Presentation delivered during 9th Seminar on Media and the Digital Economy (21-22 March 2019). http://fsr.eui.eu/event/annual-scientific-seminar-on-media-and-the-digital-economy-9th-edition/

1. Benefits of regulation vs.
competition where inequality is high:
The case of mobile telephony in South
Africa
Ryan Hawthorne1 Lukasz Grzybowski2
1University of Cape Town, Acacia Economics
2Telecom Paristech
Annual Scientific Seminar on Media and the Digital Economy
Florence, 21-22 March 2018

2. Introduction
Widening income inequality is being debated globally.
South Africa is the most unequal country in the world as a
consequence of Apartheid.
Inequality could be reduced by lowering prices of goods on
which poor consumers spend their income through greater
competition & regulation.
Competition in telecommunications is determined through
state control, and therefore a key policy lever to reduce
inequality.
The benefits from telecommunications are beyond lower
prices: access to information and services which are otherwise
not availalble.

3. Poor consumers spend up to 5% of income on
mobile
3.6
2.7 2.3 2.2
012345
Wave 1 (2008)
3.4
2.8 2.7
1.8
012345
Wave 2 (2010-11)
4.7
4.2
3.4
2.2
012345
Wave 3 (2012)
4.2 3.8
3.1
2.4
012345
Wave 4 (2014-15)
<R2,500pm 2,500-5,000pm
5,000-10,000pm >10,000pm
Income quantiles
Source: Computations using the National Income Dynamics Survey (NIDS) in years 2008-2015

4. Our approach
We develop a structural model of demand & supply for mobile
voice services.
The model is estimated using individual-level survey data on
ca. 134,000 consumers in years 2009-2014 who choose
between four network operators.
We use the model to conduct counterfactual simulations:
(i) exclude recent entrants from the market (as if there was no
entry);
(ii) assume there was no regulation of mobile termination rates
(which reduced them by 90% in years 2009-2014).
For these scenarios, we recompute equilibrium prices,
penetration & consumer surplus for different consumer
segments based on income.

5. Related literature
Demand estimation for telecommunications services based on
discrete choice framework (e.g. Cordona et al., 2009;
Grzybowski et al., 2014).
The role of telecommunications for economic development
(e.g. Roller & Waverman, 2001; Czernich et al. 2011).
The impact of entry into markets for telecommunications
services on prices (e.g. Economides et al., 2008; Nicolle et al.,
2018).
The impact of regulation of mobile termination rates (MTR)
on retail prices (e.g. Genakos & Valletti, 2015).
But there is no research on the distributional consequences of
entry & regulation, which is made possible in this case by
using individual-level data.

6. New entrants target high-income consumers
3000400050006000
HerfindahlHirschmanIndex
0 10000 20000 30000
Household income (mean)
HHI Fitted values
Source: Surveyed consumers grouped by regions (which have different income levels).

7. Concentration lower in areas with higher
incomes
(1) (2) (3)
Towns -311.28*** -318.98*** -236.02***
Cities -408.07*** -424.06*** -276.15***
HH income (mean) -0.01* -0.00
High income % pop -360.27+
Coloured % pop -724.27***
White % pop -376.44*
Indian % pop -2074.03***
Black % pop (base)
Constant 4660.84*** 4615.95*** 4805.16***
R-Square 0.11 0.11 0.29
Number of obs 420 420 420
+ p<0.10, * p<0.05, ** p<0.01, *** p<0.001

8. Mobile voice prices declined over the period
0.511.52
Averageprices(Randsperminute)
2009 2010 2011 2012 2013 2014
Year
Cell C
MTN
Telkom Mobile
Vodacom
Average prices
0.511.52
Terminationrates(Randsperminute)
2009 2010 2011 2012 2013 2014
Year
MTN,Vodacom peak
MTN,Vodacom offpeak
Telkom(M),Cell C peak
Telkom(M), Cell C offpeak
Termination rates

9. Random utility framework
Vijt = xjtβi − αi pijt + ijt
where:
xjt is a J × 1 vector of network dummy variables
βi is a J × 1 vector of coefficients denoting the
individual-specific valuations for the different networks
pijt is price paid by consumer i for calling on network j
αi is a random coefficient for individual-specific valuation of
price
ijt is individual-specific valuation for network j (iid extreme
value distributed)

10. Observed and unobserve heterogeneity in
consumer preferences
βi
αi
=
β
α
+ ΠDi +
0
σα
νi , νi ∼ N(0, 1)
where:
(β, α) is (J + 1) × 1 vector of mean valuations
Di is d × 1 vector of individual chars: gender, age, race, lang,
province, income, (self)employed, tel home or work, computer
Π is (J + 1) × d matrix of parameters, capture impact of
individual chars. on valuations for J network dummies xjt and
price variable pjt
νi is randomly drawn vector from standard normal distribution,
captures unobserved individual heterogeneity regarding price
σα is vector of std. devs around mean valuations (no
unobserved heterogeneity when σα = 0).

11. Maximum simulated likelihood
The model is estimated by taking R draws for νi from the standard
normal distribution to obtain the average choice probability per
individual:
Pij =
1
R
R
r=1
exp xj β − (α + σνr
i )pij + (xj , pij )ΠDi
k∈Ci
exp xkβ − (α + σνr
i )pik + (xk, pik)ΠDi
In the special case of no unobserved heterogeneity (σ = 0), this
expression reduces to the multinomial choice probability.
The parameters are estimated by maximizing the log-likelihood
function:
L(θ) = yij
N
i j
log(Pij ).
where yij = 1 if individual i chose alternative j and yij = 0
otherwise.

12. Price elasticities computed after demand
estimation
Aggregate elasticity of demand for subscriptions to network j with
respect to price of network k defined as:
εjk =
1
N
i
∂Pij
∂pik
pik
1
sj
= i (−αi )Pij (1 − Pij )pij / i Pij k=j
i αi Pij Pikpik/ i Pij k=j
.
where:
αi is the individual-specific price coefficient
pij is the observed price faced by consumer i for operator j
Pij is the computed probability that consumer i chooses
operator j
sj is the aggregate market share for network j, given by
sj ≡ i Pij /N, where N is no. of consumers in sample in a
year

13. Competition on the supply side
The profits of firm j are given by:
Πj (p) = (pj − cj ) sj (p)L
where:
cj is the marginal cost
sj (p) is firm’s j’s market share as a function of the price vector
market size is denoted by L
Assume firms choose prices to maximize profits. The FOCs are:
sj (p) + (pj − cj )
∂sj (p)
∂pj
= 0.

14. Computing marginal costs
In vector notation, the FOCs which yield equilibrium prices are
given by:
p = − ΘF
∆
−1
s(p) + c
where:
p and s(p) are J × 1 price and market share vectors
∆(p) ≡ ∂q(p)/∂p is J × J matrix of own- & cross-price
derivatives
ΘF is J × J matrix, 1 s for products of same firm, 0 s
otherwise
denotes element-by-element multiplication of two matrices.

15. Counterfactual simulations
Assessing the impact of competition:
1 Remove new entrants Telkom Mobile (2010) and Cell C
(2001) from consumer choice set.
2 Use iterated best-responses to compute equilibrium prices of
the remaining firms.
3 In parallel, compute choice probabilities for each consumer
and aggregate market shares of firms.
4 Calculate change in consumer surplus by comparing ‘old’ and
‘new’ equilibrium.
Assessing the impact of MTR regulation is analogous, but the
assumption is that MTR rates remained unchanged since 2009.

16. Computing change in consumer surplus
∆E(CSit) =
β
1
|αi |
ln
j
exp(V 1
ijt) − ln
j
exp(V 0
ijt) dθi
where:
αi is the individual-specific price coefficient
V 0
ijt is the observed part of the utility function before and
V 1
ijt after entry/regulation

17. Control function to account for
endogeneity of price
Prepaid Medium High
Coef. (Std.) Coef. (Std.) Coef. (Std.) Coef. (Std.)
MTR 1.30*** (0.16)
MTN 0.80*** (0.16) -0.28 (0.22) -0.21 (0.22) -0.22 (0.22)
Telkom 0.36* (0.18) -0.42+ (0.25) -0.21 (0.25) -0.27 (0.25)
Vodacom 0.50** (0.16) -0.07 (0.22) -0.14 (0.22) -0.19 (0.22)
Prepaid 0.68*** (0.16)
Postp. M 0.18 (0.16)
Postp. H 0.44** (0.16)
N 88
R-squared 0.68
+ p<0.10, * p<0.05, ** p<0.01, *** p<0.001

18. High income consumers are less price sensitive
Condition. logit RC logit
Price -2.14*** (0.08) -2.00*** (0.09)
SD Price 0.76*** (0.03)
Price*
Income 3-8 0.14*** (0.03) 0.23*** (0.04)
Income 8-16 0.07* (0.03) 0.20*** (0.04)
Income 16+ 0.03 (0.04) 0.16*** (0.04)
Black 0.36*** (0.07) 0.50*** (0.08)
Coloured 0.26*** (0.04) 0.26*** (0.04)
Indian -0.24*** (0.05) -0.32*** (0.06)
Afrikaans -0.10 (0.08) -0.13 (0.08)
English 0.11 (0.07) 0.10 (0.08)
Zulu+ -0.03 (0.04) -0.03 (0.04)
Xhosa 0.57*** (0.04) 0.62*** (0.05)
Number of obs 636,891 636,891
+ p<0.10, * p<0.05, ** p<0.01, *** p<0.001

19. Consumers in towns & cities choose entrants
Condition. logit RC logit
Control function 1.78*** (0.04) 1.47*** (0.04)
Cell C 1.92*** (0.15) 2.08*** (0.17)
MTN 4.12*** (0.16) 4.14*** (0.17)
Telkom -1.13*** (0.21) -0.97*** (0.22)
Vodacom 4.04*** (0.15) 4.14*** (0.17)
Towns*
CellC 0.62*** (0.04) 0.67*** (0.04)
MTN 0.27*** (0.03) 0.33*** (0.03)
Telkom 0.66*** (0.17) 0.71*** (0.17)
Vodacom 0.19*** (0.03) 0.25*** (0.03)
Cities*
CellC 1.03*** (0.04) 1.11*** (0.04)
MTN 0.40*** (0.03) 0.49*** (0.03)
Telkom 1.16*** (0.16) 1.23*** (0.16)
Vodacom 0.26*** (0.03) 0.35*** (0.03)
+ p<0.10, * p<0.05, ** p<0.01, *** p<0.001

20. Older, rural consumers less likely to be
connected
Condition. logit RC logit
Cell*
Age 26-50 0.05** (0.02) 0.07** (0.03)
Age 51-65 -0.50*** (0.02) -0.61*** (0.03)
Age 65+ -1.25*** (0.03) -1.55*** (0.04)
Male -0.35*** (0.02) -0.43*** (0.02)
Working 0.54*** (0.02) 0.68*** (0.03)
Self-employed 0.25*** (0.04) 0.28*** (0.05)
Telephone-home -0.25*** (0.02) -0.30*** (0.03)
Telephone-work 0.53*** (0.04) 0.62*** (0.05)
Western Cape -0.05+ (0.03) -0.08+ (0.04)
Northern Cape -0.42*** (0.04) -0.51*** (0.05)
Free State 0.01 (0.04) 0.03 (0.04)
Eastern Cape -0.46*** (0.03) -0.59*** (0.04)
KwaZulu Natal -0.03 (0.03) -0.04 (0.04)
Mpumalanga 0.47*** (0.05) 0.57*** (0.06)
Limpopo 0.09* (0.04) 0.15** (0.05)
North West 0.07+ (0.04) 0.10+ (0.05)

21. Asymmetry in substitution between operators
Cell C MTN Telkom Vodacom
Cell C -1.37 0.73 0.01 0.77
MTN 0.19 -1.35 0.01 0.74
Telkom 0.21 0.74 -1.75 0.76
Vodacom 0.19 0.72 0.01 -1.15

22. Average mark-ups (2011-2014)
Operator MTR costs Prices Mark-ups Simul. Simul.
prices mark-ups
ZAR/min ZAR/min % ZAR/min %
Cell C 0.41 1.04 62 1.15 66
MTN 0.29 1.40 79 1.38 79
Telkom 0.46 1.16 61 1.11 60
Vodacom 0.27 1.28 79 1.34 80

23. High-income consumers benefit the most
Price No Telkom No Cell C No regulation N
Price ∆CS Price ∆CS Price ∆CS
Income
<R3k 1.01 1.01 -0.00 1.05 -0.13 1.24 -0.21 18,379
R3-8k 1.12 1.12 -0.00 1.18 -0.16 1.38 -0.25 24,610
R8-16k 1.20 1.19 -0.00 1.27 -0.17 1.47 -0.27 21,638
>R16k 1.25 1.25 -0.00 1.34 -0.17 1.55 -0.29 23,080
Average 1.15 1.15 -0.00 1.21 -0.16 1.42 -0.26 87,707
Race
Black 1.16 1.16 -0.00 1.22 -0.17 1.43 -0.26 45,452
Col’d 1.03 1.03 -0.00 1.07 -0.13 1.28 -0.23 12,434
Indian 1.09 1.09 -0.00 1.13 -0.14 1.37 -0.26 5,925
White 1.21 1.21 -0.00 1.28 -0.15 1.48 -0.28 23,896
Average 1.15 1.15 -0.00 1.21 -0.16 1.42 -0.26 87,707

24. Entry & regulation result in more uptake
among poor
Actual Telkom Telkom, CellC Regulation N
Income
<3,000 74 74 70 66 18,379
3-7,999 84 84 81 78 24,610
8-15,999 90 90 88 85 21,638
>15,999 95 95 93 91 23,080
Total 86 86 84 81 87,707
Race
Black 86 86 83 81 45,452
Coloured 79 79 75 71 12,434
Indian 86 86 82 77 5,925
White 91 91 89 86 23,896
Total 86 86 84 81 87,707

25. Conclusions
We use a structural model of demand & supply of mobile voice
services to simulate the impact of competition & regulation.
Regulation had a significantly greater impact on consumer
surplus.
High income consumers benefited most from entry and
regulation of mobile services in South Africa.
At the same time, mobile penetration expanded most among
low income consumers as a result of entry and regulation.
Entrants appear to target high-income consumers in towns
and cities, and policymakers & regulators need to consider
means of directing operators to serve the poor.