The determinants of price in 3G spectrum auctions

ABSTRACT

This article examined the empirical relationship between prices paid for national 3G wireless licenses when spectrums were sold by auction. The reduced-form modelling approach was based on the premise that auction design features can, and do, affect both final (revenue) and intermediate (viz., auction competitiveness and license assignment) auction outcomes. Importantly, the analysis recognizes, and explicitly allows for, the endogeneity of bidder entry and sample selection arising from an absence of bidding. Generally, these key features have only received attention in isolation. In particular, the study addressed the principal question of: which regulator chosen auction design variables determine prices paid in spectrum auctions? The analysis showed that flexible-package formats increased prices paid, and that higher reserve prices had a dampening effect. The article also showed that longer license terms and more competitive auctions (as measured by the bidders-to-licenses ratio) enhance prices paid.

JEL CLASSIFICATION:

• D44
• L96

KEYWORDS:

• Spectrum auctions
• prices paid
• bidder participation
• sample selectivity

Acknowledgements

The authors thank Felix Chan, Christian Dippon, Alan Duncan, Bill Greene, Mark Harris, Jerry Hausman, Richard Marsden, Leslie Marx, and participants at the Curtin Econometrics Study Group Workshop in Honour of Jerry Hausman, for helpful comments. Data provided by DotEcon Spectrum Awards database is gratefully acknowledged.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ Radio spectrum is an exclusively licensed scarce (non-depletable and non-storable) natural resource (Faulhaber and Farber 2002).

² Historically assignment was by government fiat; however with the emergence of competitive mobile telephone markets more arms-length processes developed (McMillan 1994).

³ New Zealand in 1989 was the first nation to authorize the use of competitive bidding (Crandall 1998). Wireless license auctions were held in 1990.

⁴ Hazlett (2001) argues that making spectrum available without charge is inefficient as idle white spectrum or rigid allocations do not respond to changes in demand.

⁵ The requirement of profitability is somewhat strong, as evidenced by the early 3G auctions. While the guiding principle is certainly profitability, other considerations are blocking competitors and survival. There are some famous quotes after the UK and German 3G auctions that state that bidding was not primarily a question of profits but simple survival. If spectrum was not obtained, then the company would have to exit the market. This attitude impacts on the market in two ways. First, the bidder might exceed profitable levels in its bidding and then exit the market. An example of this is BT (British Telecom) which had to spin off its mobile operations. Second, when bidders pay in excess of profits, then retail prices are bound to increase. This was the concern in the AWS auction in Canada.

⁶ Estimates of deadweight loss are between 17 and 56 cents for every extra dollar raised in taxes (Ballard, Shoven, and Whalley 1985).

⁷ Also, auctions are a more transparent means to assign licenses (Cramton 2002, 608). Standard auctions at best ensure those bidders with the highest private value wins, rather than the bidder with the highest social value. Private and social values often diverge as winners compete in a marketplace. One set of winners may lead to a more collusive industry structure. For example, licenses are worth more to an incumbent than an entrant, because of greater aftermarket power an incumbent has without an entrant (Cramton 2002, 608).

⁸ Historically, the size of spectrum lots is specified by regulators prior to auctions. However, recently several regulators specified smaller lots to allow bidders to flexibily determine the size of the package licensed.

⁹ The process is slow, e.g. in 1992 the WRC IMT-2000 defined standards for 3G system deployment. Only in year 2000 was service provision designated to frequency bands (Gruber 2005, 225).

¹⁰ Herzel (1951) proposed the auction TV licenses to allow the market to decide on the colour TV standard. Later, Coase (1959) developed the property rights approach to spectrum assignment.

¹¹ Cox, Dinkin, and Swarthout (2001) experimentally test for endogenous entry in common value auctions. They posit the existence of reservation opportunity costs (expected profit from entering an auction). Clearly, agents only participate when an agents alternative income is less than or equal to the reservation opportunity cost.

¹² Deltas and Jeitschko (2007) demonstrate (Lemma 6) that decreased transaction costs increase participation.

¹³ Other formats are sometimes encountered, e.g. the Danish regulator used a 4th price sealed-bid auction. This format is similar to a standard sealed-bid auction.

¹⁴ Only Norway in 2003 and 2007 did not specify reserve prices.

¹⁵ In attempting to extract additional surplus from the highest bidder, the reservation price is set too high when the price is above the highest bidder valuation.

¹⁶ An anonymous referee is thanked for suggesting the referral of explanatory variables to their economic fundamentals that create value for spectrum licenses.

¹⁷ Specifically, costs of converting spectrums into retail services are expected to be lower when the spectrum covers a greater number of population. With the higher expected profit margin (per capita revenue less the associated technological, retailing, and other service costs), bidders are willing to place higher bids.

¹⁸ In contrast, 4G spectrum has been used increasingly more for continuous data connection than for voice call services. Thus, one can similarly argue that the price of fixed line broadband service and that of mobile broadband service should be positively or negatively associated depending on whether these services are considered complements or substitutes by consumers. An implication is that an application of the econometric model presented in this article to the analysis of 4G spectrum auction price should consider including the price of fixed broadband service as a determinant in addition to the price of fixed line voice call service.

¹⁹ The ability to resell spectrum could increase bidder’s private valuations of the licenses being auctioned. Historically, in most countries, post-auction spectrum trading was not allowed. However, more recently in some countries, regulations were introduced to permit spectrum trading within specific mobile bands. Whether regulations require modification to original licenses, or just defers to relevant regulations varies. However, all countries require that notification of any intended spectrum trades be made to the regulator, who can veto any proposed trades under certain conditions, e.g. on competition, national security, or efficient use grounds. For instance, often reselling spectrum rights is not encouraged because of concerns that resale could increase the chance of collusion in retail service aftermarkets. Importantly, none the arguments included in our spectrum price account for possibility of spectrum being resold post-auction. We thank an anonymous reviewer for altering us to this possibility.

²⁰ An auction with reserve price only results in a transaction when the reserve is met (Lucking-Reiley et al. 2007).

²¹ Klemperer (2002a) and Greenstein and Rosston (2009) list discouraging collusion an issue of primary importance in practical auction design.

²² This is to ensure that bidding is not strategically delayed and asymmetric information is not withheld by bidders until the last moments of the auction.

²³ For instance, the ascending auction for Federal Communications Commission Auction No. 31, which sold spectrum licenses in the 700 MHz band, permitted bids for any of 4095 packages of the 12 licenses.

²⁴ That is, three auctions that allocate 2G and 3G spectrums are not considered. In particular, besides 2100 MHz spectrum, the Egyptian (2006), Saudi Arabian (2007), and Netherlands (2012) auctions allocated 900 MHz and 1800 MHz spectrums. Also, because of the national focus, details of six regional licenses (i.e. Australia 2001, United States 2006, Brazil 2007, Canada 2008, United States 2008, and Mexico 2010) are not considered.

²⁵ The crucial identifying assumption is that auction design variables are independent of the unobserved component $\epsilon$ (Athey, Levin, and Seira 2011).

²⁶ In particular, for empirical estimation of model (2) in the next section, the two matrices contain x = {INCOME, POP, FCONN, FCALL, AREA, TERM, ACTIVITY, FLEXIBILITY, RESERVE, SEALED, COMP}, and w = {POP, AREA, FIXED, PERCENT, RESERVE, SETASIDE}.

²⁷ Within a generic non-linear framework, Terza et al. demonstrated that 2SRI is consistent, while two-stage predictor substitution (2SPS) is not. Terza, Basu, and Rathouz (2008, 534) asserted that the 2SRI method is first proposed by Hausman (1978) in a linear model context. Consistent 2SRI methods for specific non-linear models were developed by Blundell and Smith (1989, 1993), Newey (1987), and Rivers and Vuong (1988).

²⁸ See Athey, Levin, and Seira (2011) who authoritatively study the U.S. timber industry.

²⁹ Results from estimating the probit model for license assignment is reported in Appendix Table A1. The inverse mills ratio, $\hat{\lambda }=\varphi (r-\widehat{{\alpha }^{\prime }}\mathbf{w})(1-\mathrm{\Phi }(r-\widehat{{\alpha }^{\prime }}\mathbf{w}){)}^{-1}$, obtained from this model is then used to correct sample selection effect in the estimation of the price equation.

³⁰ This reported negative RESERVE coefficient suggests that the reserve prices were set non-optimally for revenue maximization (Elbittar and Unver 2003).

³¹ The authors thank Alan Duncan and Jerry Hausman for suggesting this test. The results are not reported, but available on request.

³² If the coefficient sign of RESID was positive, it is expected that the effect of COMP (in the uncorrected price model) is upward biased.