered to the same registrant across these seven TLDs.

Then we turned our attention to answer the question \"what if all the names registered to the same registrant across these gTLD were trademarks—what would the cost be?\" We believe there is a correlation between price and volume. We compared .info and .biz. Both TLDs were introduced at the same timeframe (year 2000). The .biz per year fee is in the $6 range and .info price is in the $2 range (depends on various promotions)—one is three times more than the other. The .biz registry has two million names in the registry, and .info has five million—one is 2.5 times more than the other. Nearly the same ratios: price to price and volume to volume. So for higher fees, we would assume the volume of registrations would decrease, and likewise for lower fees, the volume would likely increase, and by proportionally the same amount. Therefore, if ALL 6,000 of these names were trademarks then that would mean the entire trademark community was spending about $48,000 per year per TLD (assuming an $8 registration fee—about the average in-volume retail price across these TLDs), or about $336,000 per year worldwide on all these gTLDs. And, we would expect approximately this same dollar amount per TLD if the price were raised or lowered.

Of course, not all of the 6,000 same-registrant names are trademarks. Some utility (which offset the costs) flows to the registrant for having the name registered—even for a less familiar TLD like .mobi. Also, to figure the true cost, even if they were all TMs, we have to consider a) the registrants who were convinced (possibly by a zealous domain name salesperson) to register their mark across all the TLDs (to \"protect\" it) even though they may not have really needed to, and remove them from the 6,000 and b) other names not all registered to the same registrant, which may be distinctive one-holder trademarks, and increase the 6,000.

More on registrants across the gTLDs

As detailed above, our assessment reveals about 6,000 names are registered to the same registrant across all the seven gTLDs that we examined. But we also wanted to know how many names had six out of seven of the same registrants? Or four out of seven? How many names, again registered across all seven gTLDs, had none of the registrants that are the same? To determine this we again looked at the email addresses in the whois information and devised a score (see following table) with the lower score meaning the name has many different registrants across the gTLDs, and a higher score meaning the name has very few or one registrant across the gTLDs.

In the table above, each letter represents a unique email address found in the whois output for the string.TLD domain. For example, in the whois output for DomainString1.TLD1, two unique email addresses were found, email address A and email address B. In the above table, DomainString3 has the most registrants because it has the lowest score, and DomainString2 has the fewest registrants because it has the highest score.

Using this method we calculated the registrant score for each of the 194,325 names registered across all the seven TLDs. The results are plotted below.

This shows that not only are most of the names registered across all the gTLDs NOT registered to the same registrant, but also that most of them are registered to separate registrants. In other words, each of the names is likely to have seven different registrants. This makes sense, because these names, by definition, are names which are registered in all the seven gTLDs. Thus, they are probably popular names, and therefore it is likely that each would be registered to a different registrant for each TLD. The names are spread out to different registrants each of whom gets value from a unique address.

How many of the 195K names registered across all of these seven gTLDs are trademarks?

We do not have access to a list of all the trademarks world-wide, nor to all the registered trademarks worldwide, but we do have access to the database of US registered trademarks. This is a good approximation of the set of registered word marks worldwide because many, if not most, of the word marks in the USPTO database are probably also word marks in other country\'s trademark offices. So, to answer the question of how many of the domain names are trademarks, we worked with the USPTO word mark database to get the list of all live word marks (including design mark and word mark combinations).

The appropriate query to the USPTO database resulted in a list of 1,457,676 live word marks. The list has duplicate word marks because more than one trademark holder may have a trademark on \"M\" for example. Of the 1.45 million non-dead word marks, we found 1.08 million are unique strings (\"unique\" meaning that they are not on the list more than once, for example: for two identical strings in two different trademark classes—that string would be counted just once); and, for example, among the list of 1.45 million, \"GUARDIAN\" is on that list 141 times, and \"MACS\" is on it eight times, while they are both represented once among the list of 1.08 million \"unique\" word marks. We removed the non-domain characters from the list of word marks (for example, all the spaces were removed).

After comparing the list of over 1 million USPTO word marks, we found 53,327 of them were also registered across all seven gTLDs. For example \"ANDREW\" is on the US trademark list AND is registered across CNOIBUM, as is \"MICROSOFT\". Note that 1) not all of the 53,327 registered domain names found to be USPTO word marks have a single trademark holder (compare \"ANDREW\" to \"MICROSOFT\") and 2) not all of them are distinctive (again compare \"ANDREW\" to \"MICROSOFT). We found 21,366 which are not in the English dictionary and have just one trademark holder.

Other Costs?

One possible other source of costs to trademark holders does not flow to registries/registrars because it\'s not a name registration cost. It\'s for Universal Dispute Resolution Procedures (\"UDRPs\" or \"DRPs\"). The UDRP is the mechanism ICANN has contractually put in place with all registrars so that trademark holders may relatively inexpensively (compared to the court system, for example) dispute a domain name registration.

To get an idea of these UDRP costs we can look at .me as a proxy to better understand how much a new TLD may cost the entire trademark community in terms of UDRPs. We picked .me as an example because:

a)  it has a valuable meaning (\"me\" is the nominative singular pronoun in English, and is the 73rd most frequently used word in English), probably comparable to new gTLDs,

b)  though it is the country-code TLD for Montenegro, it has been positioned as a generic TLD,

c)  it has been heavily marketed by a number of registrars including the largest,

d)  the TLD was introduced recently, in early 2008, and

e)  its mid-sized at 200,000 domain names currently registered (our estimate)

Since .me was launched until now (about a year), there have been 11 UDRP cases:

All of the above were filed with the World Intellectual Property Organization (WIPO), and one was filed with National Arbitration Forum (NAF), which is: ping.me Karsten Manufacturing Corp v. pingify networks, Inc (Claim Denied)

The fees for each proceeding depends on a number of factors, such as the number of panelists chosen by the complainant (one or three, though it is very rare to use three panelists). For example, to use NAF it costs $1,300 for up to two domains and one panelist and WIPO is $1,500 for up to five domains and one panelist. If all the cases above have one panelist, which they likely do, the total UDRP fee for all of them would be $14,800. This amount does not include attorney\'s fees, which we estimate at another $2,000 each (for in-house preparation) or $22,000 total. Therefore the entire trademark community spent about $36,000 for UDRP disputes during the launch of the new generically positioned TLD \".me\" in the past year. One should also note this is a one-time cost, and assumes all complainants win (one above has not so far). The yearly retail registration fee for all of the above names is less than $200 (13 x $15.00).

Conclusion

We estimate:

• There are 194,325 names which are registered in common across these seven gTLDs: .com .net .org .info .biz .us and .mobi (CNOIBUM)
• About 3% of the names in common across CNOIBUM are registered to the same registrant.
  • That is about 6,000 names per TLD
  • If ALL these names were trademarks, it is costing the worldwide set of trademark holders about $48,000 total per year per TLD on average (at $8/domain) to have their TM registered as domain names in gTLDs.
• There are 91,332 names registered across CNOIBUM (or about 47%) that have about six or seven different registrants
  • 61,575 or about 31% are registered to about five different registrants
• At least two trademark holders, but we expect trademark holders in general, do not appear to be registering misspellings of their trademarks outside of .com across all the other six gTLDs we studied. This is because those misspellings do not appear in the other TLDs zone files, not because they do appear but are not registered to the trademark holder. We believe this is due to the fact that these six non-com TLDs do not have the type-in traffic that .com has. We believe the typo names are not being registered by \"cyber squatters\" because there is less type-in traffic in the other TLDs compared to .com. Judging by their absence, these typos are not worth \"squatting\" so they are not worth \"defending\", so they just do not get registered.
• There are 53,327 names (or about 27% of the names registered in common across these seven gTLDs) which are USPTO registered word marks.
  • 28,324 (or 53% of the word marks that are also domain names registered across all seven gTLDs) have more than one trademark holder, such as \"Andrew\". 25,003 have only one trademark holder.
  • 12,118 (or 23% of the word marks) are also English dictionary words.
  • 21,366 (or 40% of the word marks) are not in the English dictionary AND have only one trademark holder. These may be relatively distinctive word marks, or marks which comprise a combination of two or more descriptive words.
• Of 1.08 million unique USPTO word marks (\"unique\" because they were de-duped from a set of 1,457,676 word marks), only 53,327 (or about 5%) are registered across CNOIBUM
• Only 1.4% (21,366/1,457,676) of the live US trademarks have one trademark holder, are not English words, and are registered across CNOIBUM

While a small minority of USPTO trademark holders has registered their USPTO word marks across the seven gTLDs studied, the vast majority of trademark holders are not registering their trademarks in all the current gTLDs, let alone all the TLDs. Therefore, we do not expect them, in general, to register their trademarks in new gTLDs.

We\'ve looked at seven, generically positioned, relatively large, global TLDs. We believe most new TLDs will not be as general purpose as these seven TLDs, but will be more specific, such as .table. We believe that the set of trademarks which are registered across these seven TLDs, will drop off significantly for usage-specific TLDs, like .table. We do not believe that the relatively few trademark holders who do register names across these seven TLDs will continue to register names in a TLD like .table. Those trademark holders will be less interested in .table, but other users will be more interested in .table because of the very fact that the gTLD is specific. For example, a site selling billiards may want to call itself billiard.table instead of billiardtables.com, especially if billiardtables.com was unavailable. We believe that as the number of new TLDs introduced increases, the number of these niche or specific TLDs will increase faster, so the percent of niche TLDs to the total number of TLDs will continuously increase over time.

Based on

a)  the number of common registrants across the gTLDs studied (1,096 on the low end and 20,715 on the high end, with an estimate of about 6,000),

b)  the number of domains matching a word mark with one trademark holder and which are not dictionary words (21,366), and

c)  other factors discussed above,

we estimate that about 7,000 names could be expected as defensive registrations in a truly generic TLD comparable to .info or .biz or .us, with fewer defensive names expected in more specific TLDs such as .table. We suspect this number of names would be confirmed by looking at recent sunrise registrations for such generically positioned TLDs as .tel, and .me, but we do not have access to that data.

Comparing the costs of operating a TLD registry (up-front fees to ICANN, annual fees to ICANN, and technical operating costs—which is hundreds of thousands of dollars per TLD), with the revenues that would flow from defensive trademark registrations (about 7,000 or fewer names) leads us to the conclusion there is no economic incentive for an applicant to obtain a TLD for the sole purpose of making money from defensive trademark registrations.

Written by Paul Stahura, Chief Strategy Officer, Demand Media. Visit the blog maintained by Paul Stahura here.