Analysis of Factors of Demand for Domain Names in CENTR Member Registries

Transcription

1 Analysis of Factors of Demand for Domain Names in CENTR Member Registries Prepared for CENTR's 36 th General Assembly (Crete, Greece) June 5-6, 2008 By Dr. Matthew Zook, ZookNIC Inc. 1

2 Table of Contents Executive Summary... 1 Introduction... 2 Data Sources and Methods... 2 Part I: Overall Domain Name Growth Trends... 5 Part II: Factors Behind cctld Domain Registrations... 8 Part III: General Analysis of CENTR Member Registries' Growth... 9 Step 1 Price, Trend and Country Effects... 9 Step 2 Internet Use and Socio-Economic Effects Step 3 E-commerce Effects Step 3.5 Exchange Rate Effects Step 4 Registry Policy Effects (Final Model) Part IV: Price Elasticities from the General CENTR Model Part V: Summary of Findings Figures, Maps and Models Figure 1, Number of Domains Worldwide... 5 Figure 2, Aggregate Registration and Pricing Trends in 21 CENTR Registries, Figure 3, Pricing and Registration, CENTR Registries, Figure 4, Pricing and Registration, CENTR Registries, Figure 5, Relative Global Growth Rates of Selected Variables... 8 Figure 6, Example of Effects of TLD Policy Changes in TLD "A" Figure 7, Example of Effects of TLD Policy Changes in TLD "B" Figure 8, Example of Effects of TLD Policy Changes in TLD "C" Figure 9, Increase Demand for Domains from a Ten Percent Price Reduction Figure 10, Change in Revenue Resulting from a Ten Percent Price Discount Map 1, Percentage of Internet Users in Europe, February Model 1 Price, Trend and Country Effects Model 2 Internet Use and Socio-Economic Effects Model 3 E-commerce Effects Model 4 Exchange Rate and Price Interaction Model 5 Exchange Rate and.com Domains Interaction Model 6 Registry Policy Effects

3 Executive Summary This report analyzes the factors behind demand for domains offered by CENTR member registries as a function of price, economic conditions, Internet activity, exchange rates and registry level policies. The resulting general analysis (using data from 21 CENTR member registries from January 2003 to January 2008) utilizes a standard regression model with nationlevel fixed effects as well as the possibility of a linear trend across time. The models developed in this analysis are extremely robust and explain almost all of the variation within short term and long term growth. Specific results include: Price matters. Price clearly matters both for short-term demand and for long-term maintenance of domains. However, it is also evident that demand for domains is most sensitive (elastic) when price tends to be relatively high. Currently price is relatively inelastic. As the average price of domains has declined over the past five years, demand for domains is increasing tied to other factors and for most registries the big gains in terms of growth based on reasonable price reductions have already been achieved. Growth of Internet use is a key foundation for demand. One clear result from these models is that as countries increase their use of the Internet (particularly commercial utilization), demand for domain names expands alongside this growth..com domains exhibit a mild substitution effect. This dataset shows that.com domains can act as a substitute for cctld domains but the effect is rather small, i.e., each.com registration results in 0.5 percent fewer new domains per month. Policies which "open" the registry show up most strongly in the short term. Policies that expand the scope of who can register domains and what can be registered generally result in an initial surge or "land rush" as new parts of domain space is occupied. The longer term growth results, however, are mixed. More Registrars Equals More Sales. A strongly significant policy factor tied to the total number of domain registrations is the number of registrars selling a particular TLD. Findings on Major Marketing Campaigns and Low/Free Registrations are Inconclusive. While price incentives, such as low-cost or free registrations, can be shown to have a short term effect, the longer term results are inconclusive. This finding, however, does not invalidate these strategies but points to limitations within the dataset. 1

4 Introduction This report analyzes the factors behind demand for CENTR member registries as a function of price, economic conditions, Internet activity and registry level policies. The resulting general analysis, which utilizes a standard regression model with nation-level fixed effects as well as the possibility of a linear trend across time. Data Sources and Methods CENTR oversaw the collection of data from its member registries and ultimately provided ZookNIC with data from the following 21 TLDs: Table 1, Participating Registries Country Austria Canada Catalan Community Finland France Germany Ireland Italy Lithuania Norway Poland Russian Federation Slovakia Slovenia Spain Sweden Switzerland The Netherlands Turkey United Kingdom United States TLD AT CA CAT FI FR DE IE IT LT NO PL RU SK SI ES SE CH NL TR UK US 2

5 The data was assembled by the registries and included monthly figures from January 2003 to January 2008 (61 observations for each TLD). ZookNIC matched these data to a range of other variables making the final list of model parameters as outlined in Table 2. Table 2, Variables in Analysis Variable Type of Variable Source DOMAIN NAME GROWTH Total Domains Registered Ratio Registry New Domain Registrations Ratio Registry Monthly trend Ratio ZookNIC PRICE EFFECT Price Ratio Registry Average Exchange Rate Dollar to Euro Ratio Interbank Rate SOCIO-ECONOMIC EFFECTS 1 Population of Country Ratio World Bank GDP per capita Ratio World Bank Number of Internet Hosts in TLD Ratio Network Wizards Number of Internet Users in country Ratio ZookNIC E-COMMERCE EFFECTS SSL Licenses Ratio Netcraft Number of.com registrations in country Ratio ZookNIC REGISTRY POLICY EFFECTS Foreign based Registrants Allowed Binary Registry Private individuals allowed to register Binary Registry Maximum number of domains per holder Ratio Registry Second Level Registrations Allowed Binary Registry IDN Registrations Allowed Binary Registry Free one year registrations campaign Binary Registry Low cost registrations campaign Binary Registry Major (more than normal) Advertising Binary Registry Registry / registrar model Binary Registry Number of registrars Ratio Registry Not all registries provided data for the full 61 time periods and some registries provided additional data for the early months of Moreover, some observations for some registries were not included in the final analysis when registry supplied data was missing or deemed 1 The TLD.cat represents a challenge as it intended community does not correspond to national boundaries used by other data sources for collection. Therefore not all variables were obtained for this TLD. 3

6 unreliable. For example, the new domain data for certain months for some registries was simply the difference in the total amount of domains (i.e., new domains minus deleted domains). Nevertheless, the final data set was quite robust and included sufficient observations from all registries to create well specified general models. Approximately half of the price data was given in Euros while the rest was provide in a range of other national currencies. For the analysis, all prices were converted to Euros in order to increase the generalizability of the findings particularly price elasticities. 4

7 Part I: Overall Domain Name Growth Trends As of March 2008 there were approximately 164 million domains registered world wide (see Figure 1). These were divided between gtld domains (62 percent) and cctld domains (38 percent). Growth rates for both types of domains over the past two years have been phenomenal, well over 30 percent per year. Moreover, cctlds have been growing particularly fast and have increased their overall share of domain names by about three percentage points during the past 12 months. 2 There is, however, significant variation within cctlds with some exhibiting very high growth rates while others post much more modest figures. 180 Figure 1, Number of Domains Worldwide Millions of Domains cctld gtld 0 Aug 2002 Feb 2003 Aug 2003 Feb 2004 Aug 2004 Feb 2005 Aug 2005 Feb 2006 Aug 2006 Feb 2007 Aug 2007 Feb 2008 Moreover, growth rates can be extremely variable from one month to the next. All twenty one of the participating CENTR member registries exhibit a trend towards more domain registrations overtime. This mirrors the overall development exhibited in Figure 1 and suggest that rather than reaching a saturation point, the domain name market (both generally and for these specific cctlds) retains good opportunities for growth. 2 A large part of this is due to the expansion of the.cn TLD. 5

8 Figure 2 highlights this expansion and shows that the number of domains per capita in the 21 CENTR registries has more than doubled during the past five years -- increasing from 14 domains to 32.5 domains per 1,000 inhabitants. Mirroring this trend is a 65 decrease in the average price charged by these registries during the same period. Figure 2, Aggregate Registration and Pricing Trends in 21 CENTR Registries, Price (Euros) Average Registrar Price Domains per 1,000 pop Domains per 1,000 Pop Jan-03 Jul-03 Jan-04 Jul-04 Jan-05 Jul-05 Jan-06 Jul-06 Jan-07 Jul-07 Jan-08 These clear aggregate trends, however, belie the tremendous variety in pricing and domains per capita illustrated in Figures 3 and 4. While CENTR registries are moving towards more similar pricing, the overall variation in domain usage seems, if anything, to be expanding. Moreover, these figures show that price alone can account for a only a small portion of the differences between TLDs in terms of the number of domains. 6

9 Figure 3, Pricing and Registration, CENTR Registries, Average Price = 24.2 Domains per 1,000 Pop. = 16.8 Price Domains per 1,000 Pop. Figure 4, Pricing and Registration, CENTR Registries, Average Price = 12.4 Domains per 1,000 Pop. = 32.5 Price Domains per 1,000 Pop. 7

10 Part II: Factors Behind cctld Domain Registrations Understanding the strong growth rates in CENTR registries is at the heart of the research question of this report. Namely, what are the key factors driving the growth of domain name registrations? Figure 5 shows the growth rate for all domains relative to the growth of a selection of other global economic and Internet indicators. All figures are indexed to their level as of July 2002 and thus their relative change over time can be directly compared. While all indicators show increase, the non-internet variables gross national Income (GNI) and population grew at considerably lower rates than Internet related ones Secure Sockets Layer (SSL) Licenses, Internet Hosts and domain names. Clearly, growth in the Internet in general and domain names more specifically is proceeding at a much faster rate than the overall global economy. Figure 5, Relative Global Growth Rates of Selected Variables Indexed (100 = July 2002 Level) SSL Hosts Domains GNI Population Jan-01 Jul-01 Jan-02 Jul-02 Jan-03 Jul-03 Jan-04 Jul-04 Jan-05 Jul-05 Jan-06 Jul-06 Jan-07 Jul-07 Jan-08 While this insight is enlightening it does not answer the specific question of what factors are driving domain name growth. Simple economic expansion or population growth are likely aspects but are clearly not the only ones. In addition, the trends shown in Figure 2 raise important questions about the effect of pricing or registry level policy on domain name growth in the short and long term. It is to these specific questions that this report now turns. 8

11 Part III: General Analysis of CENTR Member Registries' Growth Other and earlier modeling efforts by ZookNIC shows that GDP levels, Internet users,.com registrations and price can explain 75 percent in the variation in cctld registrations worldwide as of July This basic model provides the starting point for this analysis to which a number of improvements and modifications are applied. These include: Use of monthly data from 2003 to 2008 rather than one point in time; Availability of two dependent variables Total Domains (Long Term Effects) and New domains (Short Term Effects) 3 ; A wider range of indicators of Internet activity with a specific measure of website e- commerce activity; and The inclusion of a number of registry policy variables. While these additions provide considerably more power to this analysis, it is important to note that the dataset only includes the CENTR member registries which provided data for this project (see Table 1). The high level of participation provides confidence that these models accurately capture the historical CENTR member experience but caution should be exercised in extending this to other regions or for forecasting. The following section outlines the results of the analysis of cctld registrations as a function of price, economic conditions, Internet activity and registry level policies. The resulting general analysis, which utilizes a standard regression model with nation-level fixed effects as well as the possibility of a linear trend across time, is well-behaved, and provides a number of interesting insights. Step 1 Price, Trend and Country Effects It is generally useful to follow models through their development so that one can see how the inclusion of new variables change the significance and coefficients for ones already in the model. With this in mind the analysis begins with the simple two-way relationship between price (in Euros) and short term (new domains) and long term (total domains) demand. These models include adjustments for time and country, and uses the following variables: (1) Price - The price for a new domain registration measured in Euros 4. (2) Fixed Effects - A series of binary variables indicating the registry 5. (3) Month ID - An integer variable indicating the month, allowing for a linear time trend. 3 For the most part, these variables behaved similarly and the times when they do not more weight is given to the results for total domain registrations. This is because some of the data for new registrations was bad, i.e., simply representing the difference between total registrations from one month to the next, which makes for inconsistency in the models. 4 This analysis only uses the price for new domains in its models. While some registries have different prices for new and renewed domains this policy is not consistent across registries. Moreover, since the data survey did not differentiate between new and renewal prices it is unclear whether all registries that do offer differential pricing provided this data in the survey. 5 As noted above fixed effect (dummy) variables for each registry were included in the model to capture registry/country specific effects. Given the range of registry strategies, market and socio-economic conditions it is important to control for inter-registry/country differences not captured by the independent variables. These dummy variables emerged as significant in all the models but are NOT included in the result below in order to make them more legible. 9

12 Model 1 Price, Trend and Country Effects From this model alone it appears that price does not influence demand for domains. The positive and significant coefficient in the month trend variable in both models shows that registrations follow a strong and steady rise over time and apparently there is no immediate relationship between price and new domain registrations as the price variable is not near the threshold for significance. Furthermore, the positive and significant results for price in the total domains model suggests that, in the long run, registrations increase as prices rise! Step 2 Internet Use and Socio-Economic Effects The first model while elegant in its simplicity has one particularly glaring fault. It treats all registries and their associated countries as uniform while in reality there is great variation along any number of directions, e.g., size of the country, wealth, and variations in the availability and use of the Internet. Even though the 21 registries analyzed in this report share many economic and social conditions, there is at a minimum clear differences in the overall size of their associated countries. More importantly and aptly illustrated by Map 1, there is considerable difference in the use of the Internet across CENTR member registries. 10

13 Map 1, Percentage of Internet Users in Europe, February 2008 In recognition of these important differences, the next model is therefore adjusted to include four additional variables: (4) Hosts - The number of Internet hosts operating under the TLD; (5) Users - The number of Internet users in the country; (6) Population The population of the associated country, in millions; and (7) GDP per capita GDP per capita income, logged. 11

14 Model 2 Internet Use and Socio-Economic Effects The new set of models behave much more as one would expect. As a country increases its use of the Internet (as indicated by more Internet hosts and more Internet users) the demand for domain names also increases. Interestingly, the inclusion of these measures of Internet use result in the expected relationship between price and new domains (new registrations increase as the price drops), but price and demand still reveal no significant relationship in the long-run total registration figures. Two anomalies in these models are also worth exploring. First, the negative coefficients for GDP per capita and population which mean that as a country becomes wealthier and more populous, (all other things equal), demand for domains drops. This result would seem to defy common sense unless one considers that the predictions of this model have also been adjusted to reflect Internet activity. Raw population and raw wealth do not make domains more attractive unless and until that wealth is used to set up and use the Internet. Otherwise, the variables simply represent social and economic activity taking place "offline", which in fact logically would discourage a concern with Internet domain names. 6 The other puzzle in this model is the significance of the time trend in total domain registrations but not in new registrations. Again, careful consideration of the statistical model removes the puzzle. This pattern would appear if domain registrations are growing steadily but not exponentially. Steady growth would cause the time trend to appear in total registrations over 6 The experience of Saudi Arabia and its TLD (.sa) provides a good example of this point. 12

15 the long term, more domains exist but the number of domains added to that total in the short term (i.e., new registrations in any given month) has not shown the same independent tendency to keep rising. Other things equal, it occurred at about the same rate across the data set. Step 3 E-commerce Effects The third model refines a country's Internet activity beyond the general measures of use to examine specific indicators of commercial activity on the Internet. To look at this question the model was expanded with two more country based variables: (7) SSL Licenses- The number of Secure Socket Layer 7 Licenses active in a country; and (8).com Domains The number of.com domains registered in a country. Model 3 E-commerce Effects This next pair of models confirms a relationship between e-commerce activity and demand for domains. Other things equal, the number of SSL licenses reveals both a short-term (in the form of new domains) and a long-term (in the form of total domains) positive relationship with the number of domain registrations. Notably, once the extent to which national Internet activity assumes a commercial orientation is taken into account, it is evident that the rise and fall of new 7 The Secure Socket Layer (SSL) protocol provides cryptographic protection against eavesdropping on the Internet and is therefore is a good indicator of web based commerce. 13

16 domain creation does follow the short-time fluctuations in national economic performance (as measured by GDP per capita), although the long-term registration outlook still does not. On the other hand, it appears that.com domains may act as substitute goods for CENTR member TLDs. The two forms of domain registration are negatively related to each other, suggesting that the expansion of one results in fewer of the other. This substitution, however, seems to be driven in part by the weakening US dollar (see the discussion immediately here after). Moreover the scope of this substitution effect is relatively low each.com registration reduces the number of new cctld domain registrations by only half a percentage point. Step 3.5 Exchange Rate Effects Based on comments from the presentation at the Brussels meeting, this analysis examines the effects of the weakening US. dollar on demand for CENTR TLDs. The U.S. dollar s decline may affect demand for CENTR TLDs indirectly (lessening the impact of CENTR members' price cuts) or directly (through the substitution for.com domains). In other words, as the dollar and the Euro move farther apart, there will be an indirect effect on cctld domain pricing as.com domains (priced in U.S. dollars) become relatively cheaper making cctld appear more expensive in comparison. In addition, it is also possible that there will be a direct effect as registrants decide to register a.com domain rather than a cctld one. In order to see if these potential indirect or direct impacts were taking place, versions of the last model were run with an interaction variable that allows either the price or.com domains variables to interact with the dollar-to-euro exchange rate. 8 8 These models were implemented with and without the.us TLD data included. The results were quite similar. 14

17 When looking at the interaction between the exchange rate and price (see Model 4 below) for CENTR member domains, the addition of exchange rates does not increase the explanatory power of the short term model (new domains) and it is not possible to say with confidence that the effect of price depends on the exchange rate. The exchange rate, however, does appear to have a long term effect that shows up in the analysis of total domain registrations. The negative effect of price on demand (evident in Model 3) appears to depend in part on the exchange rate, strong and negative when the two currencies approached parity, weaker to nonexistent when the U.S. dollar s value declined. Model 4 Exchange Rate and Price Interaction 15

18 Another effect of the declining U.S. dollar is an increased tendency to substitute cctld domains with.com domains which is explored in Model 5. Model 5 Exchange Rate and.com Domains Interaction The introduction of a interaction variable between the exchange rate and.com domains (see Model 5 above) in the new domain model does not increase its explanatory power and it is not possible to say with confidence that registration of new CENTR member domains versus.com domains depends on the exchange rate. The positive and significant results for the interaction term (the one showing choice between cctld and.com domains) in the total domains model means the substitution effect between the two types of domains did not appear while the dollar was strong, but has become increasingly apparent with the dollar s weakening. Thus, while exchange rate does seem to have an effect, it is difficult to cleanly isolate it from other variables, particularly price. Both exchange rate and price can cause an increase in total domain names by either expanding the overall demand for domains (indirect effect) or by capturing market share from a substitute (direct effect). This is particularly problematic at the general model using data from all 21 registries. When models were limited to data from a single TLDs, the relationship between cctld demand and the exchange rate was more direct and substantial as a strong dollar clearly helped to encourage cctld purchases. 9 As the dollar 9 This effect, however, was limited to a very small number of registries. 16

19 weakened, sales of these cctld domains did appear to slip although their overall market has not become more sensitive to price (i.e., more elastic). The best guess is that the most price sensitive portion of certain TLD markets is shifting some of its purchases to the lower cost.com domains. Due, however, to the mixed results of this analysis at the general level (particularly the effect on earlier and theoretically sound variables), the exchange rate variable is not included in the final models. Step 4 Registry Policy Effects (Final Model) The final step in the analysis is considering a range of policy variables within the control of registries. The only directly manipulable variable in the previous models is price, which obviously cuts into the bottom line when registries attract buyers through price reductions. It would be more helpful to determine factors outside of price that might influence demand. It is relatively straightforward to see clear cause and effect relations between specific registry policy actions and demand for domains. For example, the figure below shows the experience of one CENTR member TLD (details have been removed to ensure anonymity) with two policy changes. The first is when the registry allowed private individuals to register domains. This both stepped up the overall size of the registry in the short term but more importantly increased the overall rate of growth in the long term (illustrated by the different slopes on the old and new trend lines). Later on, the registry engaged in a low cost campaign which apparently only resulted in a short term gain with the longer term rate of growth remaining more or less constant. Figure 6, Example of Effects of TLD Policy Changes in TLD "A" LOW COST CAMPAIGN Number of Domains. PRIVATE REGISTRATIONS OLD TREND LINE OLD TREND LINE Time 17

20 Another example (illustrated in Figure 7) outlines similar results although in this case associated with major marketing campaigns. In this instance both marketing campaigns result in short term (an immediate bump up) and long term (a higher rate of growth) for the TLD. Figure 7, Example of Effects of TLD Policy Changes in TLD "B" Number of Domains. MAJOR MARKETING MAJOR MARKETING OLD TREND LINE OLD TREND LINE Time 18

21 Lest one get too complacent, a final (and cautionary) example is outlined in Figure 8 which illustrates a case in which low cost campaigns were not followed by any sustained increase in either the short term or long term. The registry shown below demonstrates how two separate low cost campaigns resulted in a brief step up in numbers but this increase was followed by subsequent decreases. Although it is not possible to know the reasons for this decline, the timing (especially in the first instance) strongly suggests that domains bought under the low cost campaign were not renewed at the end of their initial registration. 10 More importantly, this example also did not experience any change in the overall rate of growth. Figure 8, Example of Effects of TLD Policy Changes in TLD "C" Number of Domains. OLD TREND LINE OLD TREND LINE LOW COST CAMPAIGN LOW COST CAMPAIGN Time 10 It is also useful to remember the promotional campaigns at the end of the dot-com bubble in which registrants of.com domains were given the same domain in.org or.net. After the bust, these domains were simply not renewed resulting in a great contraction in the total size of these TLDs in late

22 Given this range of registry experience illustrated in Figures 6-8, the final set of models includes measures of a series of policies (represented by a binary variable) across time 11. These include: (9) Private Individual Registrations Allowed (1 =yes; 0=no); (10) IDN Registrations Allowed (1 =yes; 0=no); (11) Low Cost or Free Promotional Campaign (1 =active; 0=not active); (12) Major Marketing Campaign (1 =active; 0=not active); (13) Registry / registrar model (1 = registry/registrar model; 0=only registry); and (14) Registrar Count - The number of registrars available under the decentralized model Model 6 Registry Policy Effects 11 Models were created that also included policy variables on (1) whether overseas registrants were permitted, (2) whether second-level domain registrations were permitted, and (3) whether there was a restriction on the number of domains a buyer could hold. These variables did not prove significant in any cases and are therefore not included in these models. 20

23 In this final set of models, several of the policy variables work much as one might have predicted. The new domains models shows that a short term increase in demand for domains typically results when registries begin allowing individual and IDN registrations. While this short term increase translates into long-term growth in terms of IDN domains, the negative coefficient for private registrations in the total domains model is counter-intuitive. Indeed, the effect of individual registrations appears if anything to reverse. This is, however, likely simply showing that TLDs with fewer total domains were the ones that made the switch to a registrar/registry model (i.e., had more restrictive models early on). Other registries never had to make this change and thus are not reflected in this variable. Price incentives, such as low-cost or free registrations, also provide an initial push in short term sales as shown by the new domains model. The long-term effect, (as reflected in total domain registrations) however, is not clear as the variable does not appear as significant. Likewise, major marketing campaigns did not prove to be significant in either model. Part of the issue is likely tied to the quality of the data within these variable -- particularly the probable collinearity between the price and these policies (i.e., the effects are being captured by concurrent price reductions). Moreover, review of the raw data shows that these policy variables (1) represent a broad array of tactics and strategy not easily captured in statistical models and (2) registries interpreted these variables widely and provided data using different assumptions, e.g., some adjusted price during promotional periods whereas some did not. It is not possible to know whether the inconsistency reflects differences in the nature of the price incentives how comprehensive they were, for example or if it s just noise standing in the way of the analysis. While this is not unexpected (binary variables on complicated phenomenon are notoriously difficult to specify) it reduces the ability to speak authoritatively on the effects of these policies. The final two registry policies that appear as significant in the models are switching to a registryregistrar model and the number of registrars selling a TLDs' domains. As with the variable on allowing individuals to register domains, the coefficient for the first variable in the long term model is negative making for the counter-intuitive finding that switching to a registry-registrar model results in fewer total domain registration. However, it is likely again an artifact of the fact that TLDs with fewer domains were the ones that made the switch to this model while larger TLDs always operated this way during the study period. Less surprising is the positive relationship between the number of registrars and total domains registered. This clearly supports the value of making domains as widely available as possible. 21

24 Part IV: Price Elasticities from the General CENTR Model This section examines the price elasticity of demand 12 associated with the final general model developed for the CENTR member registries. It is crucial to note that this is a general price elasticity based on the overall experience represented in the database and should not be directly applied to any specific registry situation. Each registry faces a highly idiosyncratic market and these elasticities should be viewed as overall trends rather than specific operational tactics. With this caution in mind, Figure 9 shows the results of a ten percent price reduction based on the elasticities of demand from the general model. These figures are based on a hypothetical registry size of 750,000 domains although similar results (in terms of proportion) are obtained at other sizes. The analysis looks at a number of price points representing the average price of registries during the study period (some registries having identical pricing). Figure 9, Increase Demand for Domains from a Ten Percent Price Reduction 60,000 Increase Demand for Domains. 50,000 40,000 30,000 20,000 10,000-48,943 27,354 8,663 12,750 1,265 3,247 4,597 6,428 5,313 3,882 1, Price (Euros) Based on a registry of 750,000 domains 12 Price elasticity of demand (PED) measures the relationship between changes in a product's price and demand for that product. Laws of supply and demand predict that when prices rise, demand for a particular product will decrease and vice-versa. Goods that exhibit this relationship (PED > 1) are characterized as elastic. Products in which demand is relatively unresponsive to price (PED < 1) are said to be inelastic. 22

25 The results are quite striking. At the starting price point of 12 Euros, a ten percent reduction to Euros, only increases demand by 3,882 domains or little more than half of a percent. At lower starting prices the resulting demand is even more inelastic. At higher starting prices (65 or 95 Euros), demand is more responsive but the price elasticities based on the general model remain inelastic even at these levels. In short, at all these pricing levels, demand for domains is inelastic based on the general CENTR model. But increase in demand for domains is only the initial step in the question of the impact of price reductions on registry. While increased demand provides more sales, lower prices also mean that existing revenue streams will drop off. When these issues are taken into consideration, the case for price reductions become even more difficult. At a beginning price of 12 Euros a ten percent reduction to Euros decreases this hypothetical registry's revenue by 858,000 Euros (see Figure 10). Figure 10, Change in Revenue Resulting from a Ten Percent Price Discount Change in Revenue (000s of Euros). - (500) (1,000) (1,500) (2,000) (2,500) (3,000) (3,500) ,123-1,315-1,680-2,223-2,940-3, Price (Euros) Based on a registry of 750,000 domains In other words, price reductions based on the elasticities of general model consistently result in lost registry revenue. While some registries with significantly higher prices would likely profit from lowering their prices it because their markets' specific elasticities are significantly higher than the ones that emerge from the general model. 23

26 Part V: Summary of Findings The models developed in this analysis are extremely robust and explain almost all of the variation within the short term and long term growth for the 21 participating CENTR registries during the period stretching from 2003 to The specific findings are as follows. Price matters. Although it initially appeared to make little difference in demand (Model 1), this was only because it was an improperly specified model and the effect of price was suppressed by omitted variable bias. Once a fully specified model was developed, price clearly mattered both for short-term demand and for long-term maintenance of domains. However, it is also evident that demand for domains is most sensitive (elastic) when prices are relatively high. For most registries, price is relatively inelastic. The responsiveness of demand for domains is not particularly strong when they only cost a few euros. Thus, future price reductions (except for a few registries with significantly higher prices) will not increase demand substantially and risks lowering a registry's overall revenue. As the average price of domains became relatively low, demand for domains is increasing tied to other factors. This is an important point as Figure 2 shows, the average price of CENTR registry domains have dropped by more than 65 percent since For most registries the big gains in terms of growth based on price reductions have already been achieved. Growth of Internet use is a key foundation. While general economic development is clearly beyond the scope of any registry's mission, it remains an important source for increase demand for domains. One clear result from these models is that as countries increase their use of the Internet (particularly commercial utilization), demand for domain names increases alongside this growth..com domains have a mild substitution effect spurred in part by the weak U.S. dollar. The analysis of this dataset shows that.com domains can act as a substitute for CENTR member domains (particularly in the context of a weak dollar) but the substitution effect is rather small (each.com registration results in 0.5 percent fewer new domains per month). The weak dollar does seem to be providing incentives for substitution. This, however, shows up most prominently in the cases of individual TLD experience. Policies which "open" the registry show up most strongly in the short term. Policies such as allowing private individuals to register domains or allowing IDN domains, generally result in an initial surge or "land rush" as new parts of domain space is occupied. The longer term growth results, however, is mixed. In any case, CENTR registries for the most part already subscribe to "open" regimes making consideration of these policies a moot question. 24

27 More Registrars Equals More Sales. A strongly significant policy factor tied to the total number of domain registrations is the number of registrars selling a particular TLD. In short, the more places in which people can purchase a domain the more of them will be sold. Findings on Major Marketing Campaigns and Low/Free Registrations are Inconclusive. While price incentives, such as low-cost or free registrations, can be shown to have a short term effect, the longer term results are inconclusive. In similar fashion, major marketing campaigns were not shown to be significant in either the short or long term. This finding, however, does not invalidate these strategies but points to limitations within the dataset. These include (1) the collinearity between the price variable and these policies (i.e., the effects are being captured by concurrent price reductions); (2) the wide range of tactics and strategies represented by these variables; and (3) different interpretations made during the collection of this data. 25