A Basic Analysis of Entry and Exit in the US Broadband Market, 2005–2008

And, in the case of mobile broadband, where they travel.

For comparability with Xiao and Orazem (2005), we group different service types within the ZIP code for this calculation.

See also the review of broadband demand studies in Hauge and Prieger (2010).

The FCC designed its reporting rules to avoid double-counting of broadband lines. Providers report only if they are the end-user provider and only if they offer “facilities based service.” The latter clause means that if company A owns the infrastructure but rents it to company B to provide DSL, and company B uses no infrastructure of its own, then A files Form 477 and B does not. If, however, company C owns the local loop to the premises but company D has installed a DSLAM in C’s central office (the network equipment necessary to enable DSL service) and has obtained unbundled access to the local loop, then D files in Form 477 and C does not.

The reporting firm does not necessarily have a retail relationship with the end user of the broadband connection, in the case of pure resale or a facilities-based provider selling broadband connection or service to an unaffiliated ISP incorporating it into its own high-speed Internet-access service marketed to end users.

The FCC provided the Form 477 data to the authors under a confidentiality agreement covering data through 2008. Even if later data were available, there would be difficulty in merging the periods, since in 2009 the providers began reporting local service at the Census tract level instead of the ZIP code. The analysis begin in 2005 because earlier reporting requirements exempted small broadband providers (see notes to Table 1).

The “other” category in the FCC ZIP code lists includes traditional wireless services such as T1 (1.544 Mbit/s) and T3 (44.736 Mbit/s) dedicated lines (non-fiber high-capacity digital lines, also known as DS-1 and DS-3). For reasons of consistent data quality that we describe below, we drop the “other” category from our empirical investigation.

The sources of the data are FCC (2005, 2010, 2013).

Due to some violations we found of the FCC’s rule that providers should file one form combining the information for all service companies in a state under the same holding company, our final count of providers does not match official FCC reports exactly.

Within each round, we examined the lists of firms that disappeared from many ZIP codes to determine if a merger, name change, or major sales of assets explained the apparent exit. Similarly, we examined the lists of firms that newly appeared in many ZIP codes to determine if a name change or major purchase of assets explained the apparent entry. We collected information on as many mergers, etc., as we could find from newspapers, the trade press, SEC filings, and Internet sources. Nevertheless, it is important to note that we likely missed some corporate reorganizations and (particularly) asset sales among smaller firms.

Dunne et al. (1988), the seminal empirical study of firm entry and exit in the modern industrial organization literature, study entry and exit in the manufacturing sector of the US economy during the years 1963–1982.

For example, as part of the Adelphia acquisition that was effective in August 2006, Comcast and Time Warner swapped system ownership in various areas. Any apparent exit of Comcast in a market followed by subsequent apparent entry of Time Warner (or vice versa) was thus not treated as actual exit or entry.

The decision of which firm we treat as the exiting company matters in some of the measures of exit we consider below (namely, the measures of exiters’ market share and relative size). In the case of acquisitions, we treat the acquired firm as the exiter. For mergers, we judged which firm appeared to be the dominant partner in the merger (e.g., we treated SBC as the continuing firm and AT&T as the exiting firm in the SBC-AT&T merger, despite the fact that the new firm kept the AT&T moniker). In one case, three firms merged (Choice One Communications, CTC Communications, and Conversent Communications merged to form One Communications Corp. in July 2006), and so in markets where all three competed, two were marked as exiting.

For the few unwindings of 50-50 partnerships we found (e.g., the Comcast-Insight Communications Company partnership unwinding in 2007), we treat any name change in a market (e.g., an Insight system in Illinois changing to a Comcast system after the unwinding) as exit followed by entry, under the assumption that the holding company that filed the area in its Form 477 probably had the upper hand in its management under the partnership.

Most exceptions are for areas like national parks that are not important for this work. The Census Bureau defines each ZCTA to represent a five-digit ZIP code where possible. Refer to Section 2.3 of Grubesic (2008) for in-depth discussion of how ZCTAs relate to ZIP codes and other Census geography.

We mapped ZIP’s to ZCTA’s with the crosswalk files provided by The U.S. Department of Health & Human Resources (see Goodman (2005) for methodology). By far the most common substantive change in USPS ZIP code areas is splitting a ZIP code into two areas, one of which will be designated with a new number. In such cases both new areas are mapped back into the original ZCTA and combined in our methodology.

In the publicly available Form 477 data, company identities are not revealed, and only net entry can be studied [as in Xiao and Orazem (2011)]. Even then, not all net entry is observable, given that the FCC censors the publicly available data when one to three providers are in the ZIP code.

“A Small but Significant and Non-transitory Increase in Price (SSNIP) test asks whether a hypothetical monopolist could profitably impose a small increase in price. If sufficient numbers of buyers would switch to alternative products or to suppliers at other locations such that the price increase is unprofitable, then the market definition must be expanded to include at least some of those substitute products or locations” (Connolly and Prieger, 2009).

Of course, this statement is subject to falsification by reductio ad absurdum. Clearly if the price of broadband service were $1000 in Household A and $10 in neighboring Household B, there would be ways for the households to mutually benefit from “trade,” (for example, through Household A piggybacking off the wireless home network of Household B).

For a discussion of how traditional market definition may proceed for the broadband industry, see US DOJ (2010), Section 3.1.

See also the discussion of the issue of market definition in Connolly and Prieger (2009), Section 2.4.1.

Although the FCC data includes Puerto Rico, we exclude those observations from our study.

For the complete results for all market types, we refer readers to Connolly and Prieger (2013).

Since the latter includes the former, ESH is bounded between 0 and 1, inclusive.

Since the latter includes the former, XSH is bounded between 0 and 1, inclusive.

We found instances of obviously mistyped ZIP codes in the Forms (e.g., those where the ZIP code does not match the state to which the firm’s Form 477 purportedly pertains).

This happens mainly with smaller firms; we cannot be sure a non-filing firm did not have a legitimate reason for not filing.

Given our extensive checking of the provider names, we expect errors of this type to be rare.

A referee raised this issue.

Dunne et al. (1988) drop small manufacturing firms since the geographic extent of their market definition is the entire US and small firms are not very important in that context. The nature of broadband service provision, however, means that even small firms may be very important to subscribers within the area they serve. Subscribers cannot go to larger firms offering service in other areas, and the relevant options for the household or business include all market participants, no matter how small. Furthermore, excluding smaller firms would exclude virtually all rural telecommunication carriers, which are important providers of broadband in many rural areas.

The simple average of the manufacturing entry rates reported over 1963–1982 in Dunne et al. (1988) (for firms of all sizes) is 9.8%.

See Siegfried and Evans (1994: p. 127).

One unanswered question is why is the entry rate is always higher in the December filings. One possibility is that some firms did not actually update their mid-year data as they were supposed to, but instead submitted the same ZIP code list as for the previous December filing. If that were so, however, then the exit rates should always be substantially higher in the June periods, which is not the case.

By our count, there were 1226 distinct holding companies offering some sort of broadband service in June 2005, and 1344 in June 2008 (for a continuous growth rate of 3.1% per year).

See footnote 17.

To be precise, the entry rates for the US-BB market for period t in the table are calculated as ∑j=18ERj(t)8, the simple average of the entry rates across the eight broadband types j.

Dunne et al. (1988) delineate product markets at the four-digit SIC level, which is an intermediate level of aggregation between the US and US-BB markets. Mobile and wireline communications services have different four-digit SIC codes, but all types of wired communications have the same code.

ER is undefined for half of the ZIP-BB markets; in the great majority of such cases there is no entry. ER is undefined most often for the least-common broadband types. ER is undefined for BPL in virtually all markets, and is undefined for TFW in 79% of ZIP code-rounds. Other modes of service with a large number of ZIP codes-rounds with undefined ER are fiber (63%) and SDSL (61%). At the other end of the spectrum, ER is undefined for satellite service in only 13% of markets. In the working paper (Connolly and Prieger, 2013), we calculated an alternative entry rate with NT_i(t) in the denominator, which unlike ER is defined for the first entrant, but the summary statistics for entry did not differ greatly.

However, additional cost created by differences in the physical characteristics of electricity networks, as well as the fact that electricity lines are electromagnetically “noisy” environments for communications compared with the phone companies’ networks mean that the cost advantages of using existing plant is attenuated.

For the argument that wireless broadband has lower sunk cost than wired broadband, see US DOJ (2010), Section 2.5.

Four satellite providers remain in the data throughout all rounds, one of which offers service only in Alaska.

In many theoretical models of industry dynamics, the entry rate falls as the industry matures. See Prieger (2007) and Klepper (1996).

For a review of the confirmatory empirical evidence on the importance of sunk cost as an entry barrier, see Siegfried and Evans (1994), Section II.B.1.a, and references cited therein.

A firm beginning to offer ADSL may be the first, second, or third entrant in a given ZIP code in which it offers service, which explains both the higher entry rates in the ZIP-BB market as well as the great variance in the entry rate.

Since there is only one BPL provider at most in a market, the entry rate for BPL is always either zero or undefined, which is why it is ranked last in terms of ER.

It is important to note that “entry” for the satellite firms mainly reflects changes in demand, not supply, since the firms do not deploy new infrastructure to “enter” a ZIP code. However, the firms may have engaged in targeted marketing efforts to expand demand in certain areas.

The small exit rate for satellite service comes from fact that only two firms exited the national market during the period, and that the service footprints of the other providers did not change. Exit in the satellite ZIP-BB market nearly always comes from losing all customers in a ZIP code, not from making the service physically unavailable.

By conventional measures, the correlations are mostly statistically significant. However, since the data are a census of the entire population, from the viewpoint of finite population statistics the correlations are descriptive population quantities (Pfeffermann 1993), not estimates, and thus have no variance. The significance levels reported in the tables are for the interpretation, more familiar to econometricians, of the data as a finite population drawn from an infinite superpopulation.

Since NT_i(t)=NT_i(t–1)+NE_i(t)–NX_i(t), it follows that even if there is a fixed effect in the model for NT_i(t), there will not be one in ΔNT_i(t)=NE_i(t) – NX_i(t). Thus it is possible there are no fixed effects in NE and NX even when there are in NT.

The theoretical model of Asplund and Nocke (2006) shows that entry and exit rates may rise together in response to an increase in fixed costs.

See Siegfried and Evans (1994: p. 147), and references cited therein.

The pressure to remain in an unprofitable market need not take the form of formal designation as a broadband carrier of last resort. Regulators in some states, and the FCC itself, have pushed telecommunications firms for many years to expand their broadband offerings, especially in areas labeled “disadvantaged.” For example, the FCC approved the SBC acquisition of Ameritech in 1999 only subject to an agreement by the company to promote broadband Internet access (among other conditions). In particular, SBC was required to locate at least 10% of their advanced service facilities in low-income areas in the Ameritech region. State regulators in Ameritech’s operating region (Illinois, Indiana, Michigan, Ohio, and Wisconsin) also pushed the merged firm to accelerate broadband deployment. Presumably neither the federal nor the state regulators would have been satisfied to see the firm begin to offer broadband in low-income areas only to have the service offerings cease after a short period. See Prieger and Hu (2008).

See US DOJ (2010), Section II.E.

See Siegfried and Evans (1994: p. 145): “Tangible barriers to exit may include sunk costs in durable, industry-specific assets, which discourage exit because such assets do not have valuable alternative uses….” These authors conclude the balance of the empirical evidence supports this assertion.

Siegfried and Evans (1994), op. cit., Section II.B.

The National Telecommunications and Information Administration (NTIA) coordinates with the FCC to collect the data for the National Broadband Map (NBM; see http://www2.ntia.doc.gov/broadband-data). The NBM data are available semiannually since June 2010 at the level of the census block.