News

29.12.2020

No. 473: Bottom up cost modelling and statistical analysis of NGA network costs

No. 473: Bottom up cost modelling and statistical analysis of NGA network costs – clus-tering and invest calculations under consideration of already achieved rollout

Authors: Gabriele Kulenkampff, Martin Ockenfels, Konrad Zoz, Gonzalo Zuloaga

(full version only available in German)

Summary

The assessment of NGA-network costs’ is manifold. It is the objective of this paper to present three different methodological approaches towards the determination of NGA-access costs per subscriber and to give special emphasis on the determination of profitability of private investment and the requirement for state aid in order to achieve 100% coverage of broadband access by the means of FTTH P2P. A GIS-based bottom-up cost model calculation constitutes the basis for all three parts of the analysis (Greenfield approach). Besides FTTH P2P the modelling was also conducted for FTTS G.fast and FTTC Vectoring.

The cost analysis is based on 20 clusters, each cluster representing 5% of all German households, ranked according to density. The results show a relevant spread: the highest cluster represent costs per subscriber which – in comparison to the cheapest cluster – are scaled by the factor 3.2. The model results suggest profitability for 65% of all households under the assumption that the investor is able to contract 90% of the homes passed. This extreme assumption is taken in order assess a threshold with regard to the maximum reach of a profitable NGA rollout. Under these assumptions and in absence of price competition, but regional cross-subsidisation an investor at national scale could even profitably serve the whole country with FTTH P2P.

In order to interlink the analysis with the actual market situation and level of broadband coverage, we have analysed the data of the national broadband map as available by 04/2018 and matched it with the GIS data used for bottom-up modelling. As expected, we derived a lack of coverage with regard to FTTH which amounts to more than 90% of all households. Of course, CATV and FTTC show a much better coverage and require additional investment for approx. 35% of all households only. The graphical presentation of broadband availability and investment per subscriber on a map for Germany perfectly visualised that the existing access infrastructures apparently are located in the commercially most viable regions. Of course, CATV and FTTC cover more areas than FTTH due to their higher level of coverage.

In order to assess migration path related investment needs we established a further Brownfield approach, which focussed on potential savings due to the reusage of already existing infrastructure. In fact, these saving are very limited. Taking FTTC as a starting point for migrating towards FTTH P2P we have derived a maximum of 20% of savings in relation to 100% Greenfield investment in FTTH P2P. With FTTS as a starting point, a maximum of 30% was derived. In order to achieve these savings a 100% “re”-usage of feeder trenches is required.

In the second and third part of the study a statistical analysis was conducted. First, a regression model was established in order to overcome the heavy data requirements which a bottom-up modelling is facing. In a first step, we have derived a benchmark-model in order to estimate the regionally differentiated investment per subscriber on the basis of the complete data set available from the bottom-up model. In the following, this benchmark-model served as a template in order to derive proxy-variables which allow estimating the investment per subscriber in absence of data base stemming from the bottom-up model. Our analysis revealed that “households per developed km2” is considered to be superior over the plain proxy-variable “households per km2”. Unfortunately, we haven’t been able to establish a proxy regression model for estimating regionally differentiated subscriber costs on the basis of publicly available data. The proxy-models are able to explain more than 90% of the data variation in invest per subscriber, but did not pass the statistical tests for robustness.

A second statistical analysis dealt with the methodology for clustering. Taking the 20 clusters of the WIK NGA model as a starting point, comparisons were made with univariant and multivariant approaches. Furthermore, the (optimal) number of clusters was subject to analysis, resulting in the finding that a relevant reduction of the 20 clusters can be implemented without noticeably endangering the quality of results (especially with regard to univariant methods). With the application of a neighbouring concept a further cluster approach was analysed (Hot Spot – Cold Spot Analysis). This concept allowed to derive connected areas of remarkable size (exceeding NUTS 2) and can be considered as promising with regard to assessment of private investments’ profitability and funding requirements. Because the location of profitable access areas and their adjacency do matter with regard to business plans, we expect to be able to achieve a level of analysis that is even more realistic with regard to private investment decisions and need for state aid.

Discussion Paper is available for download.

To top  |  Print