Review of DBA's proposed revision of the WACC calculation for the telecom market in Denmark

Transcription

1 WIK Report Study for The Danish Business Authority Review of DBA's proposed revision of the WACC calculation for the telecom market in Denmark Authors: Dr. Werner Neu Dr. Stephan Schmitt Ing. Peter Kroon WIK Wissenschaftliches Institut für Infrastruktur und Kommunikationsdienste GmbH Rhöndorfer Str Bad Honnef Germany Bad Honnef, 16 December 2016

2 Imprint WIK Wissenschaftliches Institut für Infrastruktur und Kommunikationsdienste GmbH Rhöndorfer Str Bad Honnef Germany Phone: Fax: info(at)wik.org Person authorised to sign on behalf of the organisation General Manager and Director Director Head of Department Postal Services and Logistics Authorized Signatory Head of Department Networks and Costs Director Head of Department Regulation and Competition Authorized Signatory Head of Administration Chairman of the Supervisory Board Dr. Iris Henseler-Unger Alex Kalevi Dieke Dr. Thomas Plückebaum Dr. Ulrich Stumpf Karl-Hubert Strüver Winfried Ulmen Registered at Amtsgericht Siegburg, HRB 7225 Tax No. 222/5751/0722 VAT-ID DE

3 Review of DBA WACC I Contents 1 Introduction 1 2 Review of DBA's general approach Comments on "Initial remarks" Comments on "Preliminary results" Comments on "Basic principles for the revision" 5 3 Review of the methods for setting parameter values Risk-free interest rate Equity risk premium (ERP) Equity beta Credit risk premium Gearing Corporate tax rate Risk premium for NGA Inflation 17

4 II Review of DBA WACC

5 Review of DBA WACC 1 1 Introduction Danish Business Authority (DBA), the national regulatory authority in Denmark, is in the process of reviewing all aspects of the Weighted Average Costs of Capital (WACC). The revised WACC will be applied in DBA s cost models, which calculate wholesale access prices for fixed and mobile networks in Denmark. DBA has already consulted its first draft WACC revision with stakeholders in July Their input is currently considered and incorporated in a final draft report on the revision. This WIK report results from an objective external review by WIK of this draft report and will serve as input for a second draft WACC proposal by DBA. DBA anticipates to send this second draft proposal for a second consultation round at the beginning of 2017 to the Danish stakeholders. WIK s task is to review the basic principles proposed to be incorporated in the revised general method and the reasoning used for the approach to determine the values of specific parameters. This also involves verifying whether the proposed approach is solid from an economic and regulatory perspective. The actual determination of values for the different parameters is outside the scope of this review as well as the verification whether the positions of the stakeholders are correctly summarised. When evaluating DBA's proposed revision of the WACC, we will bring to bear our understanding based on our experience and reading of the literature. Particular attention will, however, be paid to the report prepared by The Brattle Group 1 for the European Commission (from here on the Brattle Group report), as this report may be regarded as the Commission's guideline to NRAs when determining the value of the WACC. 1 The Brattle Group, Review of approaches to estimate a reasonable rate of return for investments in telecoms networks in regulatory proceedings and options for EU harmonization, 2016.

6 2 Review of DBA WACC 2 Review of DBA's general approach DBA describes its general approach to determining the WACC in the opening sections of its draft report, in particular in the sections entitled "Initial remarks", "Preliminary results" and "Basic principles of the revision". The following comments will be on these sections. There is one more early section in the DBA draft report entitled "Consultation and early involvement of the industry" which will not be commented on. 2.1 Comments on "Initial remarks" Early in this section, DBA makes the following statement: The point of departure for the LRAIC model is the forward-looking costs in an ideally operated network and company, based on modern, efficient technology. Therefore, the costs and investments in the telecom network are included in the calculation of maximum prices, and, in the LRAIC model, these investments must yield a reasonable return. The reasonable return must equate to the WACC value. From above we take it that the value of the WACC, being equal to the reasonable return that is part of the forward-looking costs, must also be determined according to the forward-looking perspective. We take up this point later when discussing the basic principles Further on in the initial remarks, it is stated that the objective of revising the method for the WACC calculation is to create greater transparency of the process. While we do not know to what degree the previous method applied by DBA may not have been sufficiently transparent, and while the objective of transparency in general speaks for itself, it needs nevertheless to be recognised that there are also limits to transparency. Such limits may come into play whenever reasoned judgement is used to decide on the approach for determining a particular parameter (e.g. length of the time series of observations to be averaged to obtain an indicator for the risk-free interest rate, composition of the peer group, etc.). Supposedly, transparency is thought to be provided by giving the reasons for the particular decision taken. Nevertheless, some stakeholders might require for the sake of transparency that DBA also provide detailed reasons for not using a different length of the time series or a different composition of the peer group, each of which may be the preferred one by one or another of the stakeholders. Such discussions can go on ad infinitum. At some point, DBA must be able to refer to sound judgement for having taken a particular decision. The point is that DBA must not only be required to base its findings on objective criteria, that of course are laid open transparently, but also be allowed, when pushed, to insist on its ability to use judgement for making such decisions. This should in some way be made clear in the document stating the principles that are to underlie the WACC determining process.

7 Review of DBA WACC Comments on "Preliminary results" The table below is the reproduction of Table 1 of DBA's draft report showing the preliminary results due to the revision of the WACC. Parameter Mobile Fixed-line Technology Updated Risk-free rate 1.93% 1.93% All Annually Credit risk premium 1.97% 1.97% All Annually Gearing 41.59% 41.59% All Annually Equity risk premium 5.60% 5.60% All No Beta (levered) All Annually Tax rate 22.00% 22.00% All Annually NGA premium % Fibre, excluding DONG Inflation 1.77% - Mobile Annually No WACC after tax 4.69% 4.69% WACC before tax 6.01% 6.01% Fixed-line Annually including NGA premium Real WACC before tax 4.17% % Fibre, excluding DONG Annually These preliminary results come out of the equations for the WACC, which (in the notation of WIK) are as follows: WWWW aa = (1 g) r e + g r d (1 t) where WACC at = the WACC after tax, r e = the required rate of return on equity, r d = the interest rate on debt, g = gearing (the share of debt in total capital) and t = corporate tax rate. The variable r e in this equation is considered to be determined by the Capital Asset Pricing Model represented by the equation r e = r f + β e where r f = the risk-free interest rate, β = the equity beta and e = the equity risk premium. The variable r d is in turn the result of the equation r d = r f + c where c = corporate risk premium. In above table, the inflation rate also appears. It is used to determine the real WACC, given that this form of the WACC is required by one of DBA's cost models. The formula for arriving at the real WACC is WWWW rrrr = (1+WWWW nnnnnnn) 1 (1+i)

8 4 Review of DBA WACC where WACC nominal = the nominal WACC, i = the rate of inflation. The value for the real WACC shown in the above table is for the WACC before tax, which is arrived at by dividing the WACC after tax by the factor (1-t). The above model of the WACC, meaning the formal apparatus, is standard and will not be commented on any further. DBA, however, departs from this standard model when determining the WACC to be used when the cost of services provided by mobile and NGA networks is calculated. In these cases an extra risk premium is to be added to the real WACC before tax. This departure can be represented by the following equation: WWWW rrrr,bb,+rr = WWWW rrrr,bb + rr where rp = the risk premium for either mobile or NGA activities. The focus in the following will be on the methods for determining the values of each of the parameters entering into the standard model. There will also be a critical review regarding the departure from the standard version of the WACC model for taking into account the special risk of mobile and NGA activities. Since the focus is on methods, there will be no evaluation of the particular preliminary results shown in above table. Below we list the parameters involved in this review: r f = risk-free interest rate e = equity risk premium β = equity beta c = credit risk premium g = gearing t = corporate tax rate rp = risk premium on mobile or NGA activities i = rate of inflation While we carry out the reviews in this order, the greatest emphasis will be on the methods for determining the values for the beta (β), the risk-free interest rate (r f ), the gearing (g) and the extra risk premium (rp). There is less reason to provide comments on the credit risk premium (c) and the equity risk premium (e) and least on the corporate tax rate (t) and inflation (i).

9 Review of DBA WACC Comments on "Basic principles for the revision" DBA lists here the principles on which the revised method is to be based. We take these up one by one, each time with a comment on our part: Most up-to-date publicly available data to be used: Whenever it is important that most recent data be used, this principle is obvious. The requirement may, however, divert attention from the fact that, when averages over longer time series of data are used to get an indicator for the value of a parameter, any single observation, whether most recent or not, has only a relatively small weight in the average. In this case it is most important to have the averaging period appropriately specified, in which of course also the most recent available data points should be included. Consistency: This is an important requirement. To quote the DBA draft report, giving high priority to this principle "means that considerations which apply to the calculation of one parameter should also apply to the calculation of another." We would suggest to add that the consistency principle should in particular also be used to assure that as far as possible the parameter values are selected in a way to comply with the forward-looking principle. (But see on this a comment further below.) Use of information from a peer group, members of which should be as similar in important aspects as the operators regulated by DBA: We feel that a shorter and less restrictive wording like "Use of information from a suitably selected peer group" would be more appropriate, giving in particular DBA more room to use its judgement what in each case the peer group should be. International perspective: By stating that most of the parameter values will be based on information from an international peer group, the international perspective is already indicated. So emphasising the international perspective here does not add anything new. Actually, as will be noted later in this report, while DBA is pointing out here that investors have an international perspective, when selecting the type of government bond whose yield is to serve as an indicator for the risk-free rate, it uses Danish government bonds, whose yield would not necessarily be the reference for the risk-free rate of such an international investor. Same calculation period for mutually dependent variables: We suggest to soften this requirement with the added wording "wherever possible". For example, the equity risk premium (ERP) and the risk-free interest rate are mutually dependent, since the ERP is the difference between the total market return and the risk-free interest rate. Yet, the time series of observations used for determining the value for the ERP is usually very long, whereas the time series used for determining the risk free interest rate, as separately to be used in the WACC equation, is normally substantially shorter. (We take up this point in greater detail in the next section when dealing with risk-free interest rate.) Combined WACC calculations for the mobile and the fixed-line network: We believe that the parameter values may be the same, irrespective of whether the corresponding WACC is applied to a mobile or fixed-line network, whenever there are no obvious reasons for differences. We do not, however, agree with the

10 6 Review of DBA WACC statement by DBA that this should also apply to the value of the beta. We disagree for two reasons. First, DBA argues later in its report that mobile operations are riskier than fixed-line operations; such differences in risk should be expressed through different values of the beta. Second, there exist methods (and studies using them) which make it possible to estimate beta values separately for mobile and fixed-line operations, from which follows that DBA's argument that one cannot determine such separate beta values does not appear to be valid. The principle that the WACC value should be forward looking, mentioned already in the preceding section, is not included in above list. We feel that, while it is referred to in the initial remarks section of the DBA report, it should also be explicitly included as part of the basic principles. Further, that the outcome from the deliberations on the WACC will necessarily depend on DBA's ability to make sound judgements, might also find a mention here. This last point would be consistent with the recommendation in the Brattle Group report, according to which the Commission should not overly be prescriptive in respect of the application of its guideline.

11 Review of DBA WACC 7 3 Review of the methods for setting parameter values 3.1 Risk-free interest rate DBA proposes to determine the value of risk-free interest rate on the basis of yields of Danish government bonds with a maturity close to ten years, using a six-year averaging period. To this estimate it makes adjustments to compensate for the procurement programme of the European Central Bank (CB) and for the fact that the ERP is derived from historical data using a twenty instead of a ten years' term. We will discuss these choices farther below, but first take up an inconsistency in current conventional approaches to the determination of the WACC. The risk-free interest rate together with a risk premium are the principal parameters determining the two components entering into the WACC, i.e. the rate of return on equity and the interest rate on debt. Now, the return on equity equals the sum of the risk-free interest rate plus the equity risk premium (ERP) (weighted by the beta), while the interest rate on debt equals the risk-free interest rate (plus the credit risk premium). An important point in this is that, by construction, the ERP equals the total market return (TMR) minus the risk-free interest rate. Since the ERP is considered a parameter with a stable long-run value, it is determined on the basis of the long-run difference between the TMR and the risk-free interest rate. Actually, for these calculations observations are used that may run over years starting as far back as the beginning of the 20 th century. It is then consistent to use long-run values for both the ERP and this is important the risk-free interest rate, when it is the task of determining a value of the WACC that in turn is applied in costing exercises with a long-run perspective. If, in contrast, one were to take a short run-perspective, there would be no stability of the value of the ERP, as it tends to fluctuate according to macro-economic conditions, especially showing a negative correlation with the risk-free interest rate, which means that when the risk-free interest rate is low the ERP is higher than its long-run value and vice versa. 2 This means that if one were to use a short-run-oriented value of the risk-free interest rate one should also use a short-run-oriented value of the ERP. Now, it is common in current WACC determining exercises that the value of the risk-free interest rate is based on a short averaging period, while for the ERP the value is based on the long-run perspective. For example, in the Brattle Group report it is recommended that the risk-free interest rate be based on values averaged over a twelve-month period and the ERP be based on historical estimates with observations covering up to 100 years. There looms then a risk of inconsistency. This is so because a short-run estimate of the risk-free interest rate is combined with a long-run estimate of the ERP, to determine a long-run WACC. The rate of interest on debt is made dependent on the short-run value of the risk-free rate, while the return on equity is made dependent on both the long-run ERP and this short-run risk-free rate. The relevance of this 2 For a recent study making this point, see S. Kihm, A. Satchwell, P. Cappers, The Effects of Rising Interest Rates on Electric Utility Stock Prices: Regulatory Considerations and Approaches, Lawrence Berkeley National Laboratory, December 2015.

12 8 Review of DBA WACC inconsistency must be judged by considering that such a WACC is used in bottom-up cost models, in which the WACC enters into forward-looking LRIC calculations that imply that, say, the revenue requirements of 20 years hence be consistent with this WACC. DBA appears to be aware of this risk of inconsistency by proposing a longer averaging period for determining the level of the risk-free interest rate, i.e. either based on the peer group operators' average refinancing period, or on an average of periods for an appropriately defined financial cycle. We agree with DBA in that we also recommend the use of an averaging period reflecting recent business cycles. Given that the current cycle appears to be a relatively long one, with continuing depressed levels of interest rates, we suggest that even a longer averaging period be considered, one that covers two business cycles and may include 12 years of observations. This may not completely eliminate the inconsistencies pointed out, but might do so to a substantial degree. In partly recognising the arguments for a short averaging period (but really doing only partly so), using a shorter averaging period than used for calculating the ERP may be justified by the fact that companies to some extent may be able to take into considerations short-run financial conditions when making long-run decisions. We would not object to DBA compensating for the effect of the ECB's procurement programme by applying an adjustment. However, provided the two-business-cycle-long averaging period is used, the programme may be considered an extraneous factor with a downward impact which, possibly, is compensated by extraneous factors biasing the results in the other direction during the longer averaging period. In other words, there is always noise in the observations which, however, it can be hoped cancel out if the averaging period is long enough. We also agree that the estimate be increased by 40 basis points to compensate for the fact that the estimate for the ERP is based on yields of government bonds with a term of twenty instead of ten years. We would question the use of only the Danish government bonds' yield as an indicator for the risk-free interest rate. Two of the operators in Denmark, Telia and Telenor, come from neighbouring Scandinavian countries. This would suggest that yields from the government bonds of these countries also be taken into account. As for the fourth operator, Hi3G, belonging to a world-wide operating company, it would be difficult to know the appropriate indicator for the risk-free interest rate that investors in this company would refer to; therefore for want of better information the same indicators as for the three Scandinavian companies may be used. 3.2 Equity risk premium (ERP) DBA proposes two sources for determining the value of the ERP to be used when calculating the return on equity: Estimates based on historical observations from the peer group countries, carried out by Damadoran a renowned practitioner in the field.

13 Review of DBA WACC 9 Two Surveys carried out by the Danish accounting firm PWC and by Fernandez, Ortiz and Acín, a research team at the IESE Business School at the University of Navarra. 3 We broadly agree with this approach. While the standard approach in our view is the one based on historical data, using survey information allows to make use of additional information on the ERP that is more forward-looking. The requirement should be that the geographical coverage of the surveys corresponds with that of the peer group countries. This requirement is met by the survey from Fernandez et al, it is not clear to us whether it is met by the PWC survey. When scanning the Danish language reference mentioned by DBA, we did not observe any table that provides information about country specific ERP estimates. Use of survey data is criticised by The Brattle Group for being unreliable. We believe that if a survey is carried out carefully, including among their respondents a broad spectrum of experts from academia as well as professionals from the financial sector, the results may usefully round up the estimates based on historical data. From our reading of the Fernandez et al. survey, we have the impression that this survey meets this criterion. We are not sure that the criterion is met to the same degree by the PWC survey, as according to the description in the DBA report its responses come from finance departments at investment banks, consultants and institutional investors, while there is no mention of experts from academia. Irrespective of these caveats, it may be expected that all these groups of respondents use to a large extent the generally accepted standard approaches to the estimation of the ERP value, so that the unreliability feared by The Brattle Group and others may not be a real concern. In any case, the values from the two surveys as shown in DBA's report do not give any concrete reason for suspecting such unreliability. We note that the 2014 report from Fernandez et al. has been used by DBA. There exists a more recent version of their report, dated 16 May Equity beta The beta coefficient is an indicator for the relative risk of a firm with respect to the overall market. It measures the systematic, i.e. non-diversifiable, risk of an operator and can empirically be derived from the analysis of the market performance of comparable firms. In its draft report DBA proposes a revision of beta accordingly: The DBA calculates beta based on daily observations for the peer group companies. The calculation is made over a six-year period and is compared to the MSCI Europe Index as a reference index. An unlevered beta is calculated for all peer group companies and in this light, an average unlevered beta coefficient of 0.41 is calculated. 3 P. Fernandez, P. Linares, I. Fernandez Acín, Market Risk Premium used in 88 countries in 2014: a survey with 8,228 answers, 20 June 2014; PWC, Værdiansættelse af virksomheder: Sådan fastlægges afkastkravet i praksis, (Valuation of companies: how to determine the required rate of return in practice), February 2016.

14 10 Review of DBA WACC The unlevered beta is subsequently converted to a levered beta using Harris & Pringle s formula. In this, the average gearing for all peer group companies of 41.59% is used and a levered beta coefficient of 0.70 is calculated. In our view DBA s approach is reasonable and valid. However, we see some additional points for further revision. The empirical determination of the beta coefficient depends crucially, first, on the underlying peer group, second, on the analysed time horizon and, third, on the individual and average gearing ratios of the firms. In this section the focus is primarily directed on the first two points, since gearing, in particular also as its role in the determination of the beta value, is discussed in section Peer group The main reason for using a peer group to determine the beta coefficient is to increase the number of reference points by taking into account the information of comparable firms in order to have wide-ranging information about the market. If only taking into account one single firm for estimating the beta, the risk of determining an unreliable beta value due measurement errors is much higher. Accordingly, the question is not whether to build the analysis on a peer group of firms but rather on how to build a reliable sample of firms that should be taken into account. Generally, we agree with the basic principles for the group composition used by DBA after which the telecom companies included in the peer group should: provide the same type of products and services, have the same level of operational gearing and cost structure wherever possible, and operate under fairly comparable market conditions. We think, however, that the rules for the selection process for the peer group should be more comprehensive. Starting with a list of firms including all operators that might be relevant, the filter criteria such as data availability, focus on telecommunication services etc. should be described in detail. It should become clear why specific operators are in the sample and others are not. Currently, there is a Swiss operator in the sample but no Austrian one; Swedish and Finish operators are included, while the biggest Norwegian operator Telenor is missing. We are not sure whether the analysis should be restricted to only northern and central European countries. It may also be appropriate to have operators from other European countries such as from Italy or Spain in the peer group. The whole process should end up in the final peer group for the estimations. Overall, the rules for excluding companies should not be too restrictive, from which we would expect that the complete peer group list includes more than the prevailing ten operators. Another important issue regarding the beta coefficient and the peer group is how to address different risk levels of fixed-line and mobile operators. DBA states in the draft 4 As gearing enters the WACC calculation in multiple ways, we fully agree with DBA s view that it should be applied throughout in a consistent way.

15 Review of DBA WACC 11 report that the Danish mobile market is more exposed to competition than the fixed-line market as there are four mobile network operators active in the market implying a higher degree of infrastructure competition. Consequently, risks on the mobile market seem to be higher. This is the reason, why a risk premium is foreseen for mobile network operators, which is added on the final WACC. We do not object to recognizing a higher risk of mobile operators, but for this it ought to be expressed by a higher beta value, since this is the risk indicator in the WACC-CAPM model. While some European regulators apply different beta values for mobile, as e.g. Norway, other regulators do not see different risk profiles for mobile and fixed-network operators, as e.g. Germany. Although in theory it might sound rather straightforward to set different betas for mobile and fixed-network operators, in practice it may become rather difficult to estimate different beta values, as there are hardly any pure play mobile operators nor fixed-line operators. 5 Instead we find primarily integrated operators especially in Europe. Also the operators in Denmark have integrated operations. The question is then what approach to use to determine by how much the mobile beta should exceed the fixed-line beta. One feasible way to derive a beta mark-up for mobile operators is to use econometric regression techniques to analyse, whether a higher share in mobile operations (expressed as a value between 0 and 1) is correlated with a higher beta value. The coefficient of this share in mobile operations would indicate by how much the beta increases when the share in mobile operations becomes larger (in case of a pure mobile operator with a share of 1, the increase would be equal to the value of the coefficient; in case of a share between 0 and 1, the increase would be the result from multiplying the coefficient with this share). For this result to hold it would have to be true that the coefficient is positive and statistically significant. 6 If these two conditions were not met, it would be highly questionable whether a mark-up on the beta for mobile operations is justified. The quality of the results can be increased substantially if it is not only based on a single cross sectional set of data but on panel data including the estimated betas of the same operators for more than one year. The panel structure of the data set allows to increase the number of observations on the one hand and to control for firm-fixed and time-fixed effects on the other hand. If for example the peer group consists of 10 operators (like at present) and if on average data on the individual betas and mobile shares of each operator are available for 6 years, the underlying number of observations would be 60 instead of 10 and the statistical reliability of the results would increase; if the number in the peer group became larger (as we have suggested might be the case) and if the number of years became larger as well (as we are going to suggest below) the reliability of the results would correspondingly increase even more. It should be noted in the context that panel data regression models can also be applied 5 See The Brattle Group, Review of approaches to estimate a reasonable rate of return for investments in telecoms networks in regulatory proceedings and options for EU harmonization, We are aware that this kind of analysis is not sufficient to detect possible causal relation between parameters from a scientific perspective. Nevertheless, it provides important and easy achievable insights on the broad links.

16 12 Review of DBA WACC if the data set is not complete for a specific operator. A so called unbalanced panel also provides reliable results in case there are not too many missing observations. As already mentioned, panel data regression models allow to control of firm-fixed effects, i.e. for non-observable and time-invariant heterogeneity, as for example specific characteristics such as firm size, historical peculiarities or the political and regulatory environment. 7 Moreover, time dummies 8 for the analysed years should be included as additional explanatory variables to control for year-specific effects. If for example beta mark-ups for mobile operators would decrease over time, this would be thereby taken into account. Panel data regressions offer different advantages. First, they increase the degrees of freedom of the estimation with the result of more efficient estimators. Second, they reduce the omitted variable bias, which is likely to occur in case of only one explanatory variable. Third, they allow to take the time dimension into account. Thus, regressions using panel data lead to better and more robust results than regressions using cross sectional data only. Analysed time horizon The underlying time horizon of the estimations also affects the levels of the beta values. In conventional estimation procedure, two opposing effects regarding the length of the period come into play, leading to a trade-off as stated in the Brattle Group report. On the one hand the beta estimates of the operators are highly volatile, all the more if based on daily observations. This means that it has an effect on the beta value if the beta of a firm is calculated at different points in time. In order to deal with this kind of volatility, beta value estimations of the operators are typically averaged over time. In other words, a longer period implies more observations, which reduces the statistical error of the estimate. Following this line of reasoning, longer periods imply more robust and stable results. On the other hand a new day of observations has the same impact on the results as the last day that is part of the analysed period. By taking a longer period into account, the weight of every daily observation reduces and at the same time the oldest incorporated observation is further away from the present. Assuming that more recent stock market information describes the actual situation of a firm more precisely, a shorter averaging period implies a higher (forward-looking) quality of the data analysed and, thus, a better prediction of the future. These opposing effects indicate that there is no correct length of the analysed period. While the Brattle Group report favours a rather short period of only two years, DBA proposes a six year period in order to be consistent with the calculation period of the 7 From a theoretical perspective, the quality of the results can be raised even further if it is controlled for time-variant heterogeneity by including additional explanatory variables. In the practical application, however, it may be difficult to get data for the corresponding parameters and, in addition, this would make the analysis more complicated. 8 A time dummy for a specific year is a variable that is set to one for the corresponding year and equals zero for all other years that are part of the analysis.

17 Review of DBA WACC 13 risk-free rate. In our view, there exists a strong argument to extend the period up to twelve years, which dominates the other arguments. The reason for proposing such a long period is that we are currently in a financial cycle that started in 2008 and is not complete yet. As pointed out in section 3.1, a calculation period of 12 years would enclose a full investment cycle and not only parts of one, which is crucial in our view. Therefore, we suggest to extend the averaging period to 12 years, which is also consistent with the suggested period for the risk-free rate. 3.4 Credit risk premium DBA's principal decision is to fix the value of the credit risk premium based on the peer group companies' credit rating. We agree with this approach. DBA further proposes to apply a 10 % downward adjustment to the resulting value, given that the wholesale price-regulated part of the operators' business is exposed to a lesser risk than the whole. Although this appears to be an ad-hoc adjustment, we would not object to it, as such a judgement falls within a regulator's margins of discretion. As regards the details of the derivation, DBA first determines the long-term credit ratings of the peer group companies using both Moody's and S&P. DBA then refers to two sources, Damodaran and Thomson Reuters, that relate the credit risk premium to credit rating. We would suggest that DBA prefer the information from the Damodaran source, given that (a) we have more confidence in an academic source, assuming that in this case a greater level of analysis has entered the derivation, and (b) the source for Thomson Reuters does not appear to be available any more. 9 When discussing in the following section the determination of the value of gearing, we also recommend an approach that is based on the peer group companies' credit ratings. 3.5 Gearing Gearing, being the share of debt in total capital, is used (a) to transform the asset beta into the equity beta, and (b) to define the weights with which the return on equity and the interest on debt enter into the equation for the WACC. Gearing is thus a parameter impacting on the value of the required return on equity (via determining the value of the beta) and impacting on the final value of the WACC (via determining the shares with which this return on equity and the interest on debt enter into the equation defining it). In order that both, the return on equity and interest on debt, are weighted in the WACC equation in proportion to the demands made on the company by equity and debt holders, the level of gearing used should reflect the proper share that debt has - or ought to have - in the total value of the operator, and the value of the factor "one minus the level of gearing" should reflect the proper share that equity has or ought to have 9 A. Damodaran, Ratings, Interest Coverage Ratios and Default Spread, January 2016, to be found under The source for Thomson Reuters mentioned in the DBA report could not be located; the message was: "BondsOnline has closed its business".

18 14 Review of DBA WACC in the total value of the operator. To accomplish this there are two approaches that may serve the purpose of obtaining a reasonable estimate: Use of market values of the peer group companies for both debt and equity, first, to make up total value of the capital and, second, to determine the shares of equity and debt in this total, where the latter of course equals gearing. Use of the level of gearing that corresponds to the peer group companies' credit rating. This would probably be the most objective and unbiased approach. For its derivation of the level of gearing, DBA uses book values, taken from the peer group companies' financial statements, for both equity and debt, thereby not adhering to either of the two rationales developed above. DBA argues that, due to share price fluctuations caused by extraneous factors, using market values would introduce instability into the values from one period to the other. Using instead book values would make for a stable trend over time and make that gearing be based on objective and transparently obtained values. Since it deems to have with the use of book values a sound approach with no problems of data availability, it rejects the approach based on credit rating. This one would be appropriate only if there were not sufficient information which, however, were not the case given all the information on book values. We do not agree with DBA's approach on the grounds developed in the first paragraph of the section. The argument that book values would provide stable gearing levels is not a sufficient reason for using this approach. We would not object to using estimates based on market values of equity (we would accept determining the value of debt by book values, as here the difference is not likely to be large). As regards the feared instability, given that a rather large peer group is used (it counts at present 10 companies, and if our recommendation is followed, it would probably be larger) and given that observations from several years are used, having a fraction of these change from one period to the other may actually not bring that much instability to the estimate based on an average of values. The approach that we find to be most objective and that we therefore recommend, is to determine the gearing level on the basis of the levels that hold for the peer group companies in relation to their credit ratings. As mentioned, DBA considered this approach but rejected it because it regarded the one based on book values as more appropriate. We recommend it because it would be consistent with DBA's approach to determining the credit risk premium. In fact, by deriving these two parameter values in relation to the same factor, i.e. the credit rating, it is assured that these two values are mutually consistent. This follows, since the credit rating depends in general on the level of gearing and the gearing depends on the credit rating. Thus, the same as we support the credit rating approach for setting the value of the credit risk premium in Section 3.4, we also strongly support it here for determining the level of gearing. Also, the approach would be unlikely to lead to unjustified unstable gearing values from one period to the other, assuming that the companies in question will not have large swings in their credit ratings; if instead there were such swings, this would then indeed justify changes in the level of gearing.

19 Review of DBA WACC 15 Similar to the approach regarding the credit risk premium, relevant data from companies in the economy at large should be used. These data should show the relationship between companies' ratings on the one hand and their capital structure on the other. From these data and the actual ratings of the companies under review, the corresponding levels of gearing could then be derived. The data to be used would be similar to that in Table 6 in the DBA's report, reproduced below, showing the relationship, as they hold in the economy at large, between credit rating and credit risk premium. In a table to be used for the present purpose, the second column would show the levels of gearing belonging to each level of credit rating. Rating Credit risk premium Aaa/AAA 0.75% Aa2/AA 1.00% A1/A+ 1.10% A2/A 1.25% A3/A- 1.75% Baa2/BBB 2.25% Ba1/BB+ 3.25% Ba2/BB 4.25% B1/B+ 5.50% B2/B 6.50% B3/B- 7.50% Caa/CCC 9.00% Ca2/CC 12.00% C2/C 16.00% D2/D 20.00% The difference would be that the right column would show the gearing levels in lieu of the credit risk premium. As regards the source for a corresponding set of data, where the gearing level is expressed in terms of market values, we suspect that Damodaran could be such a source. In the case of the credit risk premium, DBA relies on information from two US sources, one among them Damodaran, so that relying on US data should also in this case not be an obstacle. We would further recommend that (a) the average of gearing values from the peer group companies be used for fixing the gearing level of the Danish regulated operators, and (b) each peer group company's individual gearing level be used to convert this company's equity beta value into the corresponding asset beta value. As mentioned, gearing levels based on credit ratings would most objectively reflect the degree of financial risk that this conversion is supposed to take into account.

20 16 Review of DBA WACC 3.6 Corporate tax rate DBA proposes to use the current Danish nominal corporate tax rate to determine the pre-tax value of the WACC. We fully agree with this choice and have no further comment on it. 3.7 Risk premium for NGA After reviewing opinions on the matter voiced by the European Commission, The Brattle Group and industry stakeholders, DBA believes that a two percentage point premium should be added to the WACC, when applied to fibre networks outside the DONG area. In the following, we present the analytical arguments with which it could be established whether such a premium is justified or not. We will restrain from commenting on the geographical extent of networks ("outside the DONG area") to which the WACC including such a premium should apply. Various studies and reports indicate that risks associated with investments in fibre networks are substantially higher than investment risks in fixed-line networks using conventional technologies. 10 Having in mind that the WACC-CAPM model is designed to compensate for systematic risk only, 11 it is important to consider whether the risk associated with NGA networks investment belongs to the systematic or unsystematic category. Risks due to longer investment horizon as well as higher investment and capital requirements belong to the unsystematic risk category when investing in fibre networks instead of networks with conventional technologies; they are no valid arguments for a higher WACC. In contrast, reasons for an additional systematic risk for investments in fibre networks, such as a higher ratio of fixed to variable costs for the FTTH technologies in relation to conventional broadband technologies or higher demand uncertainty for FTTH services, should have an influence on the model and the consequent WACC results. As already mentioned in section 3.3 in the discussion on a possible higher WACC for mobile operators, we think that in the WACC-CAPM model, systematic risks should be accounted for by the beta value, as this is the parameter capturing investment risks in the model. An increased beta value for fibre networks would consequently lead to a WACC, which is higher than the WACC for fixed-line networks using conventional technologies. The big practical challenge for identifying the appropriate beta value is the creation of the peer group with pure play operators offering fibre networks only. As studies carried 10 See e.g. The Brattle Group, The WACC for KPN and FTTH, prepared for ACM, July 2015; CMT, Calculation of risk premium (in Spanish language), February See The Brattle Group, Review of approaches to estimate a reasonable rate of return for investments in telecom networks in regulatory proceedings and options for EU harmonization, prepared for the European Commission, 2016.

21 Review of DBA WACC 17 out for other regulators show (as for example in the Netherlands), 12 this is not possible on a solid basis. However, there exists another, so-called second best, approach of dealing with this problem. Using DCF analysis, it is possible to estimate a WACC uplift (risk premium) for fibre networks, which is added on the WACC estimated for the network using the conventional technologies. 13 The basic idea behind this approach is that two scenarios are compared, one for a fibre rollout and one for a rollout of a network with conventional technology and therefore less uncertainty. The difference between the two reflects the risk in question. In this case the risk premium would be derived from the difference in values of the internal rate of return obtained for the two scenarios. Accordingly, this approach provides a rough estimate of the increase in the WACC due to an implied increase in the beta reflecting the increase in the systematic risk associated with fibre network investments. In our opinion, it is important to contrast the resulting risk premium for fibre networks that is derived by the DCF approach with the appropriate beta value mark-up. This value is estimated by including the new WACC which is the WACC plus the risk premium in the model and by recalculating the corresponding hypothetical beta value. Applied to DBAs proposed risk premium of 2 % for fibre networks outside the DONG area, it would imply an increase of roughly 70 % in the levered beta value from 0.70 to This value seems reasonable and not out of range from an economic point of view. However, according to the discussion in its draft report, DBA has not derived the two percentage point risk premium from a DCF analysis or a comparable approach; it appears rather to have been arrived at on the basis of benchmarking. From a regulatory perspective, we would only be in favour of a higher WACC for investment in fibre networks if it is due to higher systematic risk. Higher systematic risk leads via the beta to an increase in the WACC. If a risk premium for unsystematic risks of fibre network investments is allowed for, this would not be in line with the WACC- CAPM model methodology. In this case the risk premium has no direct basis in regulatory principles. It would rather be an industrial policy instrument for supporting the rollout of fibre networks, which, however, may be justified in case that it is politically intended. 3.8 Inflation For determining the value of inflation, needed for converting the nominal to a real WACC, DBA uses a forecast of that value for the next ten years. To arrive at this forecast it relies on two sources, an economic survey by the Danish Ministry of Finance for 2017 and statements by the European Central Bank (ECB) regarding subsequent years. Since this parameter is needed to calculate a real WACC applied in a purely Danish context (in cost models for Danish telecommunications network operators), it 12 See e.g. The Brattle Group, The WACC for KPN and FTTH, prepared for ACM, July For further details see e.g. The Brattle Group, The WACC for KPN and FTTH, prepared for ACM, July 2015; CMT, Calculation of risk premium (in Spanish language), February 2013.

22 18 Review of DBA WACC makes sense to try to use inflation information regarding the Danish environment, which would justify taking the information from the Danish Ministry of Finance for Since there is no forecast available from this source for the other nine years, for which DBA seeks a forecast, it applies the ECB target for these years. We wonder whether it might not have been more consistent to use the ECB value also for Note that the value of the real WACC, arrived at by this estimate of inflation, is going to have an effect beyond the ten years implied by DBA in its report, this being due to the fact that the real WACC is used in DBA's cost model to determine the present value of revenues expected to be earned many more than ten years in the future. This in particular would be a reason not to include in the inflation estimate the single 2017 value from the Ministry of Finance but to rely wholly on the estimate from the ECB.