SPECIAL ARTICLE 3G Spectrum Auctions in India: A Critical Appraisal
Alok Kumar
That auction should be a preferred route to allocate scarce resources such as spectrum is conditional upon getting the auction design right. We analyse the auction design employed in the spectrum allocation for third generation and broadband wireless access services, assessing its success on the parameters of revenue realisation, efficiency, post-auction market structure, and impact on consumers. While the auction has been successful in mobilising revenue for the government, and has created little adverse impact for the consumers by maintaining the level of competitiveness in the mobile services market, these gains seem to have been offset by the loss in efficiency as well as a higher complexity of bidding strategies. In making the Lowest Accepted Bid a preferred pricing rule, the government accorded primacy to revenue realisation over maximising allocational efficiencies. Thus we propose alternative design elements which would have served the stated objective of enhancing allocational efficiencies better.
I would like to thank Robert S Willig for helpful suggestions while finalising the paper, J S Deepak for discussions and Amarjeet Singh Dutt for his help in analysing the auction data. Views are personal.
Alok Kumar (alok.kumar1@gmail.com) is at the Lal Bahadur Shastri National Academy of Administration, Musoorie.
Economic & Political Weekly
EPW
1 Introduction
I
The Indian market for mobile telecom services is one of the fastest growing in the world. Prior to the conduct of the 3G and BWA auctions, the national subscriber base for mobile telephone services was reported to be 584 million (TRAI 2010) and had been growing at an astounding rate of over 17 million new subscribers every month. Simultaneously, the telecom policy has engendered strong competition among the incumbent telecom operators vying for market shares. This has led to consistently decreasing prices for mobile services, so that consumers here enjoy the cheapest call rates in the world.
That said, however, it is also important to note that despite such a huge subscriber base, the penetration of mobile services is only 52.7% as on 31 March 2010, implying a significant scope for further enhancement. Compounding matters is the highly uneven teledensity varying from a high of 128.20 in the urban areas to a low of 26.43 in rural areas. On the one hand, this has led to poor quality of service, particularly in dense urban agglomerations in part due to lack of adequate spectrum. On the other, fi nancial non-viability in sparsely populated rural areas has resulted in an inability to universalise the availability of service to all households. Hence the policy challenge is to ensure availability of spectrum1 while at the same time ensuring that services remain affordable for the highly price-sensitive consumers. As Jain (2001: 671) has pointed out:
From a regulatory and policy perspective, spectrum auctions ensure the efficient use of spectrum by allocating it to those entities that value it most, while also generating revenues for governments. But auctions may lead to unexpected outcomes due to unanticipated problems with their design leading to unexpected bidder behaviour such as collusion and over-bidding. The key challenge before regulatory agencies is to design auctions in such a way as to meet the objective of fostering competition while at the same time ensuring that bidders can effectively use the spectrum for their business.
This paper is organised as follows. In Section 2, we present a brief review of the spectrum allocation policy as it has evolved in India. This will set the context for the prevailing policy and its outcome in contrast to what has been done earlier. Section 3 briefly describes the auction design chosen for the 3G and BWA auctions. Section 4 evaluates the efficacy of the auction design in achieving the stated policy goals. Finally, we conclude by suggesting a few alternative design choices which could have better served the stated policy ends.
2 Spectrum Policy in India since the 1990s
India’s policy on spectrum allocation – as it has evolved over the years – has been riddled with inconsistencies. Some licence allocations have been done through a process of price discovery while others have been in a “beauty contest” mode, i e, the award of a licence to an operator through the use of discretion at an administratively determined price. Till the early 1990s, telecom services were provided by the Department of Telecommunications (DoT), a state-owned monopoly, which served as the policymaker, regulator and service provider rolled into one. When a decision was made to introduce private players for providing cellular telephony services in 1994-95, the Cellular Mobile Telecom Services (CMTS) licences for the four metros were awarded on a beauty contest mode. However, for non-metro circles, India became one of the early users of the auction as a means of awarding licences to private operators for providing basic and cellular telecom services.2 The potential operators had to bid on the licence fee for a given circle3 (the highest bidder being awarded the licence), with the spectrum coming bundled along with the licence. As Jain (2001) points out, despite relatively fewer problems with the award of cellular licences, the services were slow to take off, due to the high bids, slow clearances for frequency allocations and the lack of a suitable framework for managing the interconnect arrangements. Rash bidding on the part of operators, partly due to a faulty auction design and partly on account of the absence of a credible institutional mechanism to ensure adherence to the agreed contractual frameworks for making scheduled payments, plunged the sector into crisis. High tariffs for cellular services belied expectations of the projected growth in subscriber base and the operators were on the verge of defaulting on payments due. This situation forced the government to step in and bail out the operators by writing off the licence fees and permitting migration to a new revenue sharing mechanism through the New Telecom Policy (NTP 1999). All existing licence holders were allowed to “migrate” to a new regime that involved a one-time payment as entry fee (fixed as a percentage of the original bid) and an annual revenue share with the government (ibid). This change enabled the private sector operators to scale up operations at comparatively lower tariff levels.
Having decided to grant the third licence for cellular services to the incumbent government entity, the licence for the fourth operator was granted through an auction process in 2001. Learning from past mistakes, the government opted for a bidding process called the informed ascending bidding. Unlike the single round highest bid approach adopted during the 1994 auctions, the bidding process was concluded in three rounds. The highest pre-qualifi ed offer in the fi rst financial bid was treated as the reserve price for subsequent rounds of bidding with a pan-India licence garnering a revenue of Rs 1,658 crore to the government exchequer.
Despite the apparent success of the auctions in 2001, the government also parallely pursued the practice of assigning licences
122 (including the spectrum that came bundled with it) on receiving applications from the providers on a first come first served (FCFS) basis. Although the award of licences was conditional upon the payment of the one-time entry fee determined by the 2001 auction,4 this would prove to be a costly mistake on the part of the government. As the number of subscribers increased exponentially, so did the demand for spectrum on the part of the existing operators as well as firms who wanted to enter the potentially lucrative telecom market. The existing operators were accommodated by introducing the principle of subscriber-linked spectrum allocation,5 making India the only country in the world to follow such a criterion for spectrum allocation. The new operators were sought to be provided licences and the start-up spectrum (2 × 4.4 MHz) on a FCFS basis. However, the entry fee for the licence remained frozen at the price discovered in the 2001 auction regardless of the market having grown manifold and radically transformed over the period. Unfortunately, this inexplicable decision had the support of the regulator on the ostensible ground of maintaining a level playing field between the operator-incumbents and the new entrants.
Predictably, there was a huge rush of applications for allotment of licence for the 2G services as there was windfall to be made from the huge arbitrage between the administered price and the true market value of the spectrum that came bundled with it. Of the 545 applications pending for grant of licence, the DoT granted 122 licences in January 2008 on a FCFS basis. The exact fi gure for the loss to the exchequer in terms of revenue foregone on account of this dubious process of awarding spectrum is still not known.6 The “2G scam” has underscored the urgency for reforms in the allocation policy based on principles of transparency and nondiscrimination, and provision of a level playing field for different players and technologies.
With no consensus on the desirability of using a market determined mechanism for discovering the price of the spectrum for 2G services due to legacy contracting issues relating to such licences, the government took a bold step forward in announcing that the licences (and spectrum) for the 3G and BWA services will be awarded through an open and transparent bidding process. The auction was conducted in April-May 2010.
Having set the context, we now proceed to examine the auction design for assigning spectrum for provision of 3G and BWA services. Inter alia, the policy objectives for the 3G and BWA spectrum auction (DoT 2010) had been stated as:
(i) Obtain a market determined price of 3G spectrum through a transparent process.
(ii) Ensure efficiency of use of spectrum and avoid hoarding.
(iii) Stimulate competition in the sector.
(iv) Maximise revenue proceeds from auctions.
It may be borne in mind that the stated ends are synergistic as well as conflicting. Therefore, for the purpose of this analysis, we may infer that the order in which the ends have been mentioned reflect the priority of the government. Hence the primary criterion is efficiency, followed by concerns regarding collusive bidding and avoidance of undue market concentration in the post-auction market structure. Finally, subject to these is the maximisation of the auction revenue. In analysing the results of
march 26, 2011 vol xlvi no 13
Figure 1: Revenue Realised from Auctions ($/population/paired MHz)
US (AB) March-95
US (C) May-96
US (DEF) Jan-97
UK April-2000
Netherlands July-2000
Germany Aug-2000
Italy Oct-2000
Austria Nov-2000
Switzerland Nov-2000
US (C& F) Jan-2001
India May -10 0 2 4 6 8 10 12
(iii) The Indian auction realisation is limited to the 3G component of the auction in the 2.1 GHz band (and thereby excludes the BWA part and the 3G auction on the 800 MHz band for the CDMA operators). Source: Data from Cramton (2001) and author’s calculations.
the 3G auctions, we propose to use the above parameters in assessing the extent of success of the auctions. The actual bidding data in respect of the said auction has been taken from the DoT website (GoI 2010).
3 The Auction Design
While auctions have been a time-tested mechanism, the development of economic theory of auction is of a relatively recent origin. An earlier paper has a comprehensive review of the recent literature on auction theory, insofar as it pertains to spectrum auctions (Kumar 2010). Simultaneous ascending (SA) auctions have been the workhorse for spectrum auctions. It has worked reasonably well in simple situations for spectrum auction with a single geographic scheme (as in the 3G auctions in the United Kingdom). It allows for a simple price discovery process, and also allows substitution across lots.7 Further, its dynamic nature permits bidders to assemble desirable packages and reveals common value information, thereby enabling a better informed bidding process. But it also has limitations, particularly when applied to a more complex setting with multiple telecom service areas such as in India, with 71 licences in 22 circles under auction. The varying degrees of complementarity and substitutability in the licences under auction necessitates a combinatorial auction design that is able to provide necessary information to the bidders on how much they can win at what prices in order for them to make reasonable assessments for bidding on related packages.
Considering the above, the design contemplated for the 3G spectrum auction in India comprises a two stage process for allocating the lots amongst the bidders, termed clock-package auction in the literature. The first phase is called the clock phase and the second is called the assignment phase. In the fi rst phase, which is modelled on the SA auction, the auctioneer begins by setting the clock prices to the reserve price for each of the 22 telecom circles for which the licences are being auctioned (there is one clock for each telecom circle). The bidders are expected to
Economic & Political Weekly
EPW
indicate in a 1/0 format their demand for lots in the desired circles at the prevalent clock prices. In case the demand for lots exceeds the supply in a given circle, the clock prices are then increased relative to the previous round clock price according to a pre-specifi ed formula.8 The clock rounds continue simultaneously and independently in different circles with further bids being invited till the demand can be satisfied in every circle and the activity level9 of bidders has reached 100%.
It would be useful to introduce some notations to simplify further discussion. Suppose there are N bidders (i= 1,2,...N) for K identical lots in a given circle (N>K). No bidder can demand more than one lot in a circle. The auctioneer pre-announces a formula which determines the clock price at each clock round of the auction. Let the vector of prices governing the clock for the circle in question be denoted as P [P0, P1, … , Pt-1], such that P0 is the reserve price and the prices are monotonically non decreasing. We assume that the maximum valuation of the bidders is reached at round t-1 and all bidders exit at that price. The price increment at every stage P – Pt-1 is determined using a pre-specifi ed formula
t
which is a function of the excess of demand over supply as well as 10 Auction begins at round 1. At round t bidders are expected
Pt-1. to indicate their preference at the current clock price to bid qit =1 or to not to bid qit = 0. Total demand for objects at the end of round t, Q = ¦qit. If Q >K, we proceed to identify the provi
tt
sional winning bidders by assigning ranks to the considered bids11 for the circle in question. The considered bids are ranked in descending order of the clock round price when the bid was made.12 The top K bids are affirmed as provisional winning bidders, the rest are declared as provisional losers. If the number of provisional winning bids is less than the number of lots, all the bids are declared as provisionally winning bids. The auction ends in a round t such that Q <=K and there is no excess demand in any
t
other circle. If Q = K, the final uniform price for all winning
t
bidders is P and if not, the price is Pt-1 for all bidders (the objects
t
being awarded to Q active bidders; the remaining K-Q objects
tt
being awarded as per their ranks in the previous clock round). Such a pricing rule is referred to as the lowest accepted bid (LAB). In the second phase, the specific frequencies are assigned to the winning bidders randomly by the electronic auction software (EAS).
4 The Auction Results
We will assess the results of the 3G spectrum auctions on a number of parameters such as revenues realised, process of price discovery, competitiveness and market structure, auction design and impact on consumers.
4.1 Revenue Realisation
The standard metric for measuring the revenue realisation in a spectrum sale is the price realised per head of population per MHz of paired bandwidth that was on auction. So how do we benchmark the proceeds from the auction? Given that the reserve price for a pan-India licence was fixed at Rs 3,500 crore13 and the actual realisation for the same was Rs 16,750.58 crore, it would appear that the auction was successful in capturing the scarcity value of the resource. Figure 1 further buttresses the point by showing that the revenue realisation at the all-India level does not fare too badly in comparison with similar results in other countries, particularly when we make adjustments for differences in spending capacity of mobile subscribers.
Average revenue per user (ARPU)14 may be used as a reasonable proxy for the spending capacity of the subscribers. To put the above figures in perspective, it is instructive to note that the monthly ARPU raked in by Indian telecom companies in March 2010 was Rs 150.23 ($3)15 as compared to a figure of ¤34 ($50) prevalent in Europe at the time of the auctions (Yu 2009). Hence for the prevalent level of ARPUs, the revenue realisation from the
Figure 2: Winning Price per Paired MHz (Rs crore)
Andhra Pradesh Assam Bihar Delhi Gujarat Haryana Himachal Pradesh Jammu and Kashmir Karnataka Kerala Kolkata Madhya Pradesh Maharashtra Mumbai North-east Orissa Punjab Rajasthan Tamil Nadu Uttar Pradesh(E) Uttar Pradesh(W) West Bengal
0 100 200 300 400 500 600 700
Indian 3G auctions compares quite favourably with its European counterparts.16
However, when we look at the circle-wise distribution of revenue realised, the picture is a bit more sobering. Roughly 80% of the auction revenue is derived from the seven category A circles with the two metros of Mumbai and Delhi alone contributing over 40% of the total revenues. In comparison, the fi ve category C circles yielded a revenue share of less than 2.7%. The circle-wise winning prices per MHz of paired spectrum are shown in Figure 2.
The average price realisation per MHz of paired bandwidth was Rs 152 crore, whereas the median price was only Rs 64.3 crore. Thus the picture that emerges is that the high proceeds from the auction are largely a result of the perceived scarcity of spectrum by the existing 2G telecom mobile companies especially in the dense urban agglomerations with high penetration level of telecom services and higher scope for generating revenues. It is also clear that their expectations for growth in business in circles that are at a relatively lower level of economic development (category C circles) are not very optimistic. This will certainly impact their rollout plans and presents a signifi cant challenge for policymakers.
Another indicator of the success of the auction, insofar as the revenue realised is concerned, is the fact that none of the incumbents having a national presence was able to secure a pan-India licence. Due to the high auction price and their budgetary limitations, they were forced to bid for the circles where they have strong 2G positions. This aspect will be discussed further, when we take up the issue of entry and competitiveness in the post-auction market structure.
4.2 Avoidance of Collusive Outcomes
One of the weaknesses in multi-round auctions with the revelation of provisionally winning bidders has been the ability of the bidders to achieve collusive outcomes, since initial rounds could be used for signalling and deviations could be punished with relatively low cost retaliatory strategies in later rounds (Cramton et al 2000). There is a trade-off involved in revealing bid-related information at the end of each round in the SA auction. If we reveal too little information by withholding the prices quoted by provisionally winning bidders on given lots, we impede the process of price discovery. If we reveal too much information such as the identity of provisional bidders along with the prices quoted, there is a possibility of compromising the integrity of the bidding process through collusion among the bidders. By restricting the bids to quantity bids at specified clock prices and nondisclosure of identity of provisional winners, the auction precluded the possibility of using collusive and retaliatory strategies. By limiting the release of information to the excess demand at the end of each clock round and eliminating the need for disclosing the identities of the provisionally winning bidders, an effective tradeoff between releasing sufficient information to aid the pricediscovery process and prevent collusion/retaliation has been achieved.17 The high levels of revenue realisation from the auction provide a clear pointer to the fact that the bidders were unable to get into collusive agreements. However, a word of caution is in order. Since the auctioneer is privy to the identity of the provisionally winning bidders at the end of each round, paying attention to the institutional design for the conduct of auctions is very important while replicating such an auction in other settings.18
4.3 Efficiency
While the Indian auction design can be described as an unmitigated success if judged by the criterion of revenue realisation, the same cannot be said of the efficiency criterion. Here we use the broad notion of economic efficiency rather than the narrower one of technical effi ciency, thus implying putting the spectrum in its highest valued use. There are two steps in spectrum utilisation – allocation and assignment. Allocation implies pre-defi nition of the licence conditions such as the band plan (frequencies), the geographic area, the duration, and the restrictions on use. The assignment of the licences is then determined by auction. Both allocation and auction design decisions affect efficiency and need to be analysed separately.
First, as stated earlier, the pricing rule of the present auction is similar to that in a SA auction, i e, bidders face the pay as bid pricing incentive. Hence, bidders tend to shade their bids by reducing the quantity demanded. If indeed efficient allocation is the primary policy goal, we would need to amend the pricing rule to Vickrey Clarke Groves (VCG) pricing since it provides the right incentives to the bidders for truthful bidding, which is known to
march 26, 2011 vol xlvi no 13
lead to an effi cient allocation.19 In auctioning off multiple objects which are heterogeneous but similar, it has to be kept in mind that while agglomerations of some licences (covering similar frequency bands in adjoining geographical areas) are complements, others are near perfect substitutes (adjacent frequency bands in the same geographical area). One of the principal diffi culties in implementing Vickrey pricing with complementary goods is that the revenue is non-monotonic in bid and bidders. That is to say that there may be a coalition of bidders who may have offered the seller higher revenues than what he would receive under the VCG pricing scheme,20 which may be perceived as unfair. The reason for this is that with complements, the Vickrey prices may not be in the core. Cramton et al (2009) suggests a practical way of getting around this problem of revenue non-monotonicity under the VCG pricing by way of a small amendment. The revised rule (termed core optimal bid pricing) contemplates the price to be paid by the winning bidders as the one that is closest in Euclidean distance to the VCG price but one which lies inside the core. This will minimise the bidders’ incentive to distort bids in a Pareto sense, that is, there will be no other pricing rule that provides strictly better incentives for truthful bidding; while at the same time avoid the potential revenue non-monotonicity in VCG pricing.
Second, the LAB pricing rules prevent the bidders from expressing their true valuations particularly when the bid increments are large. As an illustration, let us take the case of Delhi circle. At the end of Round 181, there were four bidders in fray – Vodafone, Bharti, Tata and Reliance – at a clock price of Rs 3,316.93 crore. As per the auction rules, clock price for the circle for the next round (Round 182) would increase by 1% of the previous clock round price (by Rs 33.17 crore). In Round 182, only two bidders, Bharti and Reliance, bid the requisite price of Rs 3,350.09 crore, with the other two bidders opting to drop out of the race. Clearly, the valuation placed by the provisional winners (Bharti and Reliance) on getting this circle is higher than Rs 3,350.09 crore. Therefore, they should certainly get preference in the award of spectrum on grounds of effi ciency. Consider, however, the dilemma faced by the two provisional losers – Tata and Vodafone. Evidently, the provisional loser’s true valuation
(xi) for the circle is such that Rs 3,316.93 crore < xi < Rs 3,350.09 crore, for they both chose to exit at the lower end of the price range. They may have been persuaded to bid for the next round if and only if the expected positive pay-off from the final price being Rs 3,316.93 crore is greater than the expected negative payoff from the final price being Rs 3,350.09 crore. Clearly, now the bidding strategy has to be based on outguessing the rival bidders’ strategy. The larger the discrete bid increment, the greater the dilemma since bidding at a higher level amounts to risking a greater negative payoff with a definite positive probability.21 Provisional winners, on the other hand, face first price incentive. Hence, exiting before their value is reached, results in higher positive pay-off (but with the contingent risk of losing the object, since exiting is irrevocable).22 Continuing to bid increases the chances of winning but decreases the pay-off. LAB therefore results in asymmetric bidding behaviour from the provisional winners (who indulge in differential bid shading) and provisional
Economic & Political Weekly
EPW
losers (who are tempted to bid higher than their valuation) leading to inefficiencies and increased complexity of bidding strategies for bidders (Cramton 2009). Ultimately the bid was awarded to three bidders (Bharti, Reliance and Vodafone) at the price of Round 181 (Rs 3,316.93 crore). In the event, despite the similar decision to withdraw at the same price, Vodafone ended up winning the bid while Tata ended up losing out on ground of lower ranking in the previous clock round and we are none the wiser as to which of the two had a higher valuation for the circle in question.
An alternative auction rule which could improve effi ciencies with incentives for truthful bidding as well as simplifi ed bidding strategies is the highest rejected bid (HRB) with exit bid rule. This has the same rules as LAB except that there is no ranking. So in the example in question, we could have another exit bid restricted to the provisional losers of Round 182 (Tata and Vodafone) and decide the third winner from this exit bid. All winning bidders pay the price determined by the highest rejected bidder (bids being permitted between clock prices prevailing on the last and last but one rounds only). Since this is Vickrey pricing, bidding up to true value is a weakly dominant strategy for the bidders. Cramton et al (2009) demonstrate that in general, LAB is less effi cient and yields lower expected revenue than HRB for risk neutral23 profit maximising bidders. Admittedly, it may be reasonable to assume that keeping suitably low bid increments and increasing the number of rounds would lead to a convergence of results from the two rules, but given that bid increment is a percentage of clock prices in the previous round, they are going to be large in the final stages of a highly contested auction.24 In view of the above discussion, the choice of auction design appears to be inconsistent with the desired goals of effi ciency.25
Third, since the unit of bidding is individual lots, it exposes bidders to uncertainties in respect of being allocated some but not all the lots that are needed for assembling the package required by their business strategies. Bidding for synergistic combinations is based on guesswork and is thereby inherently risky. The bidder may end up making wrong decisions. This was obviated by having the two-stage auction design – with the fi rst stage merely settling the question of who wins how much in each area and at what price, before assignment of actual frequencies in the second stage. Lots on auction during the first phase are generic (perfect substitutes). By its very design, the auctioneer is assured that identical licences would garner similar payments.26 This also simplifies the bidding strategies, since the bidders focus their attention on licences rather than particular combination of frequencies to be allocated.27 Another significant advantage in this form of auction is that it determines the efficient band plan without the regulator/policymakers having to make an apriori decision on splitting spectrum. This comes in very handy when alternate technologies are vying for the same spectrum space – as in the instant auction.
Fourth, since the format of auction in the clock phase is quite similar to the SA auction, bidding for packages would necessarily involve guesses regarding the final prices of complementary licences, as withdrawal from currently won circles is not permitted in the assignment phase. Hence, the auction design does no better than SA auction in reducing the exposure of the bidders to agglomeration risks. The auction could have benefi ted by attempting to capture the values arising out of complementarity amongst licences by using a better approach in the assignment phase than by mere random assignation of spectral frequencies among the winning bidders by the EAS. Cramton (2009) suggests that the winning bidders from the clock phase could be asked to submit supplementary bids in the assignment phase, in which they could place bids on combinations of licences which are considered an improvement over the current assignment. This bid shall however be subjected to the constraint that the overall amount bid in the clock phase cannot be reduced by any bidder. This provides bidders an opportunity of withdrawing from unwanted combinations and assembling the best package in the supplementary phase. Since they have much better information regarding the likely prices of the objects under auction, the exposure problem is minimised to a great extent, if not completely eliminated.
Finally, we analyse the issue of allocation. The Indian auction contemplates auctioning of three to four frequency bands of 2×5 MHz each in every telecom circle. In contrast, in most of the other countries the minimum bandwidth that was provided to the operators was 2×0 MHz (in some cases it was 2×15 MHz). We have already seen the consequences of the inefficiencies arising out of the fragmentation of spectrum in the case of 2G operators due to a large number of licensees.28 Prasad et al (2009) have noted that incumbents are operating with suboptimal quantities of spectrum. The explosive growth in the subscriber base has strained the networks to the hilt. The limited supply of spectrum has forced the operators to opt for higher reuse of frequency by increasing the base transmission stations (BTS) density. In some circles the physical separation between BTS is less than 100m, which is one of the lowest in the world. Apart from increasing the capital costs, this also leads to higher operating costs in terms of electricity consumption and labour costs and by some accounts health costs to neighbouring residents.29 In addition, the network strains result in dropped calls, blocked calls and poor quality of service. For instance, TRAI (2010) documents the poor quality of service and observes, rather alarmingly, that the congestion is getting progressively worse. For instance, actual call drop rate was 4.3% against a benchmark of less than 2%. In the worst affected category, 15.2% of the cells30 have call drop rates higher than 3%. It seems that in taking a decision about the band plan, we have not learnt appropriate lessons from the 2G case. It would perhaps been better if the spectrum policy decisions were informed by the relative scarcity of inputs other than spectrum alone. Going by the 2G example, while this constraint may not pinch today, it will be a big bottleneck for efficient operations as the demand for data services shoot up. Admittedly, the incumbents may acquire spectrum through mergers and acquisitions but the transaction costs involved in such operations may not be inconsequential.
4.4 Entry and Post-auction Market Structure
The barrier for entrants was far too high relative to existing 2G incumbents with their existing customer bases, established brand names and lower cost of rolling out network (as they can ride on the existing 2G infrastructure).31 Hence it was suggested that the auction format adopted by India at its 3G auction was unlikely to prove attractive to the entrants. It was also suggested that even if the auction failed to attract entrants, it should not result in major headaches to policymakers in regard to issues such as competition during the auction and post-auction market concentration.32 How do these predictions square up to the evidence presented during the actual bidding process?
First, of the nine bidders who actually participated in the bid, none can technically be considered to be an entrant in that every
SAMEEKSHA TRUST BOOKS
Global Economic & Financial Crisis Essays from Economic and Political Weekly
In this volume economists and policymakers from across the world address a number of aspects of the global economic crisis. One set of articles discusses the structural causes of the financial crisis. A second focuses on banking and offers solutions for the future. A third examines the role of the US dollar in the unfolding of the crisis. A fourth area of study is the impact on global income distribution. A fifth set of essays takes a long-term view of policy choices confronting the governments of the world.
A separate section assesses the downturn in India, the state of the domestic financial sector, the impact on the informal economy and the reforms necessary to prevent another crisis.
This is a collection of essays on a number of aspects of the global economic and financial crisis that were first published in the Economic & Political Weekly in early 2009.
Pp viii + 368 2009 Rs 350
Available from
Orient Blackswan Pvt Ltd
Mumbai Chennai New Delhi Kolkata Bangalore Bhubaneshwar Ernakulam Guwahati Jaipur Lucknow Patna Chandigarh Hyderabad Contact: info@orientblackswan.com
march 26, 2011 vol xlvi no 13
single one of them had rolled out 2G services in at least some of the telecom circles in India. Four of them (Bharti, Vodafone, Tata and Reliance) had a significant presence in all the 22 telecom circles while two (Aircel and Idea) had a substantive market share in a large number of circles. The other three bidders (Etisalat, STel and Videocon) had been granted the licences to start rolling out 2G services in January 2008.33 The auction thus was unable to fulfil its objective of attracting new entrants.34 Some of the weaker 2G incumbents chose to not even participate in the auction (e g, HFCL and Telenor).35
Further, given that the three entities that had just started rolling out their 2G services and had low all-India market shares,36 they could have posed a potential competitive threat to the entrenched 2G incumbents for the acquisition of 3G licences. But the evidence from the bid data suggests otherwise. With the exception of STel, which ended up as a winner in three category C circles (Assam, Bihar and Himachal Pradesh), none of the others could make a dent on the bidding process. The valuations spectrum placed by these newcomers was invariably lower than that of the incumbents and the clearing price in virtually all circles was determined by the entrenched incumbents.37 In fact, the strategy of most bidders during the auction was to assemble the package of licences in the circles where they were already fi rmly entrenched (Philip and Gupta 2010).
However, anti-competitive concerns owing to most licences being won by the incumbents should not be a cause for too much worry for the regulators. This is because India has more than six to seven licensed 2G operators in every telecom circle. The business case for acquiring a 3G licence for incumbents arises more out of a promise of the resultant spectrum relieving the network capacity constraints in the short to medium term (thereby preventing revenue leaks and customer churn) rather than the incremental revenue that high-end 3G services are expected to generate.38 Thus even without an entrant the number of operators is sufficient to ensure that consumers would not be affected adversely on account of lack of competitiveness.39
4.5 Impact on Consumers
Some analysts have argued against an auction-determined pricing of spectrum on the ground that high revenue realisations for the government would translate into delayed rollout of network infrastructure, higher tariffs for the consumers and consequently low market growth. Citing the examples of countries like Korea, Japan and Singapore – which have been most successful with 3G services penetration levels – they point out that these countries have exacted the smallest or no licence fees at all because this allows carriers to invest the money into their network. While it can be plausibly argued that cheaper spectrum prices will lead to cheaper services and higher penetration, we are likely to make grave errors if we fail to take local realities into perspective. This notion that low spectrum prices translate into faster roll-out and low cost of services for the users is based upon two premises
(i) that the government can ensure compliance to roll-out obligations, and (ii) that the operators actually pass on the benefi ts of low spectrum prices to the consumers. The validity of both these premises seems to be doubtful under Indian conditions.
Economic & Political Weekly
EPW
First, there is no guarantee that the low licence fee will mean lower cost for the operators. In fact, as witnessed in the recent 2G case, the low licence fee turned out to be a huge opportunity for the political class and for unscrupulous companies to encash the available arbitrage opportunities.40 The ground realities in India in terms of the rampant corruption are simply not comparable to that of Singapore, Japan and South Korea. Second, the tariff level for much of urban India is already very low, so affordability is not a problem. As far as 3G services are concerned, initially, these will be targeted at high-end customers for whom price is not as big a consideration. Lowering the tariff further can help people in rural India and those at the bottom of the pyramid, but those are not the areas where network capacity is a problem or where 3G services will be launched. Therefore lowering the 3G spectrum price might not help these customers much. It would perhaps be better to subsidise the rollout of services in rural and remote areas through other mechanisms such as the Universal Service Obligation Fund. Finally, even with these massive revenue realisations, the operators have not paid such large amounts as to adversely affect their revenue streams. The pan-India licence fee
– when amortised over the period of licence (20 years), constitutes just about 5%-15% of the current annual revenues for the operators.41 With a burgeoning subscriber base, increasing revenues, and Earnings Before Interest Depreciation Taxes and Amortisation (EBIDTA) of service providers ranging from 25%-40% of the current revenues, this payment is hardly a cause for alarmist scenarios regarding the future financial health of the operators.
5 Conclusions and Policy Lessons
From the above discussions, it is clear that the advantages of an auction in allocating a scarce resource such as spectrum at a market-determined price to those who value it most, in a fair and transparent manner, far outweigh the disadvantages of a discretion based allocation system at an administratively defined price, particularly given the political economy realities in India. In this sense the licence for operating services should be delinked from the bundled spectrum. While the former can be awarded to any operator on fulfilment of pre-defi ned technical eligibility criteria on payment of a nominal fee, the latter has to be made available on the basis of a price discovered through an auction.
It should also be noted that while auctions may be a necessary condition for efficient allocation, they are generally not a suffi cient one. It is imperative to ensure that the auction and the supporting institutional design are appropriate in order to contain the adverse fall outs of an ill-conceived and poorly designed auction on the industry.
There are several commendable features in the auction design used for the 3G auctions in India. These include elimination of most gaming behaviour, enhancing substitution among licences and its neutrality to competing technologies for 3G services. By opting for the LAB pricing rule, the auction has been successful in far exceeding the expectations on the revenue realisation front. This was due to a combination of factors. First, by keeping the number of licences on offer lower than the number of entrenched 2G incumbents, it ensured aggressive bidding with the clearing assembling multiple licences with complementarities, without price in every circle being determined by the exit of an exposing them to agglomeration risk. entrenched incumbent. This was further strengthened by the More importantly, the lessons from the fragmentation of the fact that the bidders placed a greater premium on the regret of 2G spectrum seem to have been ignored in the determination of losing over the joy of winning. Since the bidder psychology was the predetermined spectrum management plan. We have argued characterised by a risk loving nature, their bids refl ected a that while this may not hurt now, it will result in poor quality of higher belligerency than a risk neutral profit maximising bidder service issues as the demand for 3G services rises and the (Cramton et al 2009). resultant network constraints push up capital expenditure costs
However, the design is suboptimal insofar as effi cient alloca-for the telecom operators. tion is concerned. Bidding strategies were neither straight for-The auction was unable to achieve the objective of attracting ward, nor did it lead to truthful bidding. The gain in revenue new entrants to encourage stronger competition with the incumwas offset by the loss in efficiency as well as higher complexity bents. Given the competitive disadvantage of the entrants vis-àof bidding strategies. This could have been mitigated by keeping vis the incumbents under the existing rules of the auction, the the bid increments in the final stages of the auction (when ex-newcomers in the 2G space were unable to pose a serious chalcess demand for a circle is one or lower) capped at lower levels lenge to the entrenched 2G operators. All the licences (with the (say Rs 10 crore for category A circles). While this would have exception of STel in a few circles) were bagged by the entrenched increased the number of rounds of bidding and therefore led to 2G incumbents. However, this by itself does not warrant any a prolonged bidding, it would have allowed the bidders greater significant anti-competitive concerns, given the existing market flexibility in the market price discovery process. A modifi ed condition prevalent in India. In fact, it proves to be advantaform of the VCG pricing rule suggested by Cramton et al (2009) geous in the short run, as it prevents further fragmentation of could have been used to ensure truthful bidding on the part of spectrum; and allows the incumbent to overcome the serious bidders, which could have led to a more effi cient allocation. capacity constraints and therefore quality concerns in the
There is scope for further improvement in the auction design existing 2G network. to yield higher allocational efficiencies by allowing winning bid-Finally, we have argued that the auction-determined prices ders to submit supplementary package bids in the assignment paid by the operators are not so high as to raise concerns regardphase to improve upon assignments from the clock phase instead ing the financial health of the sector as a whole, nor is it likely to of the current assignment based on random allocation EAS. As impact the consumers adversely in terms of higher prices for has been discussed earlier, this helps the successful bidders in accessing the services.
Notes
1 Unprecedented growth in subscriber base along with limited availability of spectrum is adversely affecting the quality of service, particularly in the metros.
2 The idea was to grant one additional licence for basic services apart from the existing incumbent, and two licences for mobile services in each telecom circle.
3 For service provision, the entire country was divided into 20 circles, categorised as A, B, or C depending upon their revenue potential. The circles were mostly co-terminus with the DoT’s administrative boundaries and the states. Later two more circles were added to take this number to 22.
4 This was introduced to accommodate the former CDMA operators, who were granted Universal Access Service Licenses (UASL) to migrate to a regime of seamless mobile services from that of limited mobility (Wireless in Local Loop) permitted under the original licence conditions. For a discussion on the resolution of the conundrum relating to the contentious issue of regulatory parity between the traditionally fixed services and the mobile, readers are referred to Jain (2001).
5 Additional spectrum was to be given based on the number of subscribers being served by the operator in lieu of an additional share of adjusted gross revenue, but no further entry fee.
6 The CAG has estimated a loss ranging between Rs 67,000 crore and Rs 1.76 lakh crore; while the latest report by TRAI on 9 February 2011 pegs the loss at around Rs 1.24 lakh crore. See Bamzai (2011) for details.
7 Lots shall be used to refer to individual spectrum licences that are on auction within a telecom circle/service area.
8 Bid increment for each circle is 10% of the previous round clock-price if excess demand in that circle is more than three, 5% if excess demand in that circle is two, and 1% if excess demand in that circle is one or zero (but with excess demand in another circle). If there is excess supply in the given circle (and there is excess demand in other circles), bid increment for the next round will be set at zero. Bid increments are however subject to an overall cap of Rs 40 crore, 20 crore and 7.50 crore for Category A, B and C telecom circles, respectively.
9 The activity rule was incorporated in the SA auction design to prevent bidders from engaging in the practice of “bid sniping”. As in eBay auctions, experienced bidders hide their true intent/ valuation till the very last moment. This results in bidders’ attempt at withholding information regarding their valuations of the object; hence such auctions are better modelled as sealed bid auctions rather than open ascending auctions. It entails ascribing eligibility points to bidders on the basis of their initial deposits for bidding on specifi ed number of lots. The bidder has to bid at close to maximum levels of eligibility (typically 80%100%); failing which his eligibility is curtailed in all subsequent rounds. This enforces the desired bidding behaviour of being a large bidder at the beginning of the auction to be a large bidder at the end of the auction.
10 That is to say, that P is a function of Pt-1 and the
t
excess demand in the round t-1.
11 Considered bid is the clock price corresponding to the highest bid submitted by the bidder in or before the just completed clock round.
12 In case of a tie, bids will be ranked in ascending order of the clock round when the bid was submitted (earlier rounds taking precedence). If there is a tie even by this criterion, then there will be random assignment.
13 The telecom department had originally proposed only Rs 2,020 crore, but it was subsequently revised at the instance of the fi nance ministry.
14 Defined as the total revenue per subscriber per month.
15 As reported by the Cellular Operators Association of India on their website http://www.coai.com/ revenue.php for the quarter January-March 2010 (the quarter just preceding the auctions).
16 Prima facie an argument could be made that auction revenues should be sensitive to the number of subscribers. However, the evidence presented at least in case of India seems to suggest otherwise. A regression of the winning bid price on ARPU shows that about 70% of the variation in winning bid prices across 22 telecom circles of India can be explained by the ARPU alone. Adding number of subscribers to the specification hardly increases any explanatory power. Further, in the revised specification, the coeffi cient of the number of subscribers is statistically not significant; while that of the ARPU is signifi cant even at 0.1% level.
17 A sealed bid process would achieve the same end, but this detracts from the efficient price discovery process. Since there are considerable uncertainties about the true worth of the spectrum, open bidding as in a SA auction permits the bidders to condition their bid responses to the information revealed during the bidding process. First price sealed bid auction results in bid-shading. Inabi lity to factor in signals from competing bidders may also lead to the phenomenon of the “winners’ curse” (bid being higher than the true value of the object in auction). The 1995 auction in India provides a perfect example, where this led to widespread defaults and delayed rollout of services.
18 It is interesting to note that the responsibility for the conduct and administration of the e-auction was devolved upon a consortium of reputed private agencies – N M Rothschild and Sons (India) and DotEcon. Robust and watertight rules and
march 26, 2011 vol xlvi no 13
systems were developed prior to the launch of the auction, eliminating the scope for discretionary intervention on the part of the ministry.
19 Vickrey’s seminal contribution was the insight that if the allocation rules of an auction provide for the objects in question to be awarded to the highest bidder, but at a price determined by the bid of the next highest bidder, the dominant strategy for each bidder is to bid his true value. Vickrey generalisation of the pricing rule applied to a multi-unit action of homogeneous objects requires them to be awarded to the highest bidders, but the price paid by the winning bidders is the opportunity cost of the units won. This was further generalised by Clarke and Groves to multiunit auctions for homogeneous as well as heterogeneous goods. The strengths and weaknesses of using the VCG pricing mechanism in combinatorial auctions are discussed in Ausubel and Milgrom (2006).
20 Consider a simple example from Cramton (2009). Bidder 1 bids $4 for object A, Bidder two bids $4 for object B, Bidder three bids $4 for A and B. Under VCG pricing, objects A and B would be effi ciently allocated to Bidders 1 and 2 respectively and each will pay a price of $0. Bidder 3 may have a genuine resentment against this allocation since he has offered a higher price to the seller.
21 Since the bid increment is a function of the previous clock round bid, in monetary terms the risk for a bidder is higher in highly contested circles. For instance while the final round bid increment for the Delhi circle was Rs 33.17 crore, by comparison the discrete bid increment for Bihar was only Rs 2.01 crore, for Orissa it was only Rs 0.96 crore. A cap on the maximum bid increment could have addressed this issue, but in the event the caps were fi xed so high that they did not come into play at all.
22 In case of a multi-unit auction, the bidder has the option of switching amongst objects.
23 If bidders systematically associate a greater weight on lost profit from losing at profi table prices over gained profit from successful bid shading, then the exclusion of the anticipating loser’s regret of losing at profitable prices leads to LAB yielding higher revenues than HRB.
24 For instance, the magnitude of bid increments in the last two rounds of Mumbai and Delhi was a whopping Rs 33.17 crore and Rs 32.14 crore respectively, whereas for Assam and Himachal it was a measly Rs 0.41 crore and Rs 0.37 crore. It is easier for the bidders to take risk when the discrete bid increments are lower in magnitude.
25 However, auctions with the LAB pricing rule can be demonstrated to yield higher revenues over HRB when a bidder’s fear of losing at profi table prices is sufficiently strong. This may cause bidders in the LAB auction to bid more aggressively than predicted assuming profi t-maximising bidders. See Cramton et al (2009). There were at least four or five serious bidders for the pan-India licence against the availability of three slots. The cost of not getting the 3G spectrum is very high for the leading incumbent operators due to high levels of network congestions (particularly in the metros) and their inability to provide service differentiation in the form data intensive usage facilities like mobile internet to high value customers. This prognosis appears to be vindicated by the Indian auctions.
26 See for instance, the US Advanced Wireless Services Auction when limited substitutability among lots resulted in differing prices for identical lots, in Cramton (2009), pp 7-9.
27 The number of combinatorial possibilities with 22 bids (circles) shall be much smaller than the 71 bids (lots) with a single stage auction.
28 A compact block of 10 MHz spectrum can support proportionally more than twice the number of subscribers as compared to compact block of fi ve MHz spectrum on account of trunking effi ciencies; see Prasad et al (2011).
29 The telecom companies are trying to contain the ill-effects by sharing the tower infrastructure.
30 Cells in the cellular mobile technology refers to the geographical area serviced by a single BTS.
31 Therefore entrants have a lower valuation for the spectrum than the incumbents. Since in SA it is guaranteed that the bidder with the higher valuation will win (unlike in sealed bid auction, where there is some probability of the entrant winning the auction), there is much less motivation for an entrant to participate.
32 For a detailed discussion on this, see Kumar (2010).
33 The licence had actually been granted to Swan telecom for 13 out of 22 circles, but immediately after the issue of licence and before any activity for the rollout of the network could be undertaken, Swan offl oaded a 45% stake to the UAE’s Etisalat for $900 million.
34 The European 3G auctions have shown that the higher the number of entrants, the higher has been the competitiveness on all licences, resulting in higher auction revenues. For instance, see Klemperer (2004), where he compares the UK and Netherlands 3G Auction, conducted at about same time. In UK there were five licences on offer with four 2G incumbents; where as in Netherlands fi ve licences were on offer with five 2G incumbents. The UK auction attracted 13 entrants, while the Dutch attracted only one. Per capita revenue raised from the auctions: UK £650, Netherlands: ¤170. Hence in most auctions, one licence was reserved for an entrant. This served two purposes:
35 Telenor obtained a pan-India license to operate 2G services through acquisition of 60% stake in Unitech.
36 As on 31 March 2010, at the all-India level the percentage market shares (in terms of number of subscribers) of Etisalat, STel and Videocon were 0%, 0.172% and 0.006%, respectively. The situation was not different in any specifi c telecom circle either.
37 For instance, the clearing price in Delhi was set by dropping out of Tata Telecom, in Tamil Nadu by Reliance; in Bihar by Idea, and in Orissa by Bharti. In fact the maximum bids placed by the strongest of the newcomers, Etisalat DB ranged between 15% and 75% of the winning bid (with an average of 48% over the 15 circles on which it chose to bid), clearly demonstrating that the newcomers did not really play a role in determining the auction results. STel managed to secure the circles largely by default because the incumbents ran into budget constraints to defend their turfs in the circles in which they were firmly entrenched and opted out of these category C circles due to limited revenue opportunities here; besides the fact that constraints upon the availability of the 2G spectrum are not binding.
38 In Europe, six years after the 3G auctions, only 19% of the total revenues generated were from data services, the balance coming from voice services. In 2012 this figure is projected to reach 32%.
39 World over, three to four well functioning operators per circle have adequately addressed concerns relating to market concentration. While TRAI argues for six operators per circle, some commentators have argued that having more than three operators in most B and C category circles would prevent the operators from achieving minimum efficient scales of operations besides the problem of over-fragmentation of spectrum (Prasad et al 2008).
40 For instance, immediately after award of 2G licences in January 2008, Swan Telecom offl oaded a 45% stake to the UAE’s Etisalat for $900 million (the company is now known as Etisalat DB), Unitech divested up to 67.25% in its telecom venture to Norway’s Telenor for $1.1 billion (the company is now known as Uninor).
41 I use 11% as the rate of discount since TRAI has suggested this as a benchmark rate for the Weighted Average Cost of Capital (WACC) for the telecom operators. Even at a more conservative discount rate of 15%, the amortised license cost would only be about 6%-18% of the current annual revenues.
References
Ausubel, Lawrence M and Paul Milgrom (2006): “The Lovely but Lonely Vickrey Auction” in Peter Cramton, Yoav Shoham and Richard Steinberg (ed.), Combinatorial Auction, MIT Press, Chapter 1, pp 17-40.
Bamzai, Sandeep (2011): “The Seven Deadly Sins of Sibal”, India Today, 7 March, available at http://indiatoday.intoday.in/site/Story/130242/top-stories /sibal-damages-upas-credibility.html.
Cramton, Peter and Schwartz, Jesse A (2000): “Collusive Bidding: Lessons from the FCC spectrum Auctions”, Journal of Regulatory Economics, 17, 229-52.
Cramton, Peter (2001): “Lessons Learnt from the UK 3G Spectrum Auctions”, Appendix 3, Report on the Auction of Radio spectrum for the Third Generation of Mobile Telephone commissioned by the National Audit Offi ce, UK.
– (2009): “Spectrum Auction Design”, Working Paper, University of Maryland, available at http:// works.bepress.com/cramton/32/
Cramton Peter and Sujarittanonta Pat (2009): “Pricing Rule in a Clock Auction”, unpublished, available at http://works.bepress.com/cramton/154/
DoT (2010): “Auction of 3G and BWA Spectrum: Notice Inviting Applications”, 25 February, Ministry of Communications and Information Technology, available at http://www.dot.gov.in/as/Auction% 20of%20Spectrum%20for3G%20&%20BWA/ 3G%20&%20BWA%20Auctions_Notice% 20Inviting%20Applications.pdf
Government of India (2009a): “Report of the Committee for Allocation of Access (GSM/ CDMA) Spectrum and Pricing”, Ministry of Communications and IT, Department of Telecom.
– (2009b): “Auction of 3G and BWA Spectrum – Revised Information Memorandum”, 19-20.
– (2010): 3G Auction Data Files (accessed 10/2010) available at http://www.dot.gov.in/as/Auction% 20of%20Spectrum%20for3G%20&%20BWA/ new/index.htm
Jain, R S (2001): “Spectrum Auctions in India”, Telecommunications Policy, 25, pp 671-88.
Klemperer, P (2004): “Overview of the European Auction” in Auctions: Theory and Practice, Princeton University Press, pp 151-56.
Kumar, Alok (2010): “3G Spectrum Auctions in India: Have We Learnt the Right Lessons?”, SSRN abstract id 1545045.
Prasad, Rohit and Sridhar Varadharajan (2008): “ Optimal Number of Mobile Service Providers in India: Trade-off between Efficiency and Competition”, International Journal of Business Data Communications and Networking, Vol 4, Issue 3.
Philip, Joji Thomas and Deepali Gupta (2010): “Bidding Over, Battle Begins”, The Economic Times, 21 May, available at http://telecomcircle.com/wp-content/ uploads/2010/05/3G-Economic-Times.pdf.
TRAI (2010): “The Indian Telecom Services Performance Indicators”, January-March, pp 84-88.
Yu, Eileen (2009): “License Auctions Stifl ing Mobile Development”, ZDNet Asia, 16 September, available at http://www.zdnetasia.com/license-auctions-stifl ing-mobile-development-62057816.htm accessed 13 February 2011.
Economic & Political Weekly
EPW
