Abstract: Digital convergence is quickly leading to various telecommunications and media sourcing triple play services of voice, television and internet. This article describes the extent to which bundling strategies have been recently deployed by access providers, and their impact on competition

in Western Europe. Generally speaking, we observe that the "business sorting" effect of bundling is larger than that of "business stealing" for triple-play in Western Europe. The sorting effect does not however compensate for the extra competition arising from the full convergence of access provision.

Key words: non-linear pricing, bundling, triple-play, broadband, television, telephony.

**********

The advent of media and telecommunications convergence has created new market spaces, with access providers marketing multi-play services combining internet broadband, television and telephony.

In a matter of a few years, bundling has become the rule. In the US, the first bundle offered in the market place was a double-play, composed of digital television and telephony. It was sourced by Cox, a major cable company, with some success indeed. Cox claimed that bundling reduced digital video bouquet churn rate by more than 30 percent. In Europe, "all in one" triple-play bundles including internet recently emerged. These were rapidly introduced by DSL-based local loop unbundlers. France in particular witnessed the introduction of very aggressive offers by Free, discounting the triple play offer by as much as 50% versus the telecom incumbent, France Telecom.

Bundling strategies in media and telecommunications continue to evolve, moving to the quadruple play by incorporating mobile on top of home wireline products. Cable quadruple play is being offered by Rogers in Canada deploying Wifi on top of its mobile and cable offering. In the US, Sprint set up a joint-venture with a consortium of US cable companies . NTL-Telewest in the UK bought Virgin to develop a new quadruple play company called Virgin Media.

The logic of bundling strategies is already well understood from a business perspective (STREMERSCH & TELLIS, 2002; HANSON & MARTIN, 1990). Another stream of analyses has focused on how bundling affects competition. Economists such as SCHMALENSEE (1984), McAFEE et al. (1989), SALINGER (1995), or AMSTRONG (1996) demonstrate that bundling may lead to effective price differentiation. In some cases, bundling can also act as a mechanism of market foreclosure, creating concern of anticompetitive behaviors (the reader is referred to STOLE, 2005, for an accurate and comprehensive survey).

The recent literature (CRAMPES & HOLLINGER, 2005; REISINGER, 2004) brings more "nuance" by moving from the limit case of monopoly to the less restrictive observed structure of oligopoly. In such instances, bundling can be pro-competitive and improve industry welfare. The reason is intuitively as follows: bundling is profitable as far as it makes customers more homogenous, therefore extracting rents ("the sorting effect"- see CRAMPES, 2006). However, in oligopoly, the homogenization of customers reduces the opportunity to differentiate by covering only some niches or market segments. This homogenization intensifies price competition and leads to erosion of profits ("the business stealing" effect). The net effect is not that always clear-cut.

AMSTRONG & VICKERS (2006) generalized this intuition under more general forms of consumers demand, and integrating the possibility by customers to prefer one-stop shopping. Further, the authors compared both regimes of linear/separate product price bundling and non-linear/product bundling pricing. This comparison allows to sort out the merits of both strategies and demonstrates that in general, industry profits and welfare can both increase under bundling. This happens with higher demand elasticity, higher product differentiation, higher shopping costs and higher brand preferences correlation. Again, the intuition for those results is easy to grasp.

Consider for instance high demand elasticity: under linear pricing, this would lead to lower mark-up over marginal cost. With non-linear pricing, one can segment the market between general shoppers and one-stop shoppers. This will prevent the risk of a full Bertrand price- competition.

The fast development of media and telecom access convergence seems to be a good case in point to analyze bundling strategies at the light of those new theoretical predictions. The third section develops a snapshot of bundling practices in Western Europe, while we also provide some basic facts of bundling customer preferences for one-stop shopping in the previous section. We then develop various multivariate econometric analyses of a reduced form model of bundling propensity and pricing for the recent years. In practice, the methodology is sequential. First, we test a model of the decision to offer triple play bundle; given the decision to offer those bundles, we then estimate a model of triple play pricing. The pricing equation includes the (inverse of the) Mills ratio of a provider choosing bundling as a regressor to avoid bias in estimating the joint decision of pricing and commercial bundling.

The results are summarized in the fifth section. In a nutshell, those are that the general Amstrong and Vickrey theory of "business sorting" strategies is apparent in triple play. This "sorting effect" however does not compensate fully for extra competition from converged services, leading to price deflation and thus better consumer welfare for triple play access services. Furthermore, we observe bundling tactics that are consistent with theory, for each type of provider. For instance:

* New fixed line service providers like DSL-based local loop un-bundlers would rather position their offer at very low price and limited services, possibly with superior product attributes (more bandwidth for broadband, etc). This helps to bypass the strong brand effects of telecom incumbents, with legacy telephony customers.

* On the other hand, cable companies have also the legacy benefit of offering video, plus the potential to offer full telecom services on their networks. Accordingly, they have two paths to bundle (via video, or via internet) providing them with the flexibility to cater to a larger set of core correlated brand preferences--this added flexibility means that poaching telephony may not require full product discount.

* Finally, incumbent telecom companies have the largest revenue legacy in telecom services, and will resist price decline. They tend to be price followers. Their reactions are typically to offer "tit for tat" IPTV services against cable, or to expand the product range via quadruple play including mobile to further differentiate against triple play DSL-based local loop unbundlers.

* One-stop shopping and customer brand preferences: triple-play evidence for Western Europe

A condition as to why access providers in media and telecom will resort to bundling is the demand heterogeneity arising from one-stop shopping and correlated brand preferences. Anecdotal evidence suggests that one-stop shopping and brand preferences prevail. For instance, in an open policy debate in Australia pushed by the telecom regulator about the pro-competitive aspects of bundles, the AOL answer to this debate reports the following set of facts (AOL, 2003) (1):

"[...] The value of offering bundled telecommunications services is well recognized by the telecommunications industry. Bundling reduces customer incentives to use more than one service provider and reduces customer churn [...]. In a speech given by Dr Switkowski at the Credit Suisse First Boston conference in Hong Kong in March 2001, when discussing customers who acquire bundles from Telstra, Dr Switkowski said: These customers are twice as likely to be high value and three to four times as likely to remain loyal to Telstra. [...]. In a paper presented by Adrian Chamberlain at the Australian Broadcasting Authority Conference on Radio Television and the New Media in May 2001, Mr Chamberlain said, CWO market data indicates that those customers who take a bundled set of CWO services are far less likely to churn their individual services, such as local and long distance telephony, to other service providers [...]." (source: AOL communication to competitive Australian authority, Feb. 2003).

Market research evidence is provided among others in IDC (2005) for the US consumers, and by McKinsey authors, ISERN & PEDORMO (2005) for Western Europe. Those pieces of research show that bundling is a preferred service option for many consumers, and that a prevalent driver for it includes one-stop shopping convenience.

We hereafter also report three insights from an updated McKinsey consumer research conducted in the course of the summer of 2006 (ASCARI et al., 2006) and which confirms the early results publicly available by other McKinsey authors (ISERN & PEDORMO, 2005). While this report is not necessary for the core of this article, -our article focuses on providers strategies-, this specific survey covers both the time frame (2005 and 2006) and a set of European countries ,which are precisely in the scope of this article (Germany, UK, Spain, France, Italy, The Netherlands and Poland) (2). From the reading of this survey, one finds:

There is a major preference for multi-play bundles among online consumers

With multiple choices allowed, 56% of the online users reported an interest for bundling quad play; 43% for fixed-mobile convergence, 53% for triple play telephony, internet and video (the scope of this article). Those figures are slightly higher than those reported in IDC (2006) and Isern and Pordomo (2005), demonstrating an increasing interest for bundling on the consumer side. Also one notices that the narrower the bundle, the lower the bundling preference- this means that not only bundling preferences are increasing through the years, but in favor of more comprehensive bundles.

Service components are as critical as product features for bundling

Within the sample of triple play preferences, 79% report broadband product features as key elements of the bundle choice. Services such as unified billing (71%), or unique customer sales/service point (70%) are also key drivers of bundling preferences. For the above, we thus anticipate broadband offering to be a key anchor of bundling strategies- this is confirmed by the empirical data in the next section.

There are switching costs for bundling

Compared to 56% interest in quad play, only 42% would want to switch from a current provider which is not offering bundles. In the same way, compared to 53% interested in triple play, 39% only would churn from a provider not offering bundling services. This means that a "lock-in" factor prevails for early providers such as telecom company (for telephony) or cable (for television services).

Taken in aggregate, those three insights are illustrative of the AMSTRONG & VICKREY (2006) conditions for profitable nonlinear and bundling strategies. One stop shopping effect prevails and is incomplete (56% versus 46% in quad play preferences,--a 20% advantage for bundling versus non-bundling). Correlated brand preferences are slightly more relevant than one-stop shopping: despite customer interest in triple play bundling, (53-39)/53= 27% of users will not churn away from their current mono-product provider(s), etc.

In this context, bundling is likely to prevail and create a "business sorting" effect. We now turn to a snapshot of bundling strategies observed in the market place.

* Triple-play bundling strategies in Western Europe

We concentrate our analysis for triple play within the home, that is, outside of the mobile, as the current offers of quadruple play for the time frame of our analysis (2003-2006) were still relatively scarce.

The sample uses various sources, from company web sites to industry reports and industry associations. We also checked the information via systematically calling the various providers centers for information. The sample is, we believe, the most exhaustive and comprehensive collection of the bundling and pricing strategies for up to 5 to 7 largest access providers for multi-play services in 10 European counties. The sample is fully balanced, that is, we only retained companies for which all data points were available for at least 3 years on a quarterly basis, 2004-2006. Based on this criteria, the sample includes 55 companies. The countries covered are UK, Germany, France, Spain, Italy, and smaller countries like Benelux and Scandinavia. The sample covers 82% of the annual access revenues for Western Europe in 2006.

The average company broadband access market share is 22%, for a Herfindhal index of concentration of 19%, meaning that the market resembles an oligopoly of about 5 players of equal size. This level of competition is much higher than traditionally in the fixed telephony market or in the pay-TV video market (ITU, 2006), so that we expect the convergence market to be under price pressure.

There is also considerable heterogeneity in the types of companies covered. 27% of the sample are cable companies, 16% stand for the incumbent telecom companies, and the rest are DSL-based ISPs. This heterogeneity is also visible for the countries covered, especially with regards to platform coverage (cable can have close to 100% reach in Benelux, versus less than 30% in France), to unbundling level (much more favorable in France than in Germany as a result of both regulation, and network topology), etc. Our statistical analysis will control for those differences--see infra.

Table 1 provides summary statistics on the full sample. For illustration, the data points shown in Table 1 were for 2nd Quarter of 2006. Seven key elements are worth mentioning:

Bundling offers have become mainstream

In general, bundling is used by a large majority of providers. Moreover, 72% of the access providers in our sample offer at least one form of double play bundle. 64% also offer a triple play sourcing of telephony, internet and video. If double play remains prevalent, it is not necessarily for all combinations of triple play, especially because of the television product. By mid 2006, only 20% of providers offered the complete combination of double play, with all providers however at least offering the double play of Internet broadband and telephony (100% for Internet, 82% for voice). Television has a more limited offering in bundles of double play: 45% include television in their double play bundles.

Access triple play have passed the bar of 50 Euros per month

The average price for buying the full components of triple-play access was 56 Euro per month, by second quarter of 2006. This price was slightly lower, at 53 Euro for triple play bundles. As such, the price difference is not large. While not reported in the table, the average price has decreased by close to 17% per year in our sample from 2003 to 2006.

Mixed bundling has become more prevalent than pure bundling

55% of providers are mixed bundlers, that is, they offer separate products (mono, or dual play) on top of triple play bundling. For those mixed bundlers (where narrower range of product offering can be compared), the average price discount for any other combination of double play to compose triple play is relatively significant, in the average of 19%. This price discount is much higher than what we see for the full triple play bundles (6-8%) versus triple play. This implies that mixed bundling is rather used as a marketing tactic to upsell to triple play.

Triple play comes with a valuable offer

By mid 2006, a triple play offer featured a 3 Mbps/second downstream Internet access, a basic bouquet of 40 television channels, and flat national fixed telephony rate. Broadband speed has increased by 50% a year while the television bouquet size is larger than analog cable, and flat fee has become the rule for domestic calls.

The type of offerings is however wide among providers and skewed by a few outliers: regarding television, the average offering of basic channels is 55 (reflecting outliers like Free offering a hundred channels). The average Mbps rate downstream advertised for high-speed internet is more like 7 Mbps reflecting a few companies in our sample--almost exclusively ISPS marketing already much wider-band services, such as FastWeb in Italy. The fact that ISPs offer a broader range of features for their products than incumbent telecom is consistent with the idea that ISPs must offer superior horizontal differentiation to bypass the incumbent "brand" effect.

Installation and devices are becoming free

40% of providers still asked for a service activation charge by mid 2006 (for an average of about 50 Euros). Those activation charges are already no longer existing for ISPs trying to build their market share. In the same way, self-installation has become standard practice (70% offer it), yet 75% still offer the alternative of a home technician provided the consumer clears an average fee of 50 Euros. Regarding modems, 80% are free, while TV decoders are only free in 40% of the case. The buying price for the other 60% of cases was between 70 to 120 euros.

Market regulation and competition matter

To test this, we looked at simple uni-variate correlations between the price level of triple play (offered in bundle or not) and a few variables. First, there is significantly stronger price competition in France and in countries with more favorable regulatory rules; France is known to have the most favorable full unbundling regime; the correlation between being a French provider with average triple play price is -58% and statistically significant at 5% risk level. Second, we did look at market structure and market dynamics. Telecommunication providers charge a higher price than other telephony attackers, being either cable or DSL service providers; (correlation with price is +58%, also statistically significant at 5%). There is also a positive and significant correlation between average triple play price and the country Herfindhal index for internet access (correlation of 37%).

Bundling is more differentiated by features than through price

Besides the above correlations with market structure, we also looked at correlations with bundling tactics. Those correlations are only partial correlation, so they can be spurious (see after), but they suggest the following. First, bundling is not found to be statistically correlated with lower price level (the correlation between 3-play price and 3-play bundling is even positive at +12%, while being negative, but relatively low and not significant at -14% for mixed bundling). Second, cable and DSL providers are slightly more keen to engage in bundling, yet the correlation is weak and not significant (+4%). Third, bundling is linked with wider-band and with larger offering of TV channels (statistically significant at +42%, but barely significant at +22% for TV offering however). Finally, we also observe a positive, albeit marginally significant (at 10%) correlation between free installation and bundling offering (+27%).

We propose hereafter a more complete multivariate analysis of bundling strategies.

* Multivariate evidence of triple play bundling strategies in European telecommunications

In this section, we aggregate all of the discussions above, and develop a reduced form model of bundling for triple play services in Western Europe.

We test a sequential model. First, we test a model of the decision to offer triple play bundle; given the decision to offer those bundles, we estimate a model of triple play pricing. As per the econometric literature, the first model was estimated via maximum likelihood techniques, using a multinomial probit model. The second model is a traditional least square model, including the (inverse of the) Mills ratio of a provider choosing to bundling as a regressor. The Mills ratio correction is necessary to avoid bias in the least square estimation, as pricing decision is jointly determined with the one to offer commercial bundling (HECKMAN, 1979; HECKMAN et al., 1996).

One simple condition for the second equation to be identified is that one explanatory variable is exclusive to the first model. Univariate correlation shows that triple play bundle price is not linked to the breadth of double play bundling offerings. On the contrary, the breadth of double play packages is linked to the propensity of triple play bundling. Accordingly, the extent of mixed bundling will be the exclusive variable in the first model.

The reduced form model of both equations has been estimated by pooling time series (12 observations on a quarterly basis, from 2003 to 2006) with panel data on providers strategies. The fixed effects include time dummies, T, and country effects, C,--the time dummies include unobserved effects, while the country effect tackles differences in national regulation. The Herfindhal (H) index per country captures the state of market structure and concentration (SCHMALENSEE, 1984).

Regarding provider features, we correct for the typology of the company (cable, telecommunication or ISP), as well as for the different features of the triple play offering. We also include a measure of the company size; the measure is the market share in consumer broadband access.

Triple play bundling

We report first the results on the decision to offer triple play bundles. The equation is of the form:

I(bundling) i,j,t = f (T, Hit, Ci, Pj, P[i,j,t])+u [1]

Where I = 1, 0 pending if triple play is offered by the provider or not, f(.) has been approximated by a linear function; T= 1..12 are time dummies, Hit is the Herfindhal value for the i-th country at time t; C is a country dummy (i= 1,..10), P [.,.,.] is a vector of each j-th (j= 1,55) provider features at time t, while Pj is a service provider fixed effect, capturing all unobserved features. Finally, as mentioned above, 3-play pricing is not included in the vector P[.,.,.] in this first-step estimation. The term u is the residual term, assumed to be normally distributed.

We expect: a) mixed bundling to be associated with triple play, b) Herfindhal to be negatively associated with bundling, and c) larger propensity to bundle for ISPs as attackers in the marketplace.

The results of this model as reported by equation [1] are displayed in Table 2. Only statistically significant coefficients are reported for readability (our threshold value used for statistical significance is a P-value not higher than 10%). An F-test demonstrates that both time and providers fixed-effects are significant on aggregate, but they also are not reported for clarity of table. The goodness of fit is "honest",--with an implied R-square of 36% for the panel dimension of the data outside of the fixed effects (with fixed effects included, the implied R-square improves to 72%). This fit is usually what one expects for panel data and the F-test for absence of spurious regression implies a minimal risk of spurious results of 1,1%.

Based on model being sufficiently well specified, we can comment on the regression results. In general, our hypotheses are confirmed by the data.

Let us turn first to the variables which did not come out as statistically relevant. The decision to offer 3 play bundle is not correlated with product features- this makes sense as product features are part of the horizontal differentiation strategy and more an element of pricing tactic for the triple play access provider. Second, size of company has not come as significant, contrary to bivariate correlations. This means that other correlates with size (e.g., the dummy control variable, ISP) picks the size effect better than market share.

With regard to significant variables, mixed bundling increases the marginal propensity to bundle by 26%, which is smaller than the average proportion of 55% of bundlers using mixed bundling. At the margin, thus, there is a decreasing tendency to offer mixed strategies. Being a DSL-based ISP increases the propensity to offer (mixed) 3-play bundles, again in line with our hypothesis that bundling is a tactic to differentiate offerings and build market share.

Finally, the structure of competition indeed matters. The "cable municipality" dummy identifies the cable in countries (mostly Northern Europe) where cable covers more than 90% of population and providers must carry video analog package of 30 channels or so. In those countries, satellite and DTT penetration are low, while those countries have also less favorable local loop unbundling rules, limiting the economics of ISPs. As revealed by the data, cable companies in those countries have witnessed a lower adoption of bundles. (Consider indeed countries like Germany and Scandinavia which have been late in upgrading their municipality infrastructure, contrary to DSL-based markets like in France where competition has been promoted for favorable local loop unbundling terms, inducing the emergence of full triple play players investing in local loop such as Free or N9uf Telecom). Finally, the Herfindhal index is negatively related to the propensity to bundle--again, this is consistent with the idea that bundling is more used to differentiate and create "business sorting" effects under stronger rather than lower competition (SCHMALENSEE, 1984).

Pricing triple play bundles

The second analysis concerns the determinants of 3-play bundling pricing. The equation is estimated by least square techniques on the subsample (36 companies) offering triple play bundling services. The equation is of the form:

Price [i,j',t] = f (T, Hit, Ci, Pj, P [i,j',t], M[i,j',t])+v [2]

Where Price [i,j',t] is the triple play bundling price expressed in Euro per month, f(.) has been approximated by a linear function; T= 1..12 are time dummies, Hit is the Herfindhal value for the i-th country at time t; C is a country dummy ( i= 1,..10), P [.,.,.] is a vector of each j-th (j'= 1,36) provider features at time t, while Pj is a service provider fixed effect, capturing all unobserved features. As mentioned above, we include the Mills ratio, M[i,j',t] of the propensity to bundling, computed from the first regression [1], so as to minimize estimation bias in the 3-play bundle pricing equation. The term v is the error term assumed to be normally distributed.

Based on the theory of AMSTRONG & VICKREY (2006), we hypothesize that: a) the Mills ratio will be linked to price (i.e., triple play pricing must be affected by bundling strategies); b) product features will be allowed to avoid price competition as a measure of consumer value differentiation; and c) ISPs to adopt low price signaling in order to bypass incumbents bad effects.

Table 3 provides the regression results. As for Table 2, fixed time effects are globally significant, (but not reported for clarity). We however include the country effects that are significant for discussion. Non significant variables are omitted in the table for simplicity.

The goodness of fit is relatively good, with a R-square of 62%. The risk of spurious results is very low, with F-test resulting in less than 0,1% (Is this a decimal figure in which case it should be 0.1%?)of risk of a spurious statistical relationship. The least square regression coefficients can be interpreted as the marginal Euro effect per month on the price level of triple play bundle access. The regression results are broadly in line with those expectations.

First, we see that country effects for France and Spain come as significantly negative, i.e. price competition is visible in those countries beyond bundling and providers features alone. France and Spain are quite interesting countries as they are converging to the same market dynamics for the triple play. On one hand, those countries are witnessing a consolidation of cable (Ono-Auna in Spain/ Altice in France); and likewise for pay-TV satellite (Digital + and Canal respectively). Further, a few aggressive and successful DSL-based ISPs are pushing for local loop unbundling against the telecommunication incumbent (e.g., Jazztel in Spain against Telefonica and Free in France against France Telecom). However, France has had more favorable local loop unbundling conditions provided by the regulator, furthermore, at the precise time of narrow-band upgrade, lowering switching costs for customers to leave incumbents, and thus favoring aggressive pricing by players such as Free.

Second, the Mills ratio is significant,--and interestingly, positive. Table 1 already had shown that average 3-play pricing was not that different from 3-play bundle pricing. Here, the effect shows that triple play bundle allows to increase marginally 3-play pricing by roughly 6 Euros, or roughly 10% versus average triple play price. If we combine both equations (1)--(2), the full effect of mixed bundling increases price by about 8 Euros. Otherwise stated, the multivariate evidence is that bundling allows to keep price higher at the margin than under harsh competition. This is the classical "sorting effect" emphasized by CRAMPES (2006). Obviously, this does not mean that ISPs do not price lower than incumbents. Being an ISP significantly affects both the propensity to bundle as well as the price of the 3-play. Thus, the full effect of being an ISP (computed by combining both equations) is to price down at the margin by 22 Euros. A triple play "bundling ISP" such as Jazztel and others will thus price 22-8= 14 Euros less at the margin, or a discount of 25% versus average 3-play prices.

Third, the regression results show that product features are positively linked to price. This confirms successful horizontal differentiation and extra price premium according to AMSTRONG & VICKREY (2006). There is still an interesting observation arising by looking at the marginal coefficient estimates impact on product features on price. On one hand, providers price roughly 15 cents of a Euro for offering one extra channel. This price compares well with the average price of an extra video channel in basic packages in Europe of 25-30 cents per month in Europe (as an example, the difference in our sample, between triple play pricing and double play outside video is roughly 13 Euros per month, for 50 channels, or roughly, 25 cents for best tier channels). In contrast, let us consider the price of extra bandwidth. The ITU (2006) reports in a sample of countries worldwide that the price for 100kbits/second is about 50 cents of a US dollar, with Japan on the low side (Yahoo!BB price is about 7 cents). In our sample, the price is about 52/7 = 70 cents (95) of a Euro (Dollar) per 100kbits/second which is on the high side. Now from our regression estimate from equation (2), the extra price the access user pays at the margin, is only 3 cents of a US dollar for 100kbits/second extra. Under this perspective, the extra bandwidth offered appears to be relativey cheap. This is seen in the marketing offers of DSL-based ISPs which all are emphasizing "speed" as a buying factor for broadband choice. As DSL-based ISPs are generally the ones offering mega-band, the key price differentiation of an attacker is mostly noticeable, not in absolute price, but in price per Mbps. For heavy users, this is the right metric of relevance, and we expect the DSL-based ISPs to capture heavier users than normal telecoms from this analysis.

We close on a final note. Our collected sample shows a clear hierarchy in price between types of providers, with roughly, after controlling for country, time, and market effects, a 35-40 Euros per month for ISPs, a 45-50 Euros for cable and about 60-65 Euros for telecom players. Those differences from our regression demonstrate a narrower price corridor among the type of service providers than what the gross numbers show (for instance 29.9 Euros per month for Free versus 70 Euros for France Telecom); the difference being due to market structure (say, more attackers prevail in France than in Belgium, where Belgium is a cable municipality country). This ranking of price by type of provider is what we should expect to have from the bundling strategy literature--but in general the "sorting effect" of a few Euros does not compensate for those price differences and the recently observed price reduction of 20% for access to triple play. In other worlds, bundling and product features elements have created powerful business sorting, but have not fully hedged against the price competition, -probably to the great benefit of the customer and the development of broadband in the European markets analyzed in the scope of this article.

* Conclusions

This article has analyzed the developments of triple play strategies in the wake of media and telecommunications access convergence. While typically bundling has been seen as a tactical tool to price discriminate and sort consumers, the convergence development in telecom services leads to a more competitive oligopoly. The balance between "business sorting" and "business stealing" must then be assessed empirically.

The article makes two contributions--on one hand, it has compiled some interesting stylized facts of triple play and bundling in Western Europe, concluding that bundling has become a major part of the marketing practices of the convergence industry. On the other hand, and based on a review of the literature on bundling, the article has demonstrated that bundling propensity and pricing should reflect the balance between "business stealing" and "business sorting" effects. The econometric evidence supports those theoretical predictions, demonstrating that ISPs are more price aggressive than the telecom incumbent, and differentiate, sometimes through heavy discounting of horizontal features of triple play, such as broadband speed.

References

AOL (2003): "Competitive effects of bundling telecommunications service", February reply to the Open Policy debate on competition for telecom in Australia.

ARMSTRONG M. (1996): "Multi-product Nonlinear Pricing", Econometrica, Vol. 64, no. 1, pp. 51-75.

ARMSTRONG M. & J. VICKERS (2006): "Competitive nonlinear pricing and bundling", MPRA working paper, Sept. 2006, University of Munich.

ASCARI et al. (2006): "Perspective of the T(e)lec-(h)om(es)", McKinsey internal telecommunication expertise, Frankfurt, April.

CRAMPES C. (2006): "Triple play time", University of Toulouse, Draft manuscript.

CRAMPES C. & A. HOLLANDER (2005): "Product specification, multi-product screening and bundling: the case of pay-TV", Information Economics and Policy, 17, 35-59.

HECKMAN J. (1979): "Sample selection bias as a specification error", Econometrica, 47 (1), 153-162.

HECKMAN J., H. ICHIMURA, J. SMITH & P. TODD (1996): "Sources of selection bias in evaluating social programs: An interpretation of conventional measures and evidence on the effectiveness of matching as a program evaluation method", Proceedings of the National Academy of Sciences USA, 93, 13416-13420.

HANSON W. & K. MARTIN (1990): "Optimal Bundle Pricing," Management Science, 32(2).

IDC (2005): "Consumers Demand a Better Bundle: A Cluster Analysis of Household Preferences for Voice, Data, and Video Bundles", Pub Time: 2005/10.

ISERN J. & J. PERDOMO (2005): "Bundling Europe's broadband", McKinsey Quartely Web exclusive, February 2005, from: www.mckinseyquarterly.com

ITU (2006): "Digital.life", ITU internet report, 2006.

McAFEE R.P., J. McMILLAN & M.D. WHINSTON (1989): "Multiproduct monopoly, commodity bundling, and correlation of values", Quarterly Journal of Economics, 114, 371-84, May.

PAPANDREA F., N. STOECKL & A. DALY (2001): "Impact of bundling of telecommunication services on consumers, industry and competition", Working Paper, University of Canberra, Communications and Media Policy Institute.

REISINGER M. (2004): "The effects of product bundling in duopoly", Discussion paper 2004:26, University of Munich.

SALINGER M.A. (1995): "A Graphical Analysis of Bundling", Journal of Business, 68 (1): 85-98. 138 No. 68, 4th Q. 2007

SCHMALENSEE R.L. (1984): "Gaussian demand and commodity bundling", Journal of Busines, 57, S211-S230.

STOLE L.A. (2003): "Price discrimination and imperfect competition", University of Chicago, The Handbook of Industrial Organization.

STREMERSCH S. & G. TELLIS (2002): "Strategic bundling of products and prices: a new synthesis for marketing", Journal of Marketing, Vol 86, pp. 55-72.

(1) Many other examples exist and point towards customer brand preferences and value of one-stop shopping (see among others, CRAMPES & HOLLANDER, 2005; PAPANDREA et al., 2001).

(2) The survey also extends bundling preferences to quadruple play. The sample, which was collected by TNS, a lead market research company worldwide, was also built to represent the socio-demographics of the Internet population (ASCARI et al., 2006). More details of the results can be requested confidentially from the authors.

Jacques BUGHIN

McKinsey & Company, Belgium, University of Brussels and University of Leuven

Pedro MENDONCA

McKinsey & Company, Lisbon, Portugal (*)

(*) The article reflects the authors' own opinion, not necessarily their institution. Any remaining error is their own.

Table 1--Triple-play services in Western Europe
Top 50 companies, snapshot for mid year 2006

                                              Metric          Average

Company type
* Cable                                      Percent
* Telecommunication                          Percent
* ISP                                        Percent
* Access broadband market share              Percent            22%

Triple play features
* Triple play pricing                       EUR/month           56.3
* Triple play bundle pricing                EUR/month           53.1
* Triple-play bundling offering              Percent
* Mixed triple play bundling offering        Percent
* Discount mixed bundling              Points of percentage     19%

Triple play products elements
* Brodband speed                               Mbps              7
* Number of TV channel                        Number             55
* PVR offering ( Yes/no)                   Percent Yes
* VOD offering (Yes/no)                    Percent Yes
* National telephony flat rate             Percent Yes

Triple play service features
* Service activation charge                Percent yes
* Activation fee if charged                    EUR
* Self-installation                        Percent yes
* Free modem                               Percent yes
* Free decoder                             Percent yes
* Free Wifi                                Percent yes

                                              Median         Proportion

Company type
* Cable                                                         27%
* Telecommunication                                             16%
* ISP                                                           57%
* Access broadband market share                18%

Triple play features
* Triple play pricing                          54.9
* Triple play bundle pricing                   52.9
* Triple-play bundling offering                                 64%
* Mixed triple play bundling offering                           55%
* Discount mixed bundling                      16%

Triple play products elements
* Brodband speed                                3
* Number of TV channel                          40
* PVR offering ( Yes/no)                                        37%
* VOD offering (Yes/no)                                         56%
* National telephony flat rate                                  82%

Triple play service features
* Service activation charge                                     40%
* Activation fee if charged                                      50
* Self-installation                                             70%
* Free modem                                                    80%
* Free decoder                                                  40%
* Free Wifi                                                     76%

Table 2--Propensity to offer 3-play bundles in Western Europe
Years 2003-2006 quarterly, maximum likelihood, n = 55 providers

Regression Statistics-(outside
fixed effects)

Likelihood correctness                 0.59
Implied R Square                       0.36
Standard Error                         0.33
Anova F                                0.01

                                Coefficient   T-statistic     P-value

Mixed bundling offering             0.26          1.76         0.08
ISP                                 0.07          1.69         0.10
Cable "municipality" * effect      -0.37         -2.69         0.01
Herfindhal                         -0.04         -1.82         0.06

Table 3--3-play bundle pricing, years 2003-2006 quarterly
Least squares with fixed effects, n = 36 providers

Regression Statistics

Multiple R                             0.79
R Square                               0.62
Standard Error                        10.50
Anova F                                   0


                                Coefficient   T-statistic     P-value

Mills ratio (3-play bundling)       6.45          1.78         0.08
France                            -36.04         -2.99         0.01
Spain                             -19.12         -1.68         0.10
ISP                               -20.53         -3.73            0
Mbits/s                             0.25          1.86         0.08
Number of TV Channels               0.15          1.74         0.10