ABSTRACT
A Macroeconomic Model of Federal and Commercial Spectrum Usage introduces a methodology to assess the macroeconomic impact of electromagnetic spectrum decisions and provides general results on a case study of the economic impact of shifting from fourth generation (4G) to fifth generation (5G) wireless technology. This article provides a summary introduction to this new research tool.
This research advances the state of the art of spectrum economic modeling by offering an evolution over previous work through the development of a flexible, tailorable, data-driven, and transparent model for assessing national-level economic impacts of spectrum decisions. The research tailors a best-in-class computable general equilibrium (CGE) model to the electromagnetic spectrum domain. It prototypes a network infrastructure model (NIM) as a method for incorporating technological change into the CGE model, extends the US Applied General Equilibrium (USAGE) model to incorporate a new 5G wireless services industry, and explores industry-specific impacts of 5G.
This research was made possible by developing new methodologies to integrate data sets across multiple domains including 4G and 5G wireless gross and network demand, wireless network cell sizes and throughput, demand over geographic coverage areas, downlink allocation and efficiency, mobile network operators, channelization, cost of production for 4G and 5G wireless services (capital, labor, and other intermediate goods), operating expenses, network buildout strategies, total factor of productivity assumptions, depreciation rates, useful life of capital, residual value of assets, national income product account data, and Census and Bureau of Economic Analysis data on population, labor supply, gross domestic product (GDP) components, inflation indices, and exchange rates.
Although there are data limitations and unknown details about the future 5G wireless services industry, the research team addresses these by using a CGE model and treating the new 5G industry as nearly identical to the existing wireless industry, with the major difference being that the new 5G industry sells its services to the existing wireless industry, which in turn resells that service to its customers. The team also makes two critical assumptions. The first assumption is about how much more efficient the wireless services industry is in the 4G+5G future compared to the future with just 4G technology. The second assumption details how much of the composite industry is made up of 5G technology. In addition to building the model, the team examines relationships between parameters and variables of the different components to test the overall sensitivity and develops initial findings. This approach facilitates forward progress, industry engagement, and trade space analysis.
Results suggest that by making data cheaper and more ubiquitous, 5G could increase US GDP between $347 billion and $536 billion as of 2030, depending on the relative price of 4G to 5G and the amount of data carried by 5G. While additional marginal spectrum for 5G does have some initial impact on network cost in 2025, it has a negligible impact on real GDP by 2030. This is because the additional spectrum does little to delay the densification of the network based on the anticipated demand profiles.
Given that 5G is expected to be a general-purpose technology, it is likely that all industries in the economy will become more efficient. This could add between $35 billion and $165 billion to real GDP in 2030.
Applications of the model could inform future federal spectrum sharing and allocation policies, 5G and sixth generation (6G) network evolution strategies, and balancing the value of federal and nonfederal spectrum. This research contributes to addressing spectrum issues with an impartial, transparent, and consistent methodology across stakeholders, in the public interest, and as a whole-of-nation.1
All wireless communications depend on effective and efficient access to the electromagnetic spectrum. Spectrum is a critical national resource that enables national and economic security. Demand for spectrum is growing, whereas its supply is practically limited.
A Macroeconomic Model of Federal and Commercial Spectrum Usage is one of the multiple internally-funded research projects on the topic of Spectrum Innovation for the Whole-of-Nation. This research is developing innovative toolsets and capabilities to enable more data-driven decisions that help meet the United States (US) objective of using electromagnetic spectrum as efficiently and effectively as possible while balancing economic, national security, science, safety, and other national goals now and in the future.
The project team built a capability to conduct objective and transparent macroeconomic analysis of the impact of spectrum decisions on the US economy. This initial proof-of-concept model addresses the following questions: How will spectrum affect the US GDP, based on a shift from 4G to 5G wireless technology? What is the marginal value of additional spectrum for 5G?
To address these questions rigorously and objectively, important details and assumptions must be made about the future telecommunications industry. For example, how 5G’s compositions of costs, investments, and customers differ from 4G will be very important in computing the economic impact. Unfortunately, such details are unknown and difficult to forecast without access to confidential telecommunications data. Other research that predicts the economic impact of spectrum using methodologies like input–output faces the same data limitations, yet still make rather bold projections about the economic impact.
One of the purposes of this research is to understand the sensitivity of such calculations to the myriad of assumptions often overlooked by other analyses. To do this, the team developed a modeling framework that consists of two primary components:
A wireless NIM that estimates the relative costs of 4G and 5G services given different allocations of spectrum.
The CGE model that translates the relative costs of wireless services into economy-wide impacts.
When combined, these two components trace the economic benefits of spectrum to providing more efficient wireless services. In addition to impacting the price of data, 5G will also likely influence the efficiency and productivity of industries. The team addressed the effect of 5G on productivity with a generalized approach and a rough order of magnitude approach.
Following this introduction, the article includes sections on methods, findings, testing and sensitivity analysis, and discussion.
Methods
This section describes the model’s framework, called the Spectrum Macroeconomic Model. The research team designed this model to be flexible, tailorable, data-driven, and transparent. The Spectrum Macroeconomic Model has five key elements, outlined in Figure 1.
Spectrum Pool Assignment: Historical Federal Communications Commission (FCC) auctions and other planned bands or emerging analyses are evaluated to determine the baseline allocation of spectrum that will be available to 4G and 5G providers. The economic impact of additional spectrum is assessed by deviating from the baseline allocations.
Data Demand: Using publicly available information, the total amount of data serviced by both 4G and 5G technologies is forecast through 2030. Beyond 2030, the next generation of wireless technologies is expected to debut and begin the decline of 5G.
NIM: The NIM uses the data demand forecasts and spectrum pool assignments as inputs in deriving the capital expenditures and operating costs for delivering 4G and 5G telecommunication services. Specifically, the NIM calculates the relative unit cost of 5G services and the share of total wireless industry costs attributed to 5G.
Use Case Scenarios: Qualitative assessments characterizing 5G’s potential impact on industry productivity are developed. Each assessment quantifies how much 5G may reduce cost and/or increase output. Six different industries are assessed: real estate, healthcare, accommodation, trucking, agriculture, and electric power generation.
CGE Model: The CGE model uses the output of the NIM and any use case assumptions to assess the macroeconomic impact of 5G and any marginal changes to the spectrum allocation. Like all CGE models, it describes the economy as a series of interconnected markets, each of which allocates finite resources by adjusting prices until demand equals supply. In this specific version, the wireless services industry is decomposed so that 5G is an explicit component of that industry. The model was originally developed by Peter Dixon and Maureen Rimmer at Victoria University.2 It is an extension of the USAGE model.
Key elements of the model are described in the following sections.
Spectrum Pool Assignment
Although 5G networks began rolling out in 2020, they do not yet deliver the promised gigabit level speeds that will come with mature 5G infrastructure.3 In reality, the 2020 deployment could be characterized as 4.5G, where providers are re-farming spectrum from discontinued second generation (2G) and third generation (3G) networks and applying new technology solutions to improve the network speeds as operators transition to deploying full 5G in the 3.5 GHz and millimeter wave frequencies.
With these observations in mind, the commercially available spectrum used to provide wireless services in 2019 is categorized into three cell sizes (micro, macro, and pico—see Table 1) and assigned to a generation of technology (2G, 3G, 4G, and 5G—see Table 2). This assignment is based on the first technology generation to use the frequency and assumes that the frequency can be reused by all subsequent generations. The spectrum assigned to 2G and 3G applications is assumed to be aggregated into the 4G pool. This means that when this article refers to 4G spectrum, it refers to the pool of spectrum used for 2G, 3G, and 4G wireless services.
Bands made available post-2019 (i.e., Citizens Broadband Radio Service [CBRS], millimeter wave [mmWave],4 C-band) are assumed to comprise the core spectrum for 5G wireless services and applications. Educational Broadband Service/Broadband Radio Service (EBS/BRS), while likely the core of the T-Mobile 5G network, was first introduced as a 4G band. Table 3 presents the total spectrum allocated to each generation and serves as a baseline assumption used by the NIM to calculate infrastructure requirements. In the case where a specific auction bridges two frequency bands, such as AWS-1 in the 1.7 GHz/2.1 GHz frequencies, the lower of the two frequencies was used. As discussed next, this assumption also influences the forecasts for annual demand served by the 4G and 5G networks.5
Cell Assumptions
. | Definition . |
---|---|
Macro Cell | Suited for spectrum up to 1,800 MHz |
Micro Cell | Suited for spectrum between 1,801 and 10,000 MHz |
Pico Cell | Suited for spectrum higher than 10,000 MHz |
. | Definition . |
---|---|
Macro Cell | Suited for spectrum up to 1,800 MHz |
Micro Cell | Suited for spectrum between 1,801 and 10,000 MHz |
Pico Cell | Suited for spectrum higher than 10,000 MHz |
Spectrum Assignment Table
FCC Auction . | Approximate Frequency (MHz) . | Total MHz Offered . | Initial Technology Generation (Cellular Frequencies in the United States 2021) . | Cell Type . | Source . |
---|---|---|---|---|---|
Cellular | 850 | 50 | 2 | Macro | (Musey and Keener 2018) |
PCS | 1,900 | 90 | 2 | Micro | FCC Auctions 4 & 11 (FCC n.d.) |
AWS-1 | 1,710 | 90 | 3 | Macro | FCC Auction 66 (FCC n.d.) |
SMR | 900 | 18.4 | 2 | Macro | FCC Auctions 1, 3,7, 21, 41, 42 (FCC n.d.) |
700 MHz | 700 | 86 | 4 | Macro | FCC Auctions 33, 44, 73 (FCC n.d.) |
Additional PCS | 1,900 | 10 | 2 | Micro | FCC Auction 96, PCS-H Block (FCC n.d.) |
WCS | 2,400 | 30 | 4 | Micro | (Musey and Keener 2018) |
AWS-4 | 2,000 | 40 | 3 | Micro | (Musey and Keener 2018) |
EBS/BRS | 2,500 | 194 | 4 | Micro | (Musey and Keener 2018) |
AWS-3 | 1,700 | 65 | 3 | Macro | FCC Auction 97 (FCC n.d.) |
600 MHz | 600 | 70 | 4 | Macro | FCC Auction 1002 (FCC n.d.) |
24 GHz | 24,000 | 1,100 | 5 | Pico | FCC Auctions 56, 102 (FCC n.d.) |
28 GHz | 28,000 | 1,150 | 5 | Pico | FCC Auctions 17, 101 (FCC n.d.) |
37 GHz | 37,000 | 1,000 | 5 | Pico | FCC Auction 103 (FCC n.d.) |
39 GHz | 38,000 | 1,400 | 5 | Pico | FCC Auctions 30, 103 (FCC n.d.) |
47 GHz | 47,000 | 1,000 | 5 | Pico | FCC Auction 103 (FCC n.d.) |
CBRS Auction 105 | 3,500 | 70 | 5 | Micro | FCC Auction 105 (FCC n.d.) |
C-band Auction 107 A Block | 3,700 | 100 | 5 | Micro | FCC Auction 107 (FCC n.d.) |
C-band Auction 107 BC and ABC Blocks | 3,700 | 180 | 5 | Micro | FCC Auction 107 (FCC n.d.) |
SENSR | 1,300 | 30 | 5 | Macro | TBD |
FCC Auction . | Approximate Frequency (MHz) . | Total MHz Offered . | Initial Technology Generation (Cellular Frequencies in the United States 2021) . | Cell Type . | Source . |
---|---|---|---|---|---|
Cellular | 850 | 50 | 2 | Macro | (Musey and Keener 2018) |
PCS | 1,900 | 90 | 2 | Micro | FCC Auctions 4 & 11 (FCC n.d.) |
AWS-1 | 1,710 | 90 | 3 | Macro | FCC Auction 66 (FCC n.d.) |
SMR | 900 | 18.4 | 2 | Macro | FCC Auctions 1, 3,7, 21, 41, 42 (FCC n.d.) |
700 MHz | 700 | 86 | 4 | Macro | FCC Auctions 33, 44, 73 (FCC n.d.) |
Additional PCS | 1,900 | 10 | 2 | Micro | FCC Auction 96, PCS-H Block (FCC n.d.) |
WCS | 2,400 | 30 | 4 | Micro | (Musey and Keener 2018) |
AWS-4 | 2,000 | 40 | 3 | Micro | (Musey and Keener 2018) |
EBS/BRS | 2,500 | 194 | 4 | Micro | (Musey and Keener 2018) |
AWS-3 | 1,700 | 65 | 3 | Macro | FCC Auction 97 (FCC n.d.) |
600 MHz | 600 | 70 | 4 | Macro | FCC Auction 1002 (FCC n.d.) |
24 GHz | 24,000 | 1,100 | 5 | Pico | FCC Auctions 56, 102 (FCC n.d.) |
28 GHz | 28,000 | 1,150 | 5 | Pico | FCC Auctions 17, 101 (FCC n.d.) |
37 GHz | 37,000 | 1,000 | 5 | Pico | FCC Auction 103 (FCC n.d.) |
39 GHz | 38,000 | 1,400 | 5 | Pico | FCC Auctions 30, 103 (FCC n.d.) |
47 GHz | 47,000 | 1,000 | 5 | Pico | FCC Auction 103 (FCC n.d.) |
CBRS Auction 105 | 3,500 | 70 | 5 | Micro | FCC Auction 105 (FCC n.d.) |
C-band Auction 107 A Block | 3,700 | 100 | 5 | Micro | FCC Auction 107 (FCC n.d.) |
C-band Auction 107 BC and ABC Blocks | 3,700 | 180 | 5 | Micro | FCC Auction 107 (FCC n.d.) |
SENSR | 1,300 | 30 | 5 | Macro | TBD |
Spectrum Allocations Across Generations
Generation . | Macro MHz . | Micro MHz . | Pico MHz . | Generation Total MHz . |
---|---|---|---|---|
2/3/4G | 379.4 | 364 | N/A | 743.4 |
5G | 30 | 350 | 5,650 | 6,030 |
Grand Total | 409.4 | 714 | 5,650 | 6,773.4 |
Generation . | Macro MHz . | Micro MHz . | Pico MHz . | Generation Total MHz . |
---|---|---|---|---|
2/3/4G | 379.4 | 364 | N/A | 743.4 |
5G | 30 | 350 | 5,650 | 6,030 |
Grand Total | 409.4 | 714 | 5,650 | 6,773.4 |
Data Demand
The forecasts of data served by 4G and 5G systems are estimated by calculating the total US traffic demand, and then allocating that total into 4G and 5G networks. Historical US data traffic in Table 4 is estimated by retrieving the annual historical North American non-offloaded9 mobile data and Internet traffic10 and prorating by 96%11 to get US yearly traffic estimates. Per industry reports, total mobile data traffic12 is predicted to increase 31%13 annually through 2030.14
Historical Data Used in the United States6
EB of Non-Offloaded Traffic per Year . | 2012 . | 2013 . | 2014 . | 2015 . | 2016 . | 2017 . | 2018 . | 2019 . |
---|---|---|---|---|---|---|---|---|
Cisco North America | 2.7 | 4.7 | 5.497 | 6.7 | 16.4 | 15.1 | 21.6 | 28.48 |
US Portion | 2.56 | 4.55 | 5.27 | 6.42 | 15.77 | 14.53 | 20.78 | 27.2 |
The basis for disaggregating total traffic consists of projecting 4G’s efficiency rates (exabytes [EB] provisioned per MHz of spectrum) and applying those rates to the amount of spectrum allocated to 4G in the baseline. More specifically, 4G’s efficiency rates are calculated in 2009, 2014, and 2019 from historical traffic levels and spectrum allocations. As 4G was introduced and grew rapidly between 2009 and 2014, its efficiency rate improved 57%. But as it matured and stabilized between 2014 and 2019, the efficiency rate improved just 33%, a growth rate decline of 42% since its first five years. Continuing this assumption of a maturing technology, the baseline forecast assumes 4G’s efficiency rate increases by 19.6% between 2019 and 2024 and by 11.0% between 2024 and 2029.
Next, the amount of spectrum allocated to 4G is projected for the future. It is assumed that 10% of the 4G spectrum capacity in the initial allocation will transition to 5G each year starting in 2021. This supports activities such as T-Mobile using EBS as the core for its 5G network. This assumption implies that approximately 65% of the 4G spectrum will be transitioned to supporting 5G by 2030. Similar to the sunsetting of 2G/3G with the emergence of 5G, it is expected that 4G will sunset in 2030–2035 as 6G emerges, further supporting the 10% transition rate assumption. Multiplying the forecasts of 4G’s spectrum pool and efficiency rates yields its future traffic forecast (EB/year). The delta between the 4G estimated traffic and the total future traffic is assumed to be the 5G annual traffic. Table 5 lists the traffic forecasts for 4G and 5G in the first two columns.
The average monthly demand in Table 5, columns five and six, is calculated by assuming that the number of wireless subscriptions grows proportionally with the United States.15 It also assumes that 5G captures 50% of all subscriptions by 2025 and 75% of all subscriptions by 2030. By inspection, the average data demanded per subscription is relatively in line with expected rates.16
Because the final network demand depends on not just the average demand per user but the number of users who are simultaneously trying to access a network (e.g., the total instantaneous demand), the rate of simultaneously transmitting users for both 4G and 5G is captured in the final two columns of Table 5. The 4G rate was provided by subject matter experts. The 5G rates were asserted to begin with the 4G rate and double by 2030 to reflect new verticals and the Internet of Things.
Baseline Forecast of 4G and 5G Demand
Year . | Total 4G Demand (EB) . | Total 5G Demand (EB) . | Percentage of Total Subscriptions Using 4G . | Percentage of Total Subscriptions Using 5G . | 4G Avg. Demand/Subscription/Month (GB) . | 5G Avg. Demand/Subscription/Month (GB) . | 4G Simultaneous Emitter Rate . | 5G Simultaneous Emitter Rate . |
---|---|---|---|---|---|---|---|---|
2019 | 27 | 0.3 | 0.99 | 0.01 | 6.87 | 6.25 | 0.02 | 0.02 |
2020 | 33 | 3.1 | 0.9125 | 0.0875 | 8.86 | 8.79 | 0.02 | 0.021 |
2021 | 35 | 11.7 | 0.825 | 0.175 | 10.47 | 16.42 | 0.02 | 0.022 |
2022 | 38 | 23.5 | 0.7375 | 0.2625 | 12.52 | 21.87 | 0.02 | 0.023 |
2023 | 41 | 39.5 | 0.65 | 0.35 | 15.19 | 27.46 | 0.02 | 0.024 |
2024 | 44 | 61.3 | 0.5625 | 0.4375 | 18.78 | 33.82 | 0.02 | 0.025 |
2025 | 44 | 93.7 | 0.5 | 0.5 | 21.06 | 44.94 | 0.02 | 0.026 |
2026 | 44 | 136.2 | 0.45 | 0.55 | 23.33 | 59.00 | 0.02 | 0.027 |
2027 | 44 | 191.9 | 0.4 | 0.6 | 26.17 | 75.73 | 0.02 | 0.028 |
2028 | 44 | 264.9 | 0.35 | 0.65 | 29.83 | 95.91 | 0.02 | 0.029 |
2029 | 45 | 360.7 | 0.3 | 0.7 | 34.71 | 120.50 | 0.02 | 0.03 |
2030 | 43 | 488.0 | 0.25 | 0.75 | 39.79 | 151.28 | 0.02 | 0.031 |
Year . | Total 4G Demand (EB) . | Total 5G Demand (EB) . | Percentage of Total Subscriptions Using 4G . | Percentage of Total Subscriptions Using 5G . | 4G Avg. Demand/Subscription/Month (GB) . | 5G Avg. Demand/Subscription/Month (GB) . | 4G Simultaneous Emitter Rate . | 5G Simultaneous Emitter Rate . |
---|---|---|---|---|---|---|---|---|
2019 | 27 | 0.3 | 0.99 | 0.01 | 6.87 | 6.25 | 0.02 | 0.02 |
2020 | 33 | 3.1 | 0.9125 | 0.0875 | 8.86 | 8.79 | 0.02 | 0.021 |
2021 | 35 | 11.7 | 0.825 | 0.175 | 10.47 | 16.42 | 0.02 | 0.022 |
2022 | 38 | 23.5 | 0.7375 | 0.2625 | 12.52 | 21.87 | 0.02 | 0.023 |
2023 | 41 | 39.5 | 0.65 | 0.35 | 15.19 | 27.46 | 0.02 | 0.024 |
2024 | 44 | 61.3 | 0.5625 | 0.4375 | 18.78 | 33.82 | 0.02 | 0.025 |
2025 | 44 | 93.7 | 0.5 | 0.5 | 21.06 | 44.94 | 0.02 | 0.026 |
2026 | 44 | 136.2 | 0.45 | 0.55 | 23.33 | 59.00 | 0.02 | 0.027 |
2027 | 44 | 191.9 | 0.4 | 0.6 | 26.17 | 75.73 | 0.02 | 0.028 |
2028 | 44 | 264.9 | 0.35 | 0.65 | 29.83 | 95.91 | 0.02 | 0.029 |
2029 | 45 | 360.7 | 0.3 | 0.7 | 34.71 | 120.50 | 0.02 | 0.03 |
2030 | 43 | 488.0 | 0.25 | 0.75 | 39.79 | 151.28 | 0.02 | 0.031 |
Wireless NIM
The wireless NIM, illustrated in Figure 2, was developed in Analytica for its ability to quickly demonstrate visual dependencies and add model dimensionality without losing clarity in formulas. The NIM is designed to answer the following questions for the CGE model:
How many towers are needed to provide services on 4G networks and on 5G networks?
What will be the future Facilities & Equipment (F&E) and Operations & Maintenance (O&M) costs?
What will be the relative unit costs of providing wireless services using 5G networks compared to 4G networks ()?
The NIM estimates the optimal number of towers needed to meet exogenous demand forecasts. In general, fewer towers are needed as more spectrum is made available.
To estimate system demand, the model requires several pieces of information. First, the NIM uses the aggregate data demand projection described earlier as an exogenous input. How and where that data will be demanded within any location is assumed to be independent of specific 5G use cases (i.e., Internet of Things, Machine to Machine, and Enhanced Mobile Broadband [EMBB]). The model assumes that these characteristics are similar between 4G and 5G, and that most of the data will move over to EMBB.
The projected counts of 4G and 5G subscriptions are then distributed across national carriers and across three different types of geography that vary by population density: Dense Urban Area, Urban/Suburban Area, and Rural Area—see Table 6. This yields the subscription density per national carrier per square kilometer (km2) for each geographic area. Final network demand (megabit/second [Mbit/sec] per km2) is calculated by multiplying the number of simultaneously transmitting users by their average demand. Once the final network demand is established, the necessary network size is determined.
Geographical Divisions within the NIM
. | Dense Urban . | Urban/Suburban . | Rural . |
---|---|---|---|
Definition | Greater than 7,000 pop/km2 | Less than 7,000 pop/km2 and greater than 400 pop/km2 | Less than 400 pop/km2 and greater than 1 pop/km2 |
Area Coverage | 754 km2 | 151,962 km2 | 6,503,685 km2 |
Average Pop/km2 | 13,922 | 927 | 32 |
. | Dense Urban . | Urban/Suburban . | Rural . |
---|---|---|---|
Definition | Greater than 7,000 pop/km2 | Less than 7,000 pop/km2 and greater than 400 pop/km2 | Less than 400 pop/km2 and greater than 1 pop/km2 |
Area Coverage | 754 km2 | 151,962 km2 | 6,503,685 km2 |
Average Pop/km2 | 13,922 | 927 | 32 |
To split the United States into geographic areas, the research team used a dense urban threshold of 7,000 pop/km2, which encompasses five counties covering New York City and San Francisco. The urban/suburban threshold was set to 400 pop/km2 which covers 167 counties, most major metropolitan areas, and roughly one-third of the US population. The final threshold of 1 pop/km2 was selected based on the Verizon 4G network coverage of 2.68 million square miles (6,940,000 km2).17 Using the 1 pop/km2 threshold, the NIM total network coverage area is 6,656,400 km2 and approximately to the Verizon coverage. With the US covering 9.8 million km2, the NIM represents a 67% geographic coverage of the United States.
Coverage within each geographic area could be handled by three types of cells—macro, micro, and pico—the optimal mix is primarily determined by spectrum availability. This approach is informed by the EBU Technical Review - Cost Analysis of Orchestrated 5G Networks for Broadcasting.18 The radius of each cell varies across three different geographic topologies and three cell sizes. Table 7 lists assumptions used to define macro, micro, and pico cells.
Cell Assumptions
. | Definition . | Dense Urban Cell Radius (km) . | Urban/Suburban Cell Radius (km) . | Rural Cell Radius (km) . |
---|---|---|---|---|
Macro Cell | Suited for spectrum up to 1,800 MHz | 3 | 7 | 18 |
Micro Cell | Suited for spectrum between 1,801 and 10,000 MHz | 0.5 | 1.5 | 4 |
Pico Cell | Suited for spectrum higher than 10,000 MHz | 0.07 | 0.22 | 0.59 |
. | Definition . | Dense Urban Cell Radius (km) . | Urban/Suburban Cell Radius (km) . | Rural Cell Radius (km) . |
---|---|---|---|---|
Macro Cell | Suited for spectrum up to 1,800 MHz | 3 | 7 | 18 |
Micro Cell | Suited for spectrum between 1,801 and 10,000 MHz | 0.5 | 1.5 | 4 |
Pico Cell | Suited for spectrum higher than 10,000 MHz | 0.07 | 0.22 | 0.59 |
The NIM model also assumes that spectrum usage is split evenly between uplink and downlink, but the new spectrum (i.e., the spectrum shocks) is solely allocated to downlink.19Table 8 summarizes spectrum channelization and throughput assumptions.
Spectrum Channelization and Throughput
Cell . | Channelization20 . | 4G Throughput21 . | 5G Throughput(Spectral Efficiency: 5G-NR and 4G-LTE 2018) . |
---|---|---|---|
Macro Cell | 10 MHz/Channel | Macro 1.7 Mbps/Channel | Macro 2.55 Mbps/Channel |
Micro Cell | 20 MHz/Channel | Micro 2.5 Mbps/Channel | Micro 3.75 Mbps/Channel |
Pico Cell | 100 MHz/Channel | Pico 6.3 Mbps/Channel | Pico 9.45 Mbps/Channel |
Cell . | Channelization20 . | 4G Throughput21 . | 5G Throughput(Spectral Efficiency: 5G-NR and 4G-LTE 2018) . |
---|---|---|---|
Macro Cell | 10 MHz/Channel | Macro 1.7 Mbps/Channel | Macro 2.55 Mbps/Channel |
Micro Cell | 20 MHz/Channel | Micro 2.5 Mbps/Channel | Micro 3.75 Mbps/Channel |
Pico Cell | 100 MHz/Channel | Pico 6.3 Mbps/Channel | Pico 9.45 Mbps/Channel |
Using these factors, the model determines the number of each type of cell that would be required to meet the anticipated demand. The model builds the network starting with macro cells, then checks if any traffic remains unmet. If there is excess demand, the model assumes that micro cells will be utilized next. If there is excess demand on the 4G network, then the NIM densifies 4G’s micro cell network away from the micro cell’s optimal coverage size. If excess demand exists on the 5G network, then the model increases the number of pico cells until demand is met.
Table 9 provides a snapshot of the number of towers projected by the model for the years 2022 and 2030.
Number of Projected Towers by Type in the Baseline Case
Projection Year . | Cell Type . | 4G . | 5G . | Total . |
---|---|---|---|---|
2022 | Macro | 35,118 | 35,118 | |
Micro | 157,915 | 98,885 | ||
Pico | - | - | ||
Total | 193,033 | 134,003 | 327,036 | |
2030 | Macro | 35,118 | 35,118 | |
Micro | 185,673 | 730,685 | ||
Pico | - | 1,668,685 | ||
Total | 220,791 | 2,434,488 | 2,655,279 |
Projection Year . | Cell Type . | 4G . | 5G . | Total . |
---|---|---|---|---|
2022 | Macro | 35,118 | 35,118 | |
Micro | 157,915 | 98,885 | ||
Pico | - | - | ||
Total | 193,033 | 134,003 | 327,036 | |
2030 | Macro | 35,118 | 35,118 | |
Micro | 185,673 | 730,685 | ||
Pico | - | 1,668,685 | ||
Total | 220,791 | 2,434,488 | 2,655,279 |
The team compared the NIM estimate of 327,036 sites to a Statista.com report: “In 2020, there were 417,215 mobile wireless cell sites in the United States, with a large amount of investment going toward 5G-ready cell sites and antennas as per the source. There were 395,562 mobile wireless cell sites in the United States in 2019.”22 The discrepancies are reasonable since the NIM does not over-densify (which is common in urban areas), do not model areas with <1 pop/km2 which can include major cross-country roadways that are typically covered by wireless services, and is not budget limited which would certainly push tower construction earlier to meet the quantity needed by 2030.
Once the number of cells is determined, the total capital and annual operating costs for the network are calculated. Table 10 shows the cost per cell site, independent of the technology generation. Cost estimates for towers vary widely, due to location, height, locality, and many other factors. For a standalone 700 MHz public safety tower, an FCC cost model23 indicated $223K–$395K per new site in 2010$ which at 2% inflation, normalizes to $260K–$450K in 2017$. Other sources indicate per tower installation costs of $350K,24 $175K for 4G,25 $100K–$350K with potential to escalate up to $1M based on 2015 costs,26 and $100K–$300K.27 An additional source indicates macro cells to typically cost $200K and estimates small cells to cost under $10K.28 The 2012 Small Cells World Summit estimates that small cells will cost 1/10th the cost of a macro cell.29 A T-Mobile TV advertisement (unrecorded) indicated a rollout and investment yielding $35K per 5G tower. Asserting that macro towers, used in 2G and 3G rollouts generally face larger costs due to backhaul and other capex costs and will likely continue to do so, the NIM uses a high-end estimate of $500K per tower for the macro, an average rate of $200K per tower for micro (low-end macro and average 4G), and $35K for pico cells based on using 1/10th the frequent $350K tower cost citation and the T-Mobile advertisement.
Spectrum Cell Tower Cost
Cell . | 4G and 5G Tower Cost (2017$K) . | 4G Tower Annual Operations Cost (2017$K) (15% shown) . | 5G Tower Annual Operations Cost (2017$K) . |
---|---|---|---|
Macro Cell Tower | 500 | 75 | 27.75 |
Micro Cell Tower | 200 | 30 | 11.1 |
Pico Cell Tower | 35 | 5.25 | 1.943 |
Cell . | 4G and 5G Tower Cost (2017$K) . | 4G Tower Annual Operations Cost (2017$K) (15% shown) . | 5G Tower Annual Operations Cost (2017$K) . |
---|---|---|---|
Macro Cell Tower | 500 | 75 | 27.75 |
Micro Cell Tower | 200 | 30 | 11.1 |
Pico Cell Tower | 35 | 5.25 | 1.943 |
Tower Operation Costs are assumed to be a percentage range based on the initial cost and is estimated at 17%,30 ranging from 16.5% to 28% by state.31 Another source estimates a range of 12.8%–19%.32 The NIM assumed a simple uniform range of 15%–19%. OPEX cost reduction for 5G is estimated to range from 33%33 to 63% of 4G OPEX.34
Average construction costs of each cell type are multiplied by the number of towers to determine the tower-cost impact on capital stock needed to meet the future 4G and 5G networks. Non-tower capital costs are assumed to be allocated between 4G and 5G based on the number of subscribers. The NIM also assumes that the operational costs are proportional, that all telecommunication companies have a 7% cost of capital, and that a 7% depreciation rate applies to all telecommunication equipment. Given these assumptions, future telecommunication capital costs are projected for both the 4G and 5G networks.
Between 2019 and 2030, the NIM predicts that the 4G towers network will require approximately $20 billion of investment to grow the network to meet demand, but that an additional $43 billion is spent simply to offset depreciation. By contrast, the tower-related capital expenditures on the 5G network sums to $72 billion over 2019 and 2030, only $16 billion of which is due to financing annual depreciation. The capital stock related to non-tower equipment is held fixed at $315 billion but is allocated between 4G and 5G based on the number of subscribers.
Non-tower capital expenditures forecasts are developed by backing the annual 4G network capital costs for 2019 out of the annual capital costs for the wireless industry from the CGE model. These costs are then proportionally allocated between the 4G and 5G industries based on the percentage of subscribers for each network.35
Finally, note that the assumed wireless demand and spectrum allocation decisions are not affected by the infrastructure costs. In reality, firms will consider the expected impact of spectrum on their infrastructure costs before bidding on spectrum, and the demand for wireless services will be affected by the costs of providing that service. However, the first three steps in this framework are not intended to capture a market equilibrium. Instead, they are meant to approximate the relative costs of providing 4G and 5G services—and how marginal changes in spectrum assignment change those relative costs. As will be explained, those approximations are input into the CGE model, where the demand for wireless services follows the expected behavior.
Use Case Model
Because 5G is considered a general-purpose technology, it may produce secondary impacts to GDP due to productivity gains across industries. This is accounted for in one of two ways. The first method simply assumes that total factor productivity (TFP) across industries increases uniformly due to 5G. The second method, described in the Appendix, identifies specific sectors and use cases where 5G is expected to generate new benefits, then calculates a TPF estimate.
The Australian Bureau of Communications and Arts Research’s review of the literature finds that mobile technology and information and communication technologies (ICT) more generally have increased TFP, but that there is a wide range. They find that TFP estimates for broadband and ICT cases range between 0.01% and 0.75% depending on the technology and method of analysis.36 Because the higher values in this range refer to definitions of technology that are much broader than 5G (e.g., all information technology, broadband), the team selected three lower values in that range, 0.06%, 0.15%, and 0.28% to represent low, medium, and high productivity scenarios for 5G. In each scenario, the TFP assumption is applied uniformly across all industries in the CGE model to represent 5G’s potential impact.
CGE Model
CGE models are structural representations of an entire economy. The system of equations that define these models are derived from core economic theory—such as the first-order conditions resulting from profit-maximizing firms and utility-maximizing households. There is a finite supply to a country’s productive resources, such as its labor, capital stock, and natural resource endowments. The owners of these resources earn returns (e.g., wages) that are typically determined in competitive markets. As with all markets represented in these models, the price for each productive resource adjusts until the supply of that resource equals its demand.
CGE models are also data intensive. They integrate the national (or regional) economic accounts with details on the inputs needed and output generated in each industry. Behavioral parameters within the model are usually based largely on empirical econometric estimates found in the CGE literature, as well as model calibration.
One of the practical advantages of using CGE models is that the structure of the model assesses the ramifications of more direct and measurable policy changes. This is especially useful when data does not exist or when there is a high degree of uncertainty around important features of a policy. In the former case, statistical or econometric approaches are impossible, and in the latter case, the structure enables sensitivity testing to assess the importance of the assumptions. Because the transition to 5G is in an uncertain, early stage, CGE is the tool that can address these practical limitations. Another practical advantage of CGE models is that they can be tailored to focus on research areas.37
While different CGE models are available to researchers, this study uses the commercially available USAGE model to estimate the economic impact of spectrum and 5G. Developed by researchers at Victoria University, the USAGE model is a 392-sector, recursive dynamic CGE model of the US economy. It has been used to analyze policies for numerous federal agencies, including the Federal Aviation Administration, the International Trade Commission, the US Department of Agriculture, Economic Research Service, and others. Most of its underlying data comes from the 2012 benchmark accounts of the Bureau of Economic Analysis. Recent documentation of the USAGE model is readily available online through Victoria University’s Center of Policy Studies’ website,38 including detailed technical assumptions of the theoretical construction of the USAGE model.39
While a detailed discussion of the technical assumptions behind the USAGE model is available,40 the scope of that exercise is far too large for this report. Instead, this section will highlight some of the most important technical features of the model. First, USAGE assumes that production in every industry follows a nested, constant elasticity of substitution (CES) production function, which assumes constant returns to scale. The top level of the nest separates the value-added components (capital, land, and labor) from the intermediate goods, and is the only level to assume zero elasticity between the components (fixed proportions). All other nests at or below the second level have nonzero elasticities. Second, the model assumes firms are homogenous and markets are perfectly competitive. This means that the price in the market equals every firm’s marginal cost, and technological gains that lower costs are passed on to consumers. A third important characteristic is forecasted changes to the technology of production and consumer tastes adjust smoothly over time in response to movements in relative prices, unless the model is specifically targeted otherwise.
All three of the above characteristics are important when considering the application of the USAGE model to the future of wireless services. While the simplifications give straightforward derivations of optimal economic behavior, they also push the model away from reality that should be considered. First, the large investments in providing new wireless technologies likely mean that the telecommunications companies are facing increased returns to scale. But by letting 4G and 5G’s impact on firm costs and data utilization be distinct inputs to the model, rather than as endogenous outcomes of the model, a wide range of plausible scenarios can be estimated. In particular, the increasing returns to scale implied by the NIM becomes the targeted cost efficiencies in the CGE.
The fact that all firms in USAGE operate under perfect competition is another departure from the reality of oligopolistic competition within the telecommunications sector. One implication is that cost efficiencies are completely passed on to consumers in perfect competition, while oligopolies would likely increase profits by preventing prices from falling as much as in perfect competition. The less that telecommunications companies pass these benefits on to consumers, the smaller the overall economy benefits. The implication is that the benefits calculated by USAGE due to reducing telecommunications costs should be considered as taking place under ideal conditions.
The third highlighted characteristic of the model—that there are no sudden shocks to business input demands or household purchases, also seems at odds with the promise of 5G technologies. Indeed, USAGE does not predict that 5G will be revolutionary in specific parts of the economy, thereby sharply changing input composition of household and business inputs. However, the model is flexible enough to model those changes exogenously, so it is up to the researcher to determine those changes to model. Although there is a wide degree of uncertainty for such disruptive effects, the use case scenarios offer one possible example. Alternative industry forecasts could also be incorporated into this framework.
In this report, the USAGE model is extended to incorporate a new 5G wireless services industry. When a new industry is created within a CGE model, the industry of interest is usually aggregated into a parent industry, and the modeler needs to distinguish the myriad of ways in which the industry of interest differs from the parent industry. This includes the compositions of intermediate inputs, factors of production, investments, and the destination of sales, in addition to key parameters such as capital depreciation, wage rates, elasticities of substitution and transformation, household preferences, and so on. Unfortunately, such details are unknown for the 5G sector and are extremely difficult to forecast.
The solution to these data limitations is to treat the new 5G industry as nearly identical to the existing wireless industry. The major difference is that the new 5G industry sells its services to the existing wireless industry, which in turn resells those services to its customers. Under this framework, it is more appropriate to call the existing wireless industry a composite industry. That is because it can still create its own wireless services as it has always done by using 4G technology. This means it will continue to generate output directly by hiring labor, renting capital, purchasing other intermediate inputs, and so on. However, it may also increase its sales simply by making purchases directly from the 5G industry.
Figure 3 depicts the costs of producing wireless services in two future states. The left part of the figure reflects a future that does not utilize 5G technology. The composite industry makes no purchase of the negligibly small 5G industry and therefore only produces wireless services using 4G technology. The right part of the figure shows an alternate future, in which the 5G industry sells practically all output to the composite industry. The composite industry still directly utilizes labor, capital, and other intermediate inputs to generate output using legacy 4G technology, but the purchase of the 5G industry enables it to resell those wireless services to its customers.
Cost Decomposition of the Composite and 5G Industries in Two Alternative Future States.
Cost Decomposition of the Composite and 5G Industries in Two Alternative Future States.
To implement this change within USAGE, there are at least two critical assumptions to make. The first assumption is how much more efficient the wireless services industry is in the 4G+5G future compared to the future with just 4G technology. This is calculated as the percent change in the average unit costs between the baseline and policy simulations.41
where represents output generated by industry , defines the cost of producing , and j denotes either the 4G or 5G industry.42
The second assumption details how much of the composite industry is made up of 5G technology. This can be expressed as:
For notation’s sake, the baseline assumption is that the composite industry produces wireless service only through utilizing legacy technologies, denoted as , and pays an average unit cost of . In the policy simulation, the composite industry utilizes both 4G and 5G technologies to produce total output equal to
As a practical matter, and cannot be directly incorporated into USAGE without adding more structure. Conditions must change within the model to explain why the composite industry purchases more intermediate inputs from the 5G industry and reduces the inputs purchased to make wireless services using 4G technology. This structure is provided by a relationship between two technical efficiency parameters and an equation that explicitly targets the composite industry’s cost share attributed to the 5G purchase. Specifically, the USAGE model operationalizes these assumptions with essentially the following two equations:43
where is the percent change in technical efficiency across all possible inputs of the composite industry. (See technical appendix for details regarding technical change variables.) is the amount of input purchased by the composite industry, and is the price that is paid for it.44 reflects the percent change in the purchase of 5G wireless services as an input to the composite industry, and reflects the percent change in the price the composite industry pays to obtain it. is the percent change in the technical efficiency of 5G wireless services purchased as an intermediate input of the composite industry.
When is activated in the model, the first equation instructs USAGE how to change the technical efficiency of the composite industry’s inputs to reach the required change in overall efficiency. It does this through a combination of changing the efficiency across all existing inputs and a specific change in the technical efficiency of the 5G services purchased as an intermediate input.
When is activated, the second equation indicates that the change in the cost share of the composite industry that is attributed to 5G is completely determined by the amount that the composite industry spends on 5G wireless service. This equation just equals 5G’s original, near-zero share of the composite industry’s costs multiplied by the growth in the value of the industry’s 5G purchases.
In the baseline simulation, and are both endogenous variables, and the two technical efficiency parameters are treated as exogenous. Thus, the baseline simulation continues to expand the existing wireless services industry without any improvement in efficiencies from the 5G industry. In the policy simulation, this closure rule is reversed. and are both exogenously shocked, and the two technical coefficients adjust endogenously within the model to accommodate those assumptions. Thus, the policy simulation shows a future state where 5G becomes a core ingredient in delivering wireless service to customers.
Of course, the magnitude of both assumptions must be determined to activate and . To determine these magnitudes, note that each definition can be divided by in the numerator and denominator without causing a material change. Doing so recasts both assumptions in terms of two alternate variables. Specifically, and can be reexpressed as functions of the share of wireless services delivered via 5G technology () and the ratio of 5G unit costs relative to 4G unit costs ().
For operational purposes, both variables are taken from the NIM’s forecast. is set to equal the forecasted share of annual EBs that would be serviced on 5G networks. is set to equal the ratio of the average total costs per EB of data expected to be serviced in the future. Once obtained, these are converted into shocks for and . Thus, changes in the spectrum allocations affect the NIM, which converts them into CGE model inputs.
Note that varying and makes no claim about the demand-side benefits of 5G technologies. However, if assumptions are made regarding these benefits, CGE can directly incorporate them by changing other model variables. For example, if 5G is believed to help businesses improve their productivity, those concerns can be addressed by altering different technical efficiency parameters in other industries.
Findings
The results suggest that by making data cheaper and more ubiquitous, 5G could increase US real GDP between $347 billion and $536 billion as of 2030,45 depending on the relative price of 4G to 5G and the amount of data carried over 5G. While additional marginal spectrum for 5G does have some initial impact on network cost in 2025, it has a negligible impact by 2030. This is because the additional spectrum does little to delay the densification of the network based on the anticipated demand profiles. The total number of towers to satisfy forecasted demand remains the roughly same, though additional spectrum may reduce the rate at which the towers are constructed. The full macro and micro networks must both be fully developed by 2030 with cheaper pico cells filling in the additional demand.
Making Data Cheaper and More Ubiquitous
Figure 4 shows the change in real GDP that results from changing the NIM’s output. Different assumptions about 5G’s relative unit cost as compared to 4G are shown on the vertical axis, whereas the horizontal axis varies the relative proportion of wireless data provided by 5G as compared to 4G.
Sensitivity of 5G Scenario Assumptions: Change in Real GDP by 2030 ($ billions, 2019 prices).
Sensitivity of 5G Scenario Assumptions: Change in Real GDP by 2030 ($ billions, 2019 prices).
Intuitively, if 5G becomes cheaper or more widespread, the impact on GDP increases. In the most extreme version presented in the figure, it is 20x cheaper to provide one unit of wireless services using 5G compared to 4G and over 98% of wireless services use 5G. In this simulation, 5G increases real GDP by $751 billion in 2030 relative to the alternative. On the other hand, if the NIM forecasts low adoption rates and high costs of delivering 5G, then the impact on GDP is less. For example, if 5G only cuts the unit cost of delivering wireless services in half (Cost=0.5) and reaches an equal share of wireless services relative to 4G (Data=0.5), then gain to the US economy is substantially less—real GDP increases by only $103 billion from the baseline in 2030.
Under the assumption that 5G is a general-purpose technology, it is likely that all industries in the economy will become more efficient. When the three generalized TFP assumptions enter into the USAGE model, real GDP is between $35 billion and $165 billion higher in 2030. The results presented in this section seem reasonable in both direction and magnitude, and not far outside the range of findings from the literature.
Impact of Additional Spectrum
The NIM described above evaluates the impact of the following scenarios:
Baseline 4G/5G network based on anticipated spectrum pools
The addition of spectrum in 2025 that is suited for:
- –
Macro-sized cells (i.e., +50 MHz spectrum up to 1,800 MHz)
- –
Micro-sized cells (i.e., +100 MHz spectrum 1,801–10,000 MHz)
- –
Pico-sized cells (i.e., +1,000 spectrum >10,000 MHz)
- –
The removal of 5G spectrum from the baseline suited to micro-sized cells (i.e., less spectrum 1,801–10,000 MHz). This reflects the “what if” case if the recent C-band auction had not transpired.
The results of these shocks on 5G annual capital expenditures produce two interesting findings. The first is that while the addition of spectrum does have some initial impact on network cost in 2025 (particularly the addition of spectrum well suited for macro cells), it has a negligible impact over the long term (2030) given the current 5G demand assumptions.46 This is because while the additional spectrum can delay the densification rate of rural areas based on the anticipated demand profiles in every evaluated case, a full network of optimally sized macro and micro cells was always built out by 2030. The small differences in the capital costs in 2030 are based on the marginal changes in the number of pico cells that needed to be added to fully meet the anticipated demand. However, the delay in incurring these capital costs could result in meaningful cash flow implications for mobile network providers. If the additional spectrum changes how providers position their service offerings such that the 5G uptake rates or the cost of providing services changes,47 then the marginal impacts may differ from the baseline assumptions. Figure 4 demonstrates how changes to these assumptions could affect the economic impact.
The other important finding is that the recent C-band auction has a major impact on the anticipated capital costs of the future 5G network. Without it, the capital expenditures would be higher than the baseline to meet the same level of anticipated demand.
Testing and Sensitivity Analysis
Figure 5 shows the results of similar research. The estimated impact on US GDP due to 5G from eight other studies is plotted. Though the comparison is imperfect, the team’s assessment, with a reasonable set of assumptions, fits within the range of other publicly available estimates.48
The economic impact obtained from the Spectrum Macroeconomic Model depends on many factors, including the demand forecasts, spectrum allocation, NIM assumptions and output, the CGE model’s equations, the counterfactual future scenario, parameter values, choices of functional forms, and so on. Variances in any of these assumptions create different baseline values for and or other model behavior that would change the relative impact of spectrum and 5G. Testing all possible choices is impractical, but a focused approach can help identify key assumptions and factors affecting the results. This section shows the results of sensitivity analyses conducted over the following factors:
Counterfactual Costs of Utilizing Only 4G Technology
CGE Model Parameters
Counterfactual Costs of Utilizing Only 4G Technology
So far, all the simulations use the same baseline future, in which wireless services are only provided using 4G. Changes in this counterfactual future will also have large ramifications in calculating the benefits of 5G. Intuitively, if the costs of providing wireless services continue to drop, then the benefits of using 5G will increase. In the baseline 4G-only future, the assumption is that the unit cost of wireless services grows approximately 0.49% higher than the rate of inflation each year. By 2030, the real unit cost of using 4G technology is 5.5% higher than it was in 2019 in this simulation. This baseline assumption is the result of the CGE model being projected into the future based on the benchmarked structure of the US economy dynamically changing to accommodate macroeconomic forecasts of GDP’s expenditure components, inflation, working-age population, hours worked, and other aggregate factors. It is not the result of an explicit assumption about the costs of a 4G-only future.
To test the sensitivity of this assumption, four other alternative 4G-only counterfactuals are assumed, but 5G’s relative advantages over 4G are maintained. In each simulation, the counterfactual industry’s cost grows at a different rate due to changing efficiencies of providing 4G service. The first two scenarios assume that gradual technological improvements in 4G reduce the real unit costs of providing wireless service by 2.9% and 7.0% each year. In the other two scenarios, the 4G industry experiences expanding costs such that it must significantly raise the prices of its services.49 Specifically, it is assumed that the real unit cost of offering wireless services rises 6.1% and 11.0% each year instead of 0.5% as assumed in the original 4G case.
As intuition suggests, the introduction of 5G has a greater economic impact when the cost of delivering 4G services becomes more expensive. In the two increasing cost scenarios, demand for wireless services falls between 2019 and 2030. The leap to 5G technologies is more important in these counterfactuals, as evidenced by the advent of 5G increasing real GDP substantially above the baseline policy. On the other hand, if 4G becomes cheaper in the future, then 5G’s impact on GDP falls relative to the baseline.
Default CGE Parameters
In practice, policy simulations that utilize CGE models often assume that economic behavior can be described by the model’s default parameterization. Because changes to these parameters could potentially result in materially different conclusions, it is prudent to perform sensitivity tests around parameters relevant to the policy of interest.50 This section shows the results of independently varying the parameters that determine the following four behavioral relationships.
Substitutability between intermediate inputs: wireless services industry
Substitutability between intermediate inputs: Top 20 industries utilizing wireless services51
Substitutability between capital and labor: wireless services industry
Substitutability between capital and other factors of production: wireless services industry
Table 11 shows the impact on GDP using the primary assumptions obtained from NIM for the reference scenario of $496 billion. This reference scenario is selected because sensitivities are more likely to be revealed with larger shocks. Each row shows the default parameter being tested, the default value and minimum and maximum of its assumed corresponding triangular distribution, and the distribution of the 5G’s impact on real GDP in 2030. The results suggest that 5G’s impact is not especially sensitive to the default values. The results are more sensitive to the elasticity of substitution across the 20 industries that have higher utilization of wireless services, but the implied impacts are still close to the reference simulation value of $496 billion.
Real GDP Impact Due to Primary 5G Assumptions: Sensitivity to Default CGE Parameters
. | Parameter Range . | Real GDP (%Δ and $Δ in Billions) . | |||||
---|---|---|---|---|---|---|---|
Parameter Tested . | Min . | Default . | Max . | Mean . | Std. Dev (%) . | 95% CI Lower Bound . | 95% CI Upper Bound . |
Elasticity of Substitution for Intermediate Inputs: Wireless Industry | 0 | 0.2 | 0.4 | 1.81% $496 | 0.003 $2.09 | 1.80% $492 | 1.83% $500 |
Elasticity of Substitution for Intermediate Inputs: Top-20 Wireless-Using Industries | 0.02 | 0.2 | 2 | 1.84% $504 | 0.027 $7.6 | 1.79% $489 | 1.89% $519 |
Elasticity of Substitution: Labor for Other Factors of Production | 0 | 0.5 | 1 | 1.81% $497 | 0.003 $2.1 | 1.80% $492 | 1.83% $501 |
Elasticity of Substitution: Capital for Other Factors of Production | 0 | 0.5 | 1 | 1.81% $496 | 0.002 $1.5 | 1.80% $493 | 1.82% $499 |
. | Parameter Range . | Real GDP (%Δ and $Δ in Billions) . | |||||
---|---|---|---|---|---|---|---|
Parameter Tested . | Min . | Default . | Max . | Mean . | Std. Dev (%) . | 95% CI Lower Bound . | 95% CI Upper Bound . |
Elasticity of Substitution for Intermediate Inputs: Wireless Industry | 0 | 0.2 | 0.4 | 1.81% $496 | 0.003 $2.09 | 1.80% $492 | 1.83% $500 |
Elasticity of Substitution for Intermediate Inputs: Top-20 Wireless-Using Industries | 0.02 | 0.2 | 2 | 1.84% $504 | 0.027 $7.6 | 1.79% $489 | 1.89% $519 |
Elasticity of Substitution: Labor for Other Factors of Production | 0 | 0.5 | 1 | 1.81% $497 | 0.003 $2.1 | 1.80% $492 | 1.83% $501 |
Elasticity of Substitution: Capital for Other Factors of Production | 0 | 0.5 | 1 | 1.81% $496 | 0.002 $1.5 | 1.80% $493 | 1.82% $499 |
Discussion
While this modeling approach is a first iteration and can be matured, we believe it can constructively contribute to important spectrum decisions at this stage. In future iterations, for example, the demand model could be shifted from the current user-centric model to a device-centered model. This would enable the model to explore how device deployment and corresponding demand grow across the various 5G services. In addition, the demand model could be expanded to include a mechanism that would allow demand to respond more dynamically to the forecasted price. Finally, the model could explore multiple possible demand futures. For example, what if there is a flattening of 5G demand if some 5G EMBB services (e.g., virtual reality, augmented reality) do not take hold, or what if there may be a rapid shift away from 4G as legacy devices are phased out?
The infrastructure model could be improved in several ways to make it more granular and realistic. For example:
Adding additional geographic morphologies by splitting Urban and Suburban and adding highways as their own geographic areas
Rebuilding the cell sizes and throughput basis by individual spectrum bands
Differentiating throughput vs. goodput
Exploring the effects of the simultaneous emitter rate assumptions
Accounting for frequency reuse
Assuming the 5G network will build out over the 4G network
Assuming 5G will build from high to low population density
Extending the modeled time horizon to 2040 and including 6G network buildout requirements
Exploring cost questions such as if 5G produces higher OPEX costs due to increased backhaul requirements
Greater fidelity with construction cost and OPEX cost distributions
Considering cost implications of sharing of towers between 4G and 5G
Considering tower construction implications if the amount of annual CAPEX spent on new tower construction per year is capped
Considering how per tower costs might change over time for 5G as the technologies reach scale
As discussed in the findings, 5G could increase US GDP between $347 billion and $536 billion in 2030 depending on the relative price of 4G to 5G and the amount of data carried over 5G. While this estimate should not be treated as definitive, the method outlined in this study is tailorable to easily incorporate changes to the underlying assumptions. As explored in this article, 5G’s impact on GDP could be higher when industry-specific assumptions on productivity raises are considered. But, on the other hand, these values might be lower if 4G/5G operational costs are higher than assumed or if per-cell channelization and throughput or the amount of future data demanded is lower than assumed. The recent C-band auction provided significant mid-band spectrum to the spectrum pool. Without this spectrum, the anticipated capital costs of the future 5G network would be higher to meet the same level of anticipated demand.
There are a few notable caveats to this study’s conclusions. First, the impact on GDP refers to the difference between two possible futures, one with 5G technology and one without 5G. It is not measuring the difference in GDP today (i.e., 2021) versus GDP in the future (i.e., 2030). In addition, the assumptions that drive the 4G-only future will influence 5G’s impact, as discussed in the testing and sensitivity analysis section above.
Another caveat is that the costs of 5G were primarily limited to the costs of building the network infrastructure. Other costs affected by 5G, such as investment by manufacturers, do not materially change to benefit from 5G services. Moreover, 5G does not fundamentally change the production technology of other industries.
In addition, the model assumes that 5G users receive benefits from 5G by being able to purchase wireless services at a lower price. The response to cheaper wireless services is determined by the model’s characterization of firm production and household preferences—both represented by CES functions. Of course, 5G is expected to bring about fundamentally new uses of wireless services, so household and firm demand could depart significantly from the model’s CES assumption.
Finally, this study defines the quantity of wireless services as the amount of data being provided by either 4G or 5G technologies. While other differences exist between these two technologies beyond data rates, having this common unit of measure is necessary to calculate relative unit costs and to construct the industry shares. The study does not attempt to make one bit of data delivered on a 5G system qualitatively different than one delivered on a 4G system.
As spectrum access for nonfederal users continues to grow, spectrum access for federal users is similarly growing to support increased data and information requirements, as well as more complex operations. At the same time, federal agencies are working to access spectrum more efficiently and effectively. The United States has made significant licensed and unlicensed spectrum available for nonfederal wireless services, including 5G.
However, there is a tipping point where continued repurposing of federal spectrum negatively impacts the ability of federal agencies to meet their mission requirements. The ability to consider the economic impact on critical missions due to spectrum repurposing is vital. The Spectrum Relocation Fund (SRF) provides a mechanism, funded from auction proceeds, for federal agencies to be reimbursed for some of their costs to repurpose spectrum from federal agency to commercial use, with the remaining funds deposited in a general fund of the Treasury. However, costs reimbursable by the SRF do not represent the full costs or economic impact on federal missions.
The team is developing an adjunct framework to account more completely for the full economic impacts of spectrum repurposing. This framework includes a comprehensive list of risk and cost elements.
APPENDIX
CGE Technical Efficiency Parameters
Technical efficiency parameters are a common way of implementing policies within CGE models. This appendix describes technical efficiency parameters and how they are utilized in Assumptions 1 and 2. It is useful to start with a production function that converts an industry’s inputs into output. In the USAGE model, this is:
where represents the output generated in industry , is the aggregate volume of intermediate inputs that produce in industry , and is the aggregate volume of the primary factor—a composite of the factor income. The parameters , , and are strictly positive technical efficiency parameters—typically treated as exogenous. Note that when the technical efficiency parameters fall, fewer inputs are needed to produce the same amount of output.52 The top-level production function, , is fixed proportions, which means there is no substitutability between the volume of intermediate inputs and the volume of the primary factor.53
The aggregate volume of intermediate inputs and of the primary factor is also determined by separate production functions shown below:
is the quantity of intermediate input used to produce commodity , and is its corresponding technical efficiency parameter. Similarly, is determined by the amount of labor (), capital (), and land () used by industry , along with their corresponding efficiency parameters. The two second-tier production functions, and , use a CES functional form.54 The third and final tier of production functions also apply CES functional forms to determine the share of imported content in each , though such details are not relevant to this discussion. All such production functions are homogenous of degree 1, which implies that they each operate with constant returns to scale.
By sequentially finding the cost-minimizing solution at each tier of production using standard Lagrangian techniques, the corresponding set of first-order conditions yield the formulas for the demand for intermediate inputs. Expressed in terms of percent changes,55 this is:
where the lower-case variables reflect the percent change of the corresponding upper-case variable ( is the percent change for , is the percent change in , etc.). In addition, is the assumed elasticity of substitution, is the percent change in the price input (including intermediate and factors of production), and is the share of input in the total cost of production. Similar equations hold for other inputs that are allocated according to CES assumptions, including factors of production, imported vs. domestic shares, composition of investment, and so on. The technical efficiency parameters in these equations differ, but certain efficiency parameters, such as , are common to all input demand equations for an industry.
Note that by decreasing the technical efficiency parameter , the demand for all intermediate inputs and the factors of production in the composite industry share the common effect of .56 However, the only directly affects the composite industry’s intermediate demand for 5G services, . creates a relationship between these two technical efficiency parameters, whereby the change in the unit costs of the composite industry comes from its reduced demand across all non-5G inputs, provided it purchases enough 5G as an intermediate input to satisfy .
As an example, suppose that the NIM implies that the composite industry’s unit cost falls by 50% and that 5G constitutes 80% of the composite wireless industry’s total costs. Furthermore, assume that the negligibly small share of the composite industry’s costs attributed to 5G in the baseline was 0.05%. Then USAGE would add the following two constraints into its numerical solution, with the two technical efficiency parameters being endogenous.
There are a few takeaways from the above equations. First, implies that the composite industry will substantially expand its nominal purchases of 5G, by 160,000% in this example. Since is endogenous, it will need to expand significantly in the solution to accommodate . This does not lead to an inefficiency in the composite industry, however, because imposes a negative relationship between and . This means that as the composite industry demands more 5G inputs, it also reduces its demand for all other inputs.57
Tailored Macroeconomic Impact of 5G
In addition to the generalized approach, a Rough Order of Magnitude (ROM) Use Case Model was developed to translate qualitative assessments of 5G use cases into TFP estimates. The predominant 5G use cases were researched from multiple literature reviews and aligned to the six industries that are expected to receive the clearest benefits from emerging 5G technologies and capabilities: healthcare, agriculture, real estate, trucking, food and accommodation, and utilities.
The ROM Use Case Model uses a straightforward method to calculate an industry TFP based on six input values: historic industry gross output, historic gross output growth rates to project future output size, the factor of production that is affected by the use case and its percentage contribution to gross output, an efficiency impact factor, a productivity impact factor, and a market penetration rate. The selection of the input values is based on subject matter expert (SME) judgment and interpretation of the individual use case descriptions. The inputs are multiplied together to estimate the magnitude of the cost savings enabled by the use case and divided by the future gross output to estimate the improvement in the TFP.
Gross output and growth rates: Historic gross output for an industry is available from the US Bureau of Economic Analysis (Bureau of Economic Analysis 2021). For the ROM assessments performed here, the 2019 data is used as the baseline. The growth rate from 2010 to 2019 is used to project the future gross output for 2025, 2030, and 2035.
Factor of production and associated rate: Per CGE assumptions, gross inputs equal gross outputs, so the production factors can be considered as a share of gross output. The rates are extracted from the CGE model for the factors: labor, capital, and intermediate.
Efficiency impact factor: This is an SME-selected value based on perceived magnitude of impact of the use case on the industry to which it pertains [Not Affected; Slightly Affected; Marginally Affected; Dramatically Affected]. These efficiency rates are tuned by comparing explicit use case savings calculations within the agricultural and healthcare industries with the ROM estimates produced by the Use Case Model.
The associated scale is as follows:
- a.
Not Affected—0
- b.
Slightly Affected—2.5%
- c.
Marginally Affected—5%
- d.
Dramatically Affected—8%
- a.
Productivity impact factor: This is a SME-selected value based on perceived magnitude of impact of the use case on each production factor. This factor is a set of ranges in 5% increments: 0–5%, 5%–10%, 10%–15%, and so on. The ranges are used to establish low/mid/high TFPs. These ranges are not labeled with qualitative assessments such as low, medium, or high due to the variability in use cases and the potential extent of their effects on factors of production (10 or more labels would be excessive to describe). Note that 0%–5% is a special case, and realistically, 1% is used as the low end in order to have a meaningful lower bound that is not 0.
Market penetration: This is an SME-estimated percentage that represents how much of the industry adopts the use case. It captures the effect that an industry will transition over to a new technology or capability over time, so benefits are pro-rated based on how well adopted the use case is within the industry. For the purposes of this analysis, use case penetration is assumed to grow in parallel with 5G’s market penetration: 50% by the year 2025, 75% by the year 2030, and 85% by the year 2035.
The 5G savings factor for each use case within its associated industry is estimated by using the industry gross output as the basis of the estimate. To illustrate how savings factors are derived, the telemedicine use case from the healthcare industry is used as an example. Healthcare total gross output in the year 2019 was $2.6 trillion. The average growth rate from 2010 to 2019 for healthcare is approximately 4.6%. This rate is used to project the healthcare gross output for the years 2025, 2030, and 2035, assuming that the gross output will grow at the same rate. In the year 2030, healthcare gross output is projected at $4.3 trillion.
It is assumed labor expenses will be most impacted by telemedicine. Labor expenses are 69% of the total healthcare cost. So, if 0%–5% factor productivity range is selected for telemedicine, 1% min with 5% max of 69% labor will be impacted for telemedicine. This factor is the portion of the total healthcare expenses (i.e., healthcare gross output) that can be applied to estimate the savings factor. At minimum, 1% of 69% of $4.6 trillion and at maximum, 5% of 69% of $4.6 trillion is the net labor expenses of telemedicine in the year 2030.
The labor savings are then estimated by selecting the magnitude of effect. If no effect is selected, obviously the savings rate will be zero. If slightly affected is selected, savings will be 2.5% of the net labor expenses of telemedicine. To account for technology adoption occurring over time, the market penetration assumption is applied to the net labor savings estimate. The ROM savings rate is projected by taking the telemedicine net labor savings estimate out of the total healthcare gross output. This approach is applied across all six industries to devise ROM projections of 5G savings rates or equivalent TFP improvements.
This qualitative approach yields slightly more optimistic projections than the generalized approach of selecting TFP values from a historical range. As an example, qualitative assessments in healthcare and agriculture are shown in Table 12. The ROM projections (“ROM”) largely fall within the range of the published studies from the generalized approach (“Study”).
Use Case Projections
Use Case . | Projected Year . | Method . | Sum of Net Industry Savings Rate Min (%) . | Sum of Net Industry Savings Rate Mid (%) . | Sum of Net Industry Savings Rate Max (%) . |
---|---|---|---|---|---|
Autonomous Farming Machine (Agriculture) | 2025 | ROM | 0.20 | 0.30 | 0.40 |
Study | - | 0.32 | - | ||
2030 | ROM | 0.44 | 0.67 | 0.89 | |
Study | - | 0.51 | - | ||
2035 | ROM | 0.47 | 0.71 | 0.95 | |
Study | - | 0.61 | - | ||
Building and Equipment Management (Agriculture) | 2025 | ROM | 0.16 | 0.25 | 0.33 |
Study | - | 0.29 | - | ||
2030 | ROM | 0.39 | 0.59 | 0.79 | |
Study | - | 0.46 | - | ||
2035 | ROM | 0.40 | 0.59 | 0.79 | |
Study | - | 0.56 | - | ||
Crop Monitoring (Agriculture) | 2025 | ROM | 0.40 | 0.60 | 0.80 |
Study | - | 0.76 | - | ||
2030 | ROM | 0.60 | 0.90 | 1.20 | |
Study | - | 1.21 | - | ||
2035 | ROM | 0.68 | 1.02 | 1.36 | |
Study | - | 1.45 | - | ||
Farming by Drone (Agriculture) | 2025 | ROM | 0.40 | 0.60 | 0.80 |
Study | - | 0.59 | - | ||
2030 | ROM | 0.60 | 0.90 | 1.20 | |
Study | - | 0.93 | - | ||
2035 | ROM | 0.68 | 1.02 | 1.36 | |
Study | - | 1.11 | - | ||
Livestock Monitoring (Agriculture) | 2025 | ROM | 0.33 | 0.50 | 0.66 |
Study | - | 0.47 | - | ||
2030 | ROM | 0.45 | 0.68 | 0.91 | |
Study | - | 0.74 | - | ||
2035 | ROM | 0.50 | 0.75 | 0.99 | |
Study | - | 0.89 | - | ||
Automatic Processing (Healthcare) | 2025 | ROM | 0.03 | 0.07 | 0.14 |
Study | - | 0.11 | - | ||
2030 | ROM | 0.04 | 0.10 | 0.21 | |
Study | - | 0.15 | - | ||
2035 | ROM | 0.05 | 0.12 | 0.23 | |
Study | - | 0.15 | - | ||
Connected Ambulance (Healthcare) | 2025 | ROM | 0.04 | 0.06 | 0.09 |
Study | - | 0.06 | - | ||
2030 | ROM | 0.04 | 0.10 | 0.21 | |
Study | - | 0.09 | - | ||
2035 | ROM | 0.05 | 0.12 | 0.23 | |
Study | - | 0.11 | - | ||
Remote Surgery (Healthcare) | 2025 | ROM | 0.01 | 0.02 | 0.03 |
Study | - | 0.02 | - | ||
2030 | ROM | 0.01 | 0.02 | 0.05 | |
Study | - | 0.02 | - | ||
2035 | ROM | 0.01 | 0.03 | 0.05 | |
Study | - | 0.02 | - | ||
Telemedicine—Hospitals (Healthcare) | 2025 | ROM | 0.01 | 0.02 | 0.03 |
Study | - | 0.01 | - | ||
2030 | ROM | 0.01 | 0.02 | 0.04 | |
Study | - | 0.02 | - | ||
2035 | ROM | 0.01 | 0.04 | 0.07 | |
Study | - | 0.03 | - |
Use Case . | Projected Year . | Method . | Sum of Net Industry Savings Rate Min (%) . | Sum of Net Industry Savings Rate Mid (%) . | Sum of Net Industry Savings Rate Max (%) . |
---|---|---|---|---|---|
Autonomous Farming Machine (Agriculture) | 2025 | ROM | 0.20 | 0.30 | 0.40 |
Study | - | 0.32 | - | ||
2030 | ROM | 0.44 | 0.67 | 0.89 | |
Study | - | 0.51 | - | ||
2035 | ROM | 0.47 | 0.71 | 0.95 | |
Study | - | 0.61 | - | ||
Building and Equipment Management (Agriculture) | 2025 | ROM | 0.16 | 0.25 | 0.33 |
Study | - | 0.29 | - | ||
2030 | ROM | 0.39 | 0.59 | 0.79 | |
Study | - | 0.46 | - | ||
2035 | ROM | 0.40 | 0.59 | 0.79 | |
Study | - | 0.56 | - | ||
Crop Monitoring (Agriculture) | 2025 | ROM | 0.40 | 0.60 | 0.80 |
Study | - | 0.76 | - | ||
2030 | ROM | 0.60 | 0.90 | 1.20 | |
Study | - | 1.21 | - | ||
2035 | ROM | 0.68 | 1.02 | 1.36 | |
Study | - | 1.45 | - | ||
Farming by Drone (Agriculture) | 2025 | ROM | 0.40 | 0.60 | 0.80 |
Study | - | 0.59 | - | ||
2030 | ROM | 0.60 | 0.90 | 1.20 | |
Study | - | 0.93 | - | ||
2035 | ROM | 0.68 | 1.02 | 1.36 | |
Study | - | 1.11 | - | ||
Livestock Monitoring (Agriculture) | 2025 | ROM | 0.33 | 0.50 | 0.66 |
Study | - | 0.47 | - | ||
2030 | ROM | 0.45 | 0.68 | 0.91 | |
Study | - | 0.74 | - | ||
2035 | ROM | 0.50 | 0.75 | 0.99 | |
Study | - | 0.89 | - | ||
Automatic Processing (Healthcare) | 2025 | ROM | 0.03 | 0.07 | 0.14 |
Study | - | 0.11 | - | ||
2030 | ROM | 0.04 | 0.10 | 0.21 | |
Study | - | 0.15 | - | ||
2035 | ROM | 0.05 | 0.12 | 0.23 | |
Study | - | 0.15 | - | ||
Connected Ambulance (Healthcare) | 2025 | ROM | 0.04 | 0.06 | 0.09 |
Study | - | 0.06 | - | ||
2030 | ROM | 0.04 | 0.10 | 0.21 | |
Study | - | 0.09 | - | ||
2035 | ROM | 0.05 | 0.12 | 0.23 | |
Study | - | 0.11 | - | ||
Remote Surgery (Healthcare) | 2025 | ROM | 0.01 | 0.02 | 0.03 |
Study | - | 0.02 | - | ||
2030 | ROM | 0.01 | 0.02 | 0.05 | |
Study | - | 0.02 | - | ||
2035 | ROM | 0.01 | 0.03 | 0.05 | |
Study | - | 0.02 | - | ||
Telemedicine—Hospitals (Healthcare) | 2025 | ROM | 0.01 | 0.02 | 0.03 |
Study | - | 0.01 | - | ||
2030 | ROM | 0.01 | 0.02 | 0.04 | |
Study | - | 0.02 | - | ||
2035 | ROM | 0.01 | 0.04 | 0.07 | |
Study | - | 0.03 | - |
Applying the above approach across all six industries, Table 13 shows the following results.
Industry Specific Savings by Year
Industry . | Projected Year . | Sum of Net Industry Savings Rate Min (%) . | Sum of Net Industry Savings Rate Mid (%) . | Sum of Net Industry Savings Rate Max (%) . |
---|---|---|---|---|
Accommodation and Food | 2025 | 0.12 | 0.19 | 0.25 |
2030 | 0.19 | 0.28 | 0.37 | |
2035 | 0.30 | 0.44 | 0.59 | |
Agriculture | 2025 | 1.49 | 2.24 | 2.98 |
2030 | 2.49 | 3.74 | 4.98 | |
2035 | 2.73 | 4.09 | 5.46 | |
Electric Power | 2025 | 0.09 | 0.13 | 0.17 |
2030 | 0.14 | 0.22 | 0.29 | |
2035 | 0.17 | 0.25 | 0.34 | |
Healthcare | 2025 | 0.09 | 0.17 | 0.28 |
2030 | 0.10 | 0.25 | 0.50 | |
2035 | 0.12 | 0.20 | 0.59 | |
Real Estate | 2025 | 0.0% | 0.08 | 0.10 |
2030 | 0.17 | 0.25 | 0.34 | |
2035 | 0.04 | 0.05 | 0.07 | |
Trucking | 2025 | 0.12 | 0.18 | 0.25 |
2030 | 0.25 | 0.37 | 0.49 | |
2035 | 0.38 | 0.57 | 0.76 |
Industry . | Projected Year . | Sum of Net Industry Savings Rate Min (%) . | Sum of Net Industry Savings Rate Mid (%) . | Sum of Net Industry Savings Rate Max (%) . |
---|---|---|---|---|
Accommodation and Food | 2025 | 0.12 | 0.19 | 0.25 |
2030 | 0.19 | 0.28 | 0.37 | |
2035 | 0.30 | 0.44 | 0.59 | |
Agriculture | 2025 | 1.49 | 2.24 | 2.98 |
2030 | 2.49 | 3.74 | 4.98 | |
2035 | 2.73 | 4.09 | 5.46 | |
Electric Power | 2025 | 0.09 | 0.13 | 0.17 |
2030 | 0.14 | 0.22 | 0.29 | |
2035 | 0.17 | 0.25 | 0.34 | |
Healthcare | 2025 | 0.09 | 0.17 | 0.28 |
2030 | 0.10 | 0.25 | 0.50 | |
2035 | 0.12 | 0.20 | 0.59 | |
Real Estate | 2025 | 0.0% | 0.08 | 0.10 |
2030 | 0.17 | 0.25 | 0.34 | |
2035 | 0.04 | 0.05 | 0.07 | |
Trucking | 2025 | 0.12 | 0.18 | 0.25 |
2030 | 0.25 | 0.37 | 0.49 | |
2035 | 0.38 | 0.57 | 0.76 |
With exception of the agriculture industry (i.e., 2.49%–4.98%), the results from the Use Case Model are generally within the range of the generalized TFP shocks (i.e., 0.06%–0.28%).
The distinction of the Use Case Model from the generalized TFP is that the use cases are industry specific, which intuitively seems more reasonable. It seems less likely that 5G technology will generate the same productivity benefits across all industries. Some industries are likely to benefit more from 5G than others, though many may receive a low general TFP boost due to general efficiencies. Replacing the generalized TFP shocks with the TFP shocks derived from the Use Case Model for specific industries may be a better representation of the productivity efficiency gains. Moreover, additional industries receiving 5G benefits, such as manufacturing and entertainment, could be explored the Use Case Model to better reflect the economic impact of 5G productivity efficiency gains. Going forward, the TFP shocks from the Use Case Model could replace the generalized TFPs currently used in the CGE model.
FOOTNOTES
The authors dedicate this article in memory of Dr. Zekarias Hussein, a thoughtful coauthor who provided valuable contributions to our team.
Whole-of-nation refers holistically to a country or state, including across various key stakeholder organizations, such as national regulators, federal agencies, industry, local governments, research institutions, and consumer associations.
Dixon and Rimmer, 2001; Dixon, Rimmer, and Waschik.
Tom’s Guide Staff.
Some mmWave spectrum was auctioned in 2004 but was purchased speculatively, because the technology was not yet available to broadly capitalize on this spectrum.
The demand forecasts are affected by transitioning the 4G spectrum to 5G purposes.
At the time of this research, data were only available through 2019. The 2021 Ericsson Mobility Report (Ericsson 2021) presents North American traffic rates for 2020 at 3.7 EB/month and for 2021 at 4.9 EB/month. Using an average offload rate of 67% (average difference between Cisco rates (Cisco 2019) and Ericsson rates (Ericsson 2019) for 2017 and 2018 [63%–72%]), and the 96% United States of North America rate, the expected US traffic for 2020 was 28.55 EB and for 2021 was 37.8 EB. The 30% jump in traffic aligns with the rollout and maturity of 5G in 2021 as well as potential increased mobile traffic due to COVID influences. The compound annual growth rate (CAGR) growth rate predictions (see footnote 10) of 35.7 EB in 2021 and 46.7 EB predicted higher than realized growth even though it was the lowest CAGR published to date.
Linear extrapolation between 2013 and 2015. Actual data not available.
31% growth rate applied to the 2018 traffic.
Offloading of mobile traffic is important to consider because a significant portion of total mobile device data and traffic is handled by alternative methods. To the best of our knowledge, Ericsson reports do not include offloading. Cisco non-offloaded demand is assumed to approximate the actual demand for wireless 4G/5G services.
Cisco n.d.; Cisco 2014; Cisco 2016; CISCO 2016; Cisco 2017; Cisco 2018; Cisco 2019; Cisco 2020; Pepper 2013.
2017 US EB/month is 1.2 EB (Cisco 2018). 2017 North American EB/month is 1.261 EB (Cisco 2019). US share is 96% of North America.
This analysis was performed in 2018–2019, therefore 2020 traffic and beyond are estimates.
This is the most pessimistic growth rate seen by the research team.
US Census, Table 1.
One important consideration is that wireless devices can and do offload traffic onto local Wi-Fi networks. Therefore, it is important to capture the portion of traffic from wireless devices that actually use 4G/5G networks. It is not known what portion of 5G device traffic will be offloaded in the future, nor is it known how much Wi-Fi traffic in the future might be routed via 5G fixed-wireless backhaul instead of physical fiber backhaul. Using 5G backhaul would add more load to the network. For simplicity, it is assumed that 5G will have offloading rates similar to 4G.
Verizon.
Lombardo.
While it is reasonable to assume that mobile network operators will continue to shift spectrum from 4G to 5G services over time, this channelization portion of the analysis assumes that the 4G and 5G spectrum pools are static. Therefore, results can be considered a “best case” for the value of additional spectrum for 5G applications.
Nokia.
Ibid.
Statista.
Federal Communications Commission, “A Broadband Network Cost Model.”
LaFuria.
Nixon.
Foster.
CellTowerInfo.com.
Goss.
The 3G4G Blog.
~17% from Annual Maintenance Capital + Total OPEX / Initial Investment (~$2B/~$12B)
CostQuest Associates, Inc.
LaFuria. The two numbers are calculated by summing the $4,662.44 monthly costs and dividing by the initial cell site CAPEX or by the initial cell site CAPEX plus the initial core CAPEX.
Gabriel.
Oughton and Frias.
As an illustration the nonnetwork capital cost pool is split 50/50 between 4G/5G in 2025 but is split 25/75 in 2030 as more subscribers move to 5G.
Department of Communications and the Arts, Australian Government, page 17.
For example, USAGE-Air is a CGE model that extends the USAGE model by incorporating a cost parameter designed to reflect costs of airspace parameters (GTAP Resources: Resource Display 2020).
The USAGE Model.
Ibid.
Note that this formula assumes that 4G’s unit cost is independent of the amount of 5G services provided, which one might expect under the assumption that all industries operate under constant returns to scale. However, repurposing spectrum from 4G to 5G to accommodate 5G’s growth likely affects the unit costs of 4G. Though this assumption of independence is therefore a simplification, this can be partially addressed by making alternative assumptions in the USAGE model about the growth in 4G’s future costs.
In other words, , where are the optimal inputs demanded to produce at input prices, .
The full equations in the USAGE model also make minor adjustments to account for the treatment of taxes and the sourcing of inputs from domestic or imported sources. Neither of these adjustments is material to the solution, and they are omitted from the equations above for clarity.
In keeping with the General Equilibrium Modelling Package (GEMPACK) convention, variables expressed in lower case typically represent percent changes in the corresponding uppercase variable. In addition, the formulas utilize another GEMPACK convention, where the sum between two percent change variables, say “x+y” is shorthand for 100*((1+x/100)*(1+y/100)-1).
This is the single-year (2030) marginal change in Real GDP in 2019 dollars.
6G could add significant disruptive demand changes, which would impact the 2030 network buildout requirements that would be larger than currently modeled to reflect the anticipation of the 6G rollout. 6G considerations are a potential future enhancement to the model.
The NIM does not dynamically model optimal firm behavior. The purpose is to provide a reasonable approximation of industry costs as inputs for the macroeconomic model. Changes in assumptions like uptake rate must be done outside the model.
This graphic is shown for broad comparison purposes only. Assumptions, time frame, level of detail, and transparency of method vary across studies. For example, it is unclear whether some reports adjust for inflation, as this article’s estimate does. In general, such differences and uncertainties in the details of what is being reported complicates the comparison.
For example, one might conjecture that the unit costs increase as the industry expands 4G service into remote locations.
The tests performed in this step assume that each parameter is drawn with uncertainty from a symmetric triangular distribution with minimum and maximum values defined by the researcher. The model is solved at discrete points along this triangular distribution, and Gaussian quadrature is used to calculate the means and standard deviation of the model variables. Note that due to the discretization of the triangular distribution, the results are approximations to the true mean and standard deviation. The confidence intervals are then developed using Chebyshev’s inequality. See (Channing and Pearson 1998) for further details regarding systematic sensitivity analysis for parameters of the CGE model.
The ranking includes all 20 industries that pay at least 0.77% of total industry costs on wireless services.
Because the technical coefficient parameters are typically exogenous, it is straightforward to create technical efficiency shocks that are common across multiple inputs. For example, is a common term that will affect the technical efficiency of all inputs to production.
where and are two weighting parameters.
where and are both nonzero parameters that are less than 1. The CES functional form of the aggregate factor volume is similar.
The firm’s behavioral response to the changes in the price, efficiency, and preferences in its decision to purchase imported and domestic goods is not shown for simplicity.
Because movements in prices and output are endogenous and subject to general equilibrium effects, differs across inputs.
Note that the first-order condition on the previous page implies that creates a common reduction in the demand for every input by the composite wireless industry .