Financial implications of mobile phone-based personal carbon trading: a case study of Safaricom

Personal carbon trading (PCT) has garnered significant interest in the literature as an alternative policy instrument to the largely unpopular carbon tax. However, it has been hindered by the cost and administrative complexity concerns as a result of covering potentially millions of emitters. This work expands on a prior study which presented a mobile phone-based PCT scheme for personal road transport in Kenya. In that study, the system design and operation was extensively developed, and distributional impact was assessed using sample data of motorists to identify equity issues. In this extension, I justify the scheme further by assessing the cost concerns using a case study approach of the mobile service provider, Safaricom. Data from the sample survey is revisited and combined with Safaricom’s financial reports to simulate the potential cost of the scheme. Results revealed running costs of less than £80 000 annually, several times lower than estimates that relied on the chip card system. Policymakers and researchers are encouraged to build on this scheme’s viability as a globally-inclusive variant of PCT.


Introduction
Climate change as a result of anthropogenic causes has resulted in the development of several policy instruments in order to mitigate rapidly increasing emissions.The most common are market mechanisms such as carbon taxes (pricing instruments) and emission trading systems (quantity instruments).Personal carbon trading (PCT) is a broad type of a downstream emissions trading system, first developed in the UK, whereby either household or both household and transport-related emissions are capped and distributed as emissions allowances (Fawcett and Parag 2010).There are various types of PCT schemes that have been developed extensively to justify the importance of downstream participation of emissions trading (see Ayres 1997, Hillman and Fawcett 2004, Raux and Marlot 2005, Starkey and Anderson 2005, Raux 2010, Starkey 2011, Raux et al 2015).
While emissions trading systems currently in operation involve the participation of a few stationary emitters in the form of industries, PCT involves potentially millions of individuals.Hence, its implementation and operation would benefit from minimizing cost and administrative complexities.Prior works have suggested that as a result of today's technological possibilities, operationalizing PCT can be cost-efficient if a smart design leveraging existing technologies is used to manage the large number of participants (Raux 2010).
However, PCT is yet to benefit from a practical, evidence-based argument to support both its administrative complexities and high running costs which has made its implementation elusive.Previous proposals have perceived certain designs to be feasible such as using chip card technology, but these studies lack realistic, evidence-based cost estimations in order to be considered by policymakers for actual implementation.For the works that have loosely estimated running costs, they also suffer from the myopia of only developed country consideration, leaving behind the possibility of a globally implementable policy option.Therefore, a simple, cost-efficient system needs to be empirically justified.
This work expands on a prior study by Al-Guthmy and Yan (2020) which presented a popular mobile money transfer platform called M-Pesa as a potential solution for PCT implementation in Kenya.In this extension, the argument for that scheme is further developed by establishing the financial implications using a case study approach of the mobile service provider, Safaricom.

Conventional approach and limitations: chip card technology
Since the mid 90s when Fleming (Fleming 1997) introduced tradable domestic quotas (DTQs), there has been a culmination of academic and political research into designs motivated by the proponents of urgent climate change mitigation.In the literature, the enabling infrastructure of PCT has traditionally focused on using the chip card system for whole economies (Starkey and Anderson 2005, DEFRA 2008, Burgess 2016), household energy (Niemeier et al 2008) and road transport (Fleming 1997, Raux and Marlot 2005, Harwatt 2008, Raux 2010, Harwatt et al 2011, Rothengatter et al 2011).From the various proposals, two key works attempted to estimate the cost of PCT implementation: DEFRA's domestic tradable quota (DTQ)-inspired PCT (Lane et al 2008) and tradable carbon permits (TCPs) by Harwatt et al (2011).
For DTQ, the UK government commissioned several prefeasibility studies, one of which was the technical feasibility and potential cost (Lane et al 2008) based on the chip card system and concluded that it was not the right time for implementation.The costs and administrative complexities were deemed too high to be acceptable at the time (DEFRA 2008).The report used simple inhouse estimates of the setup and running costs by the consultants and lacks financial implications for participating individuals.As a result, it strongly recommended more research into alternative ways to implement PCT to address the issues raised.The high costs and system complexities of a card-based system are evident from two main flaws:

Additional hardware
For road transport proposals, the motorist must obtain and carry a chip card to use at a fuel retail station in order to purchase fuel and update their quota balance.This requires every retailer to either have either a card terminal or modify existing chip card systems at the premises (if any) to accommodate the scheme.The system would connect to a dedicated server and run specialized software to manage the quota accounts.The findings from the DEFRA report concluded this not be feasible.

Unnecessary involvement of intermediaries
The trading of quotas by participants using the chip card system would need to be done through intermediaries.Harwatt et al (2011) state that the information technology available during that period was adequately sophisticated to establish and manage a national chip card system that could be used at post offices, fuel stations and even online.Other works suggest that even though there are cheaper alternative instruments, the costs of the chip card-based PCT scheme are justified by the benefits of equity, acceptability and overall efficiency (Starkey and Anderson 2005).However, they add an unnecessary layer of involvement and by default, administrative complexities and overheads.Furthermore, the private sector involvement would require significantly more compliance and regulatory checks to ensure the system is not compromised by fraudulent activities, placing a significant burden on the regulator.

Research approach: cost-estimation of a mobile phone-based PCT
This research uses the Safaricom-based PCT system proposed by Al-Guthmy and Yan (2020) to provide evidence of cost feasibility for implementation in personal road transport.In that paper, the system design and operation was extensively developed, and distributional impact was assessed using sample data of 500 motorists across different regions to identify equity issues.The platform operates using the unstructured supplementary service data feature of basic mobile phones, hence even those without smart phones are able to participate.
Motorists would be allocated carbon credits through their mobile phones by the regulator and would be required to surrender the appropriate number of credits to the pump attendant when refuelling.Users would be able to access the trading platform on their phones to buy or sell carbon credits by placing bids or offers, similar to stock trading in a centralized marketplace.All motorists are able to participate due to the interoperability of mobile networks and those without their phones may be able to access their account securely at a retail fuel station using their login credentials.The distributional impact assessment of the motorists revealed the PCT system to be significantly progressive under different scenarios.The integration into an existing mobile platform and elimination of unnecessary intermediaries would reduce the potential costs although the authors did not provide evidence of the cost-savings.
In this expansion, Safaricom's annual financial reports and performance statistics are analyzed and combined with sample survey data to provide potential running costs of operating PCT using the same infrastructure.The analysis is conducted using Monte Carlo simulation which approximates the number of transactions  a Safaricom Limited (2014,2015,2016,2017,2018,2019).
that could be made which are then combined with the estimated running costs and projected to a national scale.

Method
Using publicly available financial statements from Safaricom, and sector statistics of mobile service providers from the Communications Authority of Kenya (CAK), the per-transaction costs and revenues of operating M-Pesa are derived.The sample data used by Al-Guthmy and Yan (2020) contains the list of motorists with fuel quota deficits under 3 allocation methods: the equal per-vehicle allocation (EpVA), equal per-capita allocation (EpCA) and   needs-based allocation (NbA) methods.Assuming each purchase of quotas has a fixed fee, we must predict how often each motorist purchases quotas to clear their deficit.95% of the quota deficits under all allocation methods range between 208 and 345 L of fuel quotas.Using the Monte Carlo method, random trials are performed using a discrete uniform distribution of 5 outcomes of quota purchases of each motorist's deficit until the deficit is depleted.These are 20%, 40%, 60%, 80% and 100% of quota deficits.Figure 1 contains a flow chart which illustrates the procedures undertaken and performed using Microsoft Excel.The method was performed in 11 steps which are divided into three outputs defined below.

Estimating the per-transaction cost and revenue of M-Pesa
(Step 1) From Safaricom's annual financial reports, extract the line items of interest as shown in table 1 below.Then calculate the average growth of each line item by subtracting the previous year's value in order to identify the effect.(Step 2) Next we to divide revenues and cost by M-Pesa customers to obtain the per-customer statistics for   (2014,2015,2016,2017,2018,2019).each period.Note that the M-Pesa cost is the sum of the direct and operating costs.
where: RC is the M-Pesa revenue per customer; CC is the M-Pesa cost per customer; MR is the M-Pesa revenue; and MC is the M-Pesa cost; and NC is the number of M-Pesa customers.(Step 3) From the CAK sector statistics reports (Communications Authority of Kenya 2015Kenya -2019aKenya , 2015Kenya -2019bKenya , 2016Kenya -2019aKenya , 2016Kenya -2019b)), obtain the total number of M-Pesa transactions for each year by adding the number of person-to-person (P2P) and withdrawal transactions to the number of e-commerce transactions, given by: where: TT is the total number of transactions; P2P is the number of person-to-person transactions; W is the number of withdrawal transactions; and EC is the number of ecommerce transactions.where: RT is the M-Pesa revenue per transaction; and CT is the cost per transaction.(Step 5) The above revenue includes taxes.For PCT, we expect taxes to be excluded.Hence excise tax of 12% payable on sales and corporate tax of 30% payable on earnings before tax are deducted from the revenue to give discounted price (revenue) for PCT.where: Tax is the total tax per transaction; Etax is the excise tax component per transaction; Ctax is the corporate tax component per transaction; and PRT is the revenue per transaction expected under a PCT scheme.

Simulating the number of transactions using Monte Carlo method
(Step 6) Extract the number of motorists with quota deficits under each allocation method from the sample survey data.The number of motorists in deficits and their proportions to the sample are provided in table 2. (Step 7) Choose random numbers against the discrete uniform distribution to select the percentage of each motorist's deficit balance reduced for each transaction until depletion of the deficit.The discrete choices are 20%, 40%, 60%, 80% and 100% reduced.This means motorists reduce their quota deficits by these percentages which results in between one to five possible transactions depending on each random outcome.(Step 8) Sum up each motorists' transactions and repeat this process for 1000 trials per allocation method.(Step 9) The mean value of all trials results in the approximate number of transactions under each allocation method for a given year of running the PCT scheme.(11) where: MF is the multiplication factor; MP is the number of motorists in the population; MS is the number of motorists from the sample; PR is the population revenue; SR is the sample revenue; PC is the population cost; and SC is the sample cost.

Data summary
The Communications Authority of Kenya (CAK) publishes quarterly sector statistics of the mobile carriers which includes Safaricom.Its financial year runs from April to March each year, and its performance is reported in the CAK's reports in the order quarter 4, 1, 2 and 3. Table 3 shows the total M-Pesa-related transactions which is the sum of the P2P, withdrawal and e-commerce transactions for each financial year.M-Pesa has seen consistent year-on-year growth in active number of customers as shown in figure 2. The 2019 financial performance shown in figure 3 depicts higher values than the 6 years average between 2014 to 2019 reflecting the consistent growth of the service.
According to the firm's financial statements, the direct costs associated with M-Pesa are commissions which are paid to the agents across the country.Customers must visit them to deposit efloat (with cash) or withdraw cash (with efloat).The other cost shown in figure 3 is the operating cost which is not only for M-Pesa, but for all Safaricom's services.Of all the revenue streams, voice is the largest and is followed by M-Pesa.
Figure 4 depicts how M-Pesa has had the highest average revenue growth at 48.33% compared to other revenue streams.Operating costs increased only by 10.98%.Even though voice revenues dominate all other services, M-Pesa revenue growth rate is higher.At the current growth rate, M-Pesa revenues may surpass voice revenues in less than half a decade.
The proportion of M-Pesa's revenue growth rate (48.33%) to the overall revenue growth rate was taken from the operating costs growth rate (10.98%) to estimate the M-Pesa operating costs.In other words, the operating cost of M-Pesa is taken as 48.33% of the increase in operating costs.With the estimated M-Pesa operating cost, it was possible to assess the financial indicators of M-Pesa on a per-customer and per-transaction basis.
For the average per-customer financials (figure 5), the total cost (direct and operating cost) accounts for 45% of the revenue while the net earnings after tax account for 30%.These are for the average transactions performed by each customer annually.The number of transactions are shown in figure 6.The average transactions (for ecommerce, P2P and withdrawals) per customer per year totaled 113 for the period 2014-19.

PCT revenue per transaction
The per-transaction financial indicators in figure 7 show the average revenue of each M-Pesa transaction of £0.30 including tax.The M-Pesa cost (which is also taken as the PCT cost) is £0.14 (approximately 18 US cents).The M-Pesa revenue is lowered by excise and corporate tax to give a final per-transaction PCT price of £0.23.

Simulated number of transactions
The simulations for each of the three allocation methods resulted in several summary statistics shown in table 4.
The mean was chosen as the representative value across all allocation methods using the multiplication factor of 512 0363 /500.

Financial implications of PCT in Kenya
Figure 8 presents the simulated national scenarios for each allocation method in terms of annual total revenues generated, running costs incurred and net earnings.These figures represent the estimates for the 512 036 personally-owned motor vehicles in Kenya.Since NbA has the highest number of motorists in deficit, it has the highest values.The opposite is true for EpVA while EpCA remains in between the two.The results of the financial implications of PCT in Kenya have been presented by combining M-Pesa's financial indicators and simulating the number of possible transactions by motorists.These findings are discussed in the next section.

Alleviating cost concerns
The annual running costs for the population of motor vehicles would average no more than £80 000 based on the simulation findings which would cover the cost of running the scheme.Safaricom's return on investment (ROI) was found to be 65% and when added to the total transaction cost, the PCT revenue (transaction fee) amounted to a 25% reduction in the transaction price, further making the system more affordable than the normal M-Pesa fees as a result of PCT being tax-free.Furthermore, it ensures adoption by Safaricom as it maintains profitability for its shareholders.

Comparison with DEFRA prefeasibility report
One of the DEFRA reports was a prefeasibility report on the technical feasibility and potential cost of PCT using the chip card system in the UK (Lane et al 2008).The report suggested the set-up costs incurred by the government and intermediaries for a PCT scheme covering the whole economy of 50 million participants would be approximately £700 million to £2 billion4 .This excludes fixed or running costs incurred by participating individuals.After adjusting for inflation at an average of 2.7% per year to 2019 prices (Bank of England 2020), this range increases to £941 million to £2.7 billion.Table 5 shows the set-up cost from the DEFRA estimate for implementing the chip-card system for intermediaries which was subsequently rejected by the UK government.
The M-Pesa PCT set-up costs are nil as the government would simply provide Safaricom a lucrative business opportunity with the same ROI.In return, Safaricom would bear the setup costs in order to win the tender for this project.There would certainly be other setup costs borne by the government in designing and overseeing the project; this is treated as a constant for any alternative policy that would be introduced.These unaccounted costs may be minimal and could be absorbed as part of a profit-sharing agreement between the government (which a 35% shareholder of Safaricom) and Safaricom.DEFRA's estimate suffers from the flaws discussed in the beginning of this study.The report makes 3 key suggestions for further investigation which have been addressed in this study, namely addressing the key cost drivers, consultations for and better alternatives.Harwatt et al (2011) proposed TCP for road transport in the UK and estimated the costs involved to set up and run the scheme whereby each adult is allocated allowances.Table 6 shows the line items that were focused on, adjusted for inflation (Bank of England 2020).

Comparison with tradable carbon permits (TCP) scheme
The operating costs of £2.6 million are significantly higher than this study's estimate of £80 000 (see table 7).This could be explained by Harwatt et al's use of the UK vehicle licensing agency's operating costs as the closest estimate to running the TCP scheme and the fact that allocation is made to all adults even though the scheme is limited to road transport.
The scanning equipment required by all fuel retailers is also an unnecessary cost that a mobile phone-based PCT system would render unnecessary.

Limitations
This study also does not take into account indirect costs outside the scope of Safaricom such as the lost fuel tax revenue.These would not be as much of a concern to policymakers as the running costs.This was left out intentionally as it is assumed to be at least partially offset by the large import bills the country incurs annually in relying on fuel imports and also a price the government must be willing to pay to meet its climate change ambitions.Another limitation is that even though the comparison with the works of DEFRA and Harwatt account for inflation, the costs of setting up a chip card system may have reduced considerably.Nonetheless, the costs realized through M-Pesa are many times lower than these estimates and the mere fact that additional hardware and intermediaries are required for the chip card system diminishes this concern.

Conclusion
This study furthers the case for a potential PCT system inspired by the M-Pesa mobile money transfer platform.This is achieved by assessing the financial implications of the scheme for personal road transport.The simulation exercise provided evidence of the low running costs and tax-free transactional fee for motorists making it more affordable than the prevailing M-Pesa fees.Intermediaries are eliminated and no additional hardware is required to operate the scheme.The cost and complexity barriers which previous proposals failed to solve by using the chip card system have been addressed.The added advantage of using a developing country as a case study makes the argument for PCT even more appealing to policymakers for consideration as a globally-inclusive and feasible policy option.This research is the first to empirically show the feasibility of PCT in a developing country context and which leverages basic mobile phones.It is hoped that this will trigger more interest in developing country participation and also consideration of this scheme in developed countries.

FFigure 1 .
Figure 1.Flow chart of revenue and cost estimation method of an M-Pesa-based PCT system.

Figure 4 .
Figure 4. Average growth of revenues and costs.M-Pesa has the highest growth rate.Note the effect of revenue growth on operating cost growth.Adapted from Safaricom Limited(2014, 2015, 2016, 2017, 2018, 2019).

Figure 5 .
Figure 5. Financial indicators of M-Pesa on a per-customer basis (average of 2015-19).

Figure 6 .
Figure 6.Average number of M-Pesa transactions per customer per year.Note the increase attributed to more payment options using M-Pesa such as fuel purchase, shopping and paying utility bills in addition to P2P and withdrawal transactions.
Pesa revenue and cost estimates to the simulated transactions (Step 10) The number of transactions under each allocation method are multiplied by the PCT revenue per transaction and the M-Pesa cost per transaction.The difference between the two values provides the tax-free earnings for each of the 3 allocation methods.(Step 11) The sample results are translated into population estimates by the multiplication factor of the population.

Table 1 .
Financial data extracted from Safaricom's annual financial reports.
a Line item Description No. of active M-Pesa customers The reports provide 30 days active customers, so this is scaled to annual level to match the rest of the data M-Pesa revenues These are provided in the annual reports Direct cost Commissions paid to agents who exchange efloat and cash Operating cost (including capital expenditure) Only overall values are provided, so it was necessary to derive an estimate for M-Pesa alone as shown: MOC = MRG TRG × TOC (1) Where: MOC is the estimated M-Pesa operating cost; MRG is the average M-Pesa revenue growth (2014-19); TRG is the average total revenue growth (2014-19) and; TOC is the average total operating cost (2014-19)

Table 2 .
Frequencies of sample and population of motorists in deficit.a

Table 3 .
Total M-Pesa transactions.a a Aggregated from: Communications Authority of Kenya (, , , ). b Unavailable and estimated by extrapolating backwards the quarter's values from subsequent years.

Table 4 .
Monte Carlo simulation summary statistics for each allocation method.
Figure 8.Estimated annual revenues, costs and earnings for M-Pesa-based PCT scheme for personal road transport in Kenya.Running costs do not exceed £80 000.

Table 5 .
Bank of England (2020)d M-Pesa PCT set-up costs.aAdjustedfor inflation to 2019 prices using the inflation calculator provided by theBank of England (2020).
a Adapted from Lane et al(2008).b

Table 6 .
Estimated financial costs of TCP. a

Table 7 .
Comparison of TCP and M-Pesa PCT running cost implications.
a Adjusted for inflation to 2019 prices using the inflation calculator provided by the Bank of England (2020).