By Diane Coyle
What has happened to the price of telecommunications services in the past decade or so? It sounds like a straightforward question but our research indicates that the answer could be very different depending on some important choices made in constructing this index.
In the 2017 ESCoE Discussion Paper ‘A Comparison of Approaches to Deflating Telecoms Services Output‘ we explored why the official price index for telecoms had been almost flat during a period when the industry had experienced significant technological change including the advent of broadband data services, and business models, pricing and consumer behaviour also changing substantially.
While data services now represent the primary output of the telecommunications services sector, the existing output deflator used in the UK and elsewhere gives higher weight to traditional voice and text (SMS) services. Because the price of these traditional services is higher and demonstrated less change, using a deflator weighted towards these items implies slower growth in the real-terms output and productivity of the sector, which seems at odds with the considerable usage growth and experience of service improvements and motivated the consideration of alternatives.
One change we made in recalculating the price index then was adding in data services. We also, as a sort of thought experiment, looked at a unit value index: revenues per byte of data transferred across all the different services. Because, although the component services from text messages and voice calls to WhatsApp messages and video all involve different charges (higher per byte for the ‘traditional’ services), they all use bytes of data carried over the same networks. These two options suggested a price decline of between 37% and 97% over the years 2010 to 2017, rather than the roughly zero change in the published index.
In our new paper we have refined our approach further by taking more detailed account of the service providers’ pricing structures. For instance, fixed line services involve an access charge, currently treated as a separate charge when it would be more reasonable to allocate it across the services provided, as no-one just buys ‘access’. Bundling services has become common: Ofcom has estimated that nearly four fifths of services in the UK are ‘bundles’ of different components. Currently these are included in the index by using as weights the share of operators’ revenues for each component outside the bundles. We wanted to test what difference it would make to use the share of data volumes instead.
The aim of the exercise was to identify the key factors causing the wide range from our first paper of possible paths for the telecoms services price index; the new indices add in various factors which each in turn explains part of the difference, so bridging the wide gap, thankfully in a logically intuitive fashion: The greater the use of volume weights (ie share of data volumes, rather than revenues), the closer we get to the unit value approach. The challenge of deciding which is the appropriate one to use is a new manifestation of an old problem in constructing price indices, namely how to incorporate significant quality change or new goods in a rapidly innovating sector like telecoms.
The question is how to adequately control for quality change when there is a new or higher quality product, rapid volume growth and declining price substitute for an existing good or service. The challenge arises across the spectrum of digital services and has implications for the interpretation of the calculated deflators and thus real growth rates for such sectors. This issue may be important in the case of a number of digital goods, where bundling is becoming increasingly common. In the case of telecommunications services, the price trends differ greatly between OECD countries, although the technological advances are similar everywhere, suggesting statistical offices may be implementing a variety of approaches to the challenges we discuss.
One approach often recommended in the price index literature is to use hedonic adjustment for the changing quality characteristics. However, hedonic methods also have significant practical limitations that make them less suited for application in telecommunications services. For instance, many hedonic regressions for broadband use download and upload speeds as the main quality characteristics. However, these regressions rely on the high level tariff, rather than individual contract level data. This means using advertised, rather than actual, speeds since actual speeds can only be observed at the individual contract level. Further, whilst speeds are one of the main quality characteristics in telecommunications services, other factors are also important such as coverage and latency (the time-lag between sending and receiving, of little importance on a text or local call, potentially awkward on an international call, potentially catastrophic for a high frequency traders). Clearly the relative importance of these factors varies for individual consumers.
Over time, as consumers switch progressively from more costly traditional services to newer data-based services, the gap between revenue and volume weights and therefore between the different approaches to deflators will narrow. Our practical recommendation is that statistical offices should for now allocate fixed line access charges using volume weights, as revenue weights in this component only reflect accounting allocations rather than consumer choices, but should not apply volume weights to bundled charges for mobile services. However, the key point is to be aware of the sensitivity of the price index to the assumptions made about weights. This is not just a point about digital services. For example, around the world, prices data is produced using recent data on volume of sales to set the weights, but the recent disruption to sales patterns caused by Covid-19 won’t be relevant to future baskets of goods. NSIs around the world will, no doubt, develop answers to that challenge, but ultimately, it’s the same old story; its not just the prices, its also the weights that matter.
The full Discussion Paper can be found here.
Diane Coyle is Professor of Economics at the University of Cambridge, and an ONS Fellow.
ESCoE blogs are published to further debate. Any views expressed are solely those of the author(s) and so cannot be taken to represent those of the ESCoE, its partner institutions or the Office for National Statistics.
Monday, July 06, 2020