Chapter 2 Challenges of Empirical Observation & Estimation
The highly globalized music industry generates two important international reports, as well as several national reports, but these are not suitable for the analysis of the typical or average rightsholder, nor for small labels and publishers who do not represent a large and internationally diversified portfolio of music works or recordings. Copyright and neighboring right revenues are collected in national jurisdictions. Because BUS artists are almost never constrained by their use of language, and the UK Music Industry is highly competitive in the global music markets, even relatively less known rightsholders earn revenues from dozens of national markets. The lack of market information on music sales volumes, prices for each jurisdiction, and the unaccounted for national, domestic, and foreign revenues makes the analysis of the rightholder’s earnings, or the economics of a certain distribution channel like music streaming or media platforms, impossible.
The highly globalized music industry generates two important international reports: the Global Collection Reports of CISAC for the author/publishing side of the industry, and the Global Music Report of the IFPI, which reports the recording (producer) side of the industry. These reports are not suitable for an economic market analysis, because they do not contain prices and quantities, only aggregated revenues. There are many national market studies are available, but only few of them make an effort to quote volume or use data and price data. CISAC and IFPI are global organizations, and their reports are based on an internal survey of their members. Most industry reports are member self-reported studies.
The Hungarian, Slovak, and Croatian reports were based on independent market surveys (Antal 2015, 2017b, 2019c, 2019a). They brought up many methodological challenges that a market study in the UK music industry must solve, too. Music organization usually do not possess the information that would be desirable for analysing the market from an economic point of view. Surveying rightholders and users is very difficult, though, because neither rightsholder nor works or recordings have an authoritative description of their population. Only very advanced, inverse sampling techniques (which require very large amounts of data) can reveal price and volume movements when the analyst has no access to full transactional logs.
This is the problem that the Digital Music Observatory, a working demo of the European Music Observatory is solving. It grew out of the Central & Eastern European Music Industry Databases (CEEMID) initiative in 2014 (Artisjus et al. 2014), in which rightsholders from three countries attempted to solve this problem. It was supposed to bring the seemingly data poor Central European countries to a level of data availability that allows better price setting or the creation of better creative industry policies or business strategies, but as experience showed, Western, more advanced, and future markets like Armenia share very similar problems. The idea of CEEMID and later the Digital Music Observatory was to collect meaningful data that allows an economic analysis or economic valuation, i.e., it contains volume and price information.
In 2019 Consolidated Independent (member of the state51 music group) teamed up with CEEMID to make a showcase for a permanent, harmonized, international data collection program. The Central European Music Industry Report was prepared with the help of 60 music organizations in 12 countries, including the United Kingdom, and analysed rightsholders earnings in various distribution channels, including, but not limited to streaming, in 20 markets. The idea of this observatory was brought to the UK policy debate on music streaming by the observatory’s only (former) British users, via the Written evidence submitted by The state51 Music Group to the Economics of music streaming review of the DCMS Committee (state51 Music Group 2020; Antal 2020).
“There are instructive initiatives in other industries in which there is perhaps a clearer and longer standing recognition of the role of economic analysis. This sometimes results in initiatives such as ‘Observatories’ like the European Market Observatory for Fisheries an Aquatorial Products or the European construction sector observatory […] These tend to be collaborative endeavours, with a varying mix of government, industry, economists and in some cases funding bodies. […] To date there have been few if any entities or initiatives for music similar to the above-mentioned observatories. We suggest this is something that policy makers can support and encourage, but which ultimately needs to be driven by the industry itself. […] This is one reason we have worked with the economist Daniel Antal and his team, in particular on the Central European Music Industry Report 2020. Economists such as Daniel Antal produce data about the music industry that is consistent with international statistical standards and adhere to rigorous data ethics principles, seeking external validation through data and code repositories for underlying data and methodologies.”
The data observatory concept is derived from Earth and natural sciences, where often many research stakeholders build large observation stations, such as the Hubble telescope in space, or CERN. Data observatories are often managed by a triangular stakeholder base of business, scientific, and policy stakeholders. The music industry requires a permanent market monitoring facility to win fights in competition tribunals, because it is increasingly disputing revenues with the world’s biggest data owners.
This was precisely the role of the former CEEMID program, that was initiated by a few collective management societies after a dropped GESAC project. Starting out from three relatively data-poor countries, where data pooling allowed rightsholders to increase revenues, the CEEMID data collection program was extended by 2019 to 12 countries. It was eventually transformed into the Demo Music Observatory in 2020 (Antal 2021a), which is now open for any national rightsholder, stakeholder organization or music research institute. In the UK, the Music Creators’ Earnings project, which created the Intellectual Property Offices’ Music Creator Earnings’ Earnings in the Digital Era relied on our data, particularly, with the kind approval of Consolidated Independent, the (re-)publication of our price and volume indexes.
Our working paper is divided into three subsequent chapters and conclusions.
There are very few public sources available for a real economic analysis of the music sector, particularly in an international comparison or in a time-wise, longitudinal comparison. The global music industry has two well-known global reports: The Global Music Report of IFPI, which covers total revenues in almost all territories for producers and partly performers; and the CISAC Global Collections Report that contains total revenues from collectively managed author’s revenues. These reports were designed to set business targets for larger organizations, and they do not contain price or volume data, only revenue data, which allows very limited economic analysis. The more fragmented live music industry does not have any comprehensive global or European report. We also do not have a truly comprehensive report on global publisher revenues that are not collectively managed.
Both organizations collect (rather different) international data which is only available to their members. Because CISAC had been earlier accused of price fixing, and made an agreement with the European Commission, the organization is particularly careful about even recording price data. IFPI has more comprehensive economic analysis, and in 2008 it even published a very useful pricing guide (PwC 2008). There are also national music industry reports available with variable depth of content and analysis, like the UK Music in Numbers series. These reports contain limited information for a thorough economic analysis or valuation, partly, because the conflicts of interests within the national music industries, for example, among publishers and producers, and producers and performers, do not allow the systematic collection and dissemination of such information.
2.1 Market definition
In competition policy, a relevant market is a market in which a particular product or service is sold—an intersection of a relevant product market and a relevant geographic market. This concept is particularly useful in our case, because music has no clear-cut product, service, or geographical market.
2.1.1 Relevant product: the analysis of the value chain
A relevant product market comprises all those products and/or services which are regarded as interchangeable or substitutable by the consumer by reason of the products’ characteristics, their prices, and their intended use—experience and research in other countries shows that music streaming can be substituted with other, differently licensed music, for example, listening to music via the media platform YouTube, or from mp3 files. Even the sale of physical music products, such as microcassettes and CDs, can be substituted with music services such as streaming subscriptions.
missing
CEEMID has been mapping data sources for composer, producer and performer royalties, and other data sources since 2014, based on the standard mapping technique developed originally in the United States (Hull et al. 2011) and later adopted to the EU (Leurdijk et al. 2012; Leurdijk and Ottilie 2012). To make specific—and confidential—analysis Artisjus and other collective rights holder representatives gave confidential data for this part of the calculation. For individually paid royalties, CEEMID used its own, 6th annual Music Professional Survey, which was conducted with 2065 musicians in twelve languages in 2019. CEEMID was turned into the Digital Music Observatory—a modern, largely automated data observatory following the planned structure of the European Music Observatory—in 2020.
Missing info
2.1.2 Relevant geographical market
A relevant geographic market comprises the area in which the firms concerned are involved in the supply of products or services and in which the conditions of competition are sufficiently homogeneous. Because recorded music creates royalties from copyrights and neighbouring rights defined by national copyright law, and, with some exceptions (for example, the BIEM agreements), are set nationally, the relevant market for analysis is the national market. British music is licensed in almost 200 jurisdictions, which makes the analysis of music creators’ earnings particularly difficult. Yet, there are very few national (or domestic) data available on music sales volumes and sales prices, particularly in streaming. And the empirical observation, either via surveys, self-reporting, or via the observation of various royalty statements is very difficult, because the population of rightsholders, works and recordings is unknown, and representative sampling is difficult.
While observing their global income is relatively easy, it tells us nothing about the economics of those earnings, i.e. the change of demand, supply, volumes, and prices.
Aggregated income indicators are usually made in two versions: national or domestic. The national music creators’ earning is the earning of rightsholders who live or are registered as legal persons in the United Kingdom. The domestic income of the music industry is the earning made by entities in the United Kingdom, including the earnings of non-resident rightsholders they represent. Because in most industry sources, the domestic and the national approach is not clearly delienated, it is very challenging to carry out a proper data analysis.
Aggregated income indicators are usually made in two versions: national or domestic. The national music creator earning is the earning of rightsholders who live or are registered as legal persons in the United Kingdom. The domestic income of the music industry is the earning made by entities in the United Kingdom, including the earnings of non-resident rightsholders they represent. The earnings of a French artist residing in the United Kingdom, represented by a French label, is likely to be counted in the British national earnings, but not in the domestic earnings of the UK music industry. However, a Dutch artist residing in the Netherlands but represented by a UK label is likely reported in the domestic revenue of the UK music industry (and add value to the UK GDP.)
Defining the geographical scope of the analysis is relatively easy in the case of live music, and rather complicated in the case of recorded music. Because copyrights (and neighboring rights) are nationally granted on the basis of international treaties, rightsholders—via distributors or collective management socieites—collect their revenues from each copyright jurisdiction separately. However, this information is often lost in surveys and global reports. In some cases, tying the revenues to a country, or even a region of a country is possible—this maybe possible in the case of radio revenues or royalties connected to live performances.
The loss of information, as we will show in the next chapter, is critical, because volumes, prices and exchange rates move rather differently from the UK market in other countries. German revenues are driven by different volume growth rates, different local price fluctuations in the value of a stream, and by the EUR/GBR rate applied whenever a monthly or accrued royalty is paid. Because practically all stakeholders have non-UK revenues, without knowing how much they are affected by different international factors, they cannot know what steps in Britain can lead to a positive change in their visibility or income goals.
The UK music industry is not constraint by language barriers, and most rightsholders receive income from dozens of territories or jurisdictions. This means that British rightsholders collect from many jurisdictions, subject to various definitions of equitable remuneration, public performance, mechanical royalty regulations, and in many currencies. Because of the significant differences in streaming prices across the world, and the presence of currency rate fluctuations, a very large part, if not most of the income differences of rightsholders are caused by the different geography of their audience distribution.
The Digital Music Observatory uses data from rightsholders directly, from collective management and distributors. It is not always possible to break down the figures to national territories—for example, in the case of musician surveys, it would not be possible to ask musicians to break up their royalty statements when they self-report revenues.
This is an important problem, because each artist, label, may have a different share of British, Irisih, American, German, and other income, and therefore their reported income may contain an unknown element of foreign income, and an unknown currency conversion effect. For example, in the 2016-2020 period the British pound generally lost value against the euro, so a label or artists with flat revenues partly arriving from the eurozone could have seen a rising pound revenue.
The CEEMID-CI indices are denominated in British pound, and each revenue is exactly tied to a jurisdiction. This cannot be said of self-reported surveys like that of the UK Music in Numbers, and we believe that there is no mechanism to prevent reports to IFPI to clearly delineate revenues per market. In the case of the CISAC Global Collection Report, we have a different problem: while we know that all reported income was generated in a particular national jurisdiciton, often a large share, even majority of those collections were made on behalf of foreign rightsholders.
2.2 Volumes
The most notable problem for any economic analysis of the music sector is the lack of volume information for the most important uses: broadcasting and retransmission, and various forms of licensed and UUC streaming (mainly YouTube.) The public music industry sources do not contain the number of exploitations (uses), the hours of exploitation (uses), or even the number of users.
Detailed volume data usually exists, though it is not always available from a central source. Live music is collectively the biggest part of the music industry, but it is very fragmented, and with the exception of a few small and developed national markets, there are no real central ticketing services and central points of ticketing information. However, because live performances in most cases exploit the music creator’s copyright (with the exception of early and classical music and authentic folk music that is not, or no longer subject to copyright protection), music performances are licensed by collective management organizations. There is always a certain level of latency — some shows go unreported —, but such agencies usually have detailed volume information (number of events, their audience volumes, ticket prices or revenues, and even the actual works used for royalty payment.)
Similarly, the uses of recorded music in most cases has full transactional data. Streaming providers pay by every single use, and mechanical licensing is based on the mechancical copies made (in the form of vinyl records, CDs, or downloadable files.) In some countries, broadcating and re-transmission has a full transactional log, in other cases, it has large use samples.
The Digital Music Observatory has been following the methodological guidlines of the former ESSNet-Culture statistical working group. These methods were synthetized from the best practices found by ESSNet-Culture, a working group set up by Eurostat and 15 member states’ national statistical offices to measure cultural and creative industries. The ESSNet-Culture working group recommends the measurement of cultural access and participation (including market- and non-market forms) on the basis of the ICET model (Bína, Vladimir et al. 2012, pp 237-239).
- Information: to seek, collect and spread information on culture;
- Communication and community: to interact with others on cultural issues and to participate in cultural networks;
- Enjoyment and expression: to enjoy exhibitions, art performances and other forms of cultural expression, to practice the arts for leisure, and to create online content;
- Transaction: to buy art and to buy or reserve tickets for shows.
The ICET model is based on a long history of quantitative sociology and media research which has almost 50 years of research history. It is a well-established methodology. For more details on the ICET model and conducted cultural access and participation surveys with it see (Haan and Adolfsen 2008; Haan and Broek 2012). Following the ICET model, we created surveys that had been measuring the missing variables in a way that they could be related to transactions. (The ICET abbreviation stands for information, communication, enjoyment and transactions.)
Rightsholders often have full or near-full transaction details for market-based cultural activities, such as sales of books or concert tickets. However, private copying is not a market-based activity, and there are no sales logs present. The problem with these data sources is that they are business confidential, fragmented, and they are lacking a common collection methodology. It is possible to integrate this data (we will turn to this problem shortly), but often it is more useful to collect more aggregated forms of data. To arrive to a common demoninator of users and uses, we have been calculating notional hours of music enjoyment in the forms of attending concerts, listening to radio or streaming services, or the respondents record collection. While self-reported surveys tend to be biased, they are consistently biased. They may overstate the actual hours of use, as people like to report higher cultural participatoin than they actually engage in, but such data is not biased when expressed as the market share of radio transmission and licensed music streaming.
2.3 Prices
Price information is always missing from the public music industry reports, and critically, in the absense of volume data, even price avarages cannot be calculated from total revenues. Of course, the average price woudl be only a starting point in cases where the prices are not uniform. As we will show, the prices are very stable in some forms, such as mechanical licensing (they are set by international agreements), somewhat variable in broadcasting and retransmission (they are changing annually) and they are fluctuating monthly in streaming.
Price comparison is very difficult in the music sector because various forms of exploitation of the underlying copyright or neighboring right (the “revenue streams”) have different licensing and contracting models. The prices of mechanical reproduction, public performance and broadcasting/transmission are based on annual use and annual payments. Because some earned royalties are very small, royalty management organizations do not pay out but accumulate earnings where money transfer and accounting costs would be larger or would take most of the payout.
The streaming licenses, which are usually make a form of a hybrid between a mechanical and public performance license, have a monthly payment schedule, but the revenues cannot be treated stricly monthly. The typical revenues do not meet the monthly minimal payout tresholds—in fact, the typical (median) earning has been zero in the Consolidated Independent portfolio that was used for the xxxxx, and we believe that a larger portfolio that has more not actively distributed or promoted recordings have even larger payment lags. This means that without further data processing — which can be a tedious task, as usually we have to deal with hundreds of millions of transactions stored in various royalty statement formats—, we cannot even correctly attribute the revenue to a financial year.
Price harmonization is a complicated task, and the the ?? Price Harmonization and Valuation section of this paper expands this topic.
2.3.1 Functional currency of the analysis
Music is a global industry, and even less known artists or songs find occasionally international audience on streaming platforms. This means that practically every British rightsholder has some revenues that are not originally denominated in British pounds. In the post-Brexit period, the British pound lost value against the US dollar and the euro, which means that a British artist could have received the same 100 GBP revenue even if her euro or dollar quantities or prices were falling.
For any meaningful economic analysis, the volume, price, and exchange rate effect must be separated.
Revenues are multiples of use volumes over price expressed in a certain currency. They can be national or domestic market indicators if they contain revenues only domiciled rightsholder or neutralized rightsholders’ revenues. Without the ability set up apart foreign versus national or domestic revenues, and break up the effect of changing volumes, prices and exchange rates, they are not suitable for economic analysis, and they do not indicate the earnings of the average or typical rightsholder. They still maybe useful for business target settings, but only for organizations with a large and diversified portfolio.
There are several problems with the existing national and international music industry reports:
They usually do not separate domestic, national and foreign income.
They do not consider the exchange rate, or use currency translations that may not be adequate for every type of analysis.
They do not separate the effect of price and volume change.
Sometimes they do not even control the number of territories (jurisdictions from where royalties are received.)
These problems give rise to the problem that growth in aggregates may lead to a diminishing revenue for smaller entities or individual artists. These indicators are not adequate for the use of small publishers, labels or managers of bands, artists, because they do not indicate growth or decline in volumes (and signal for changes in promotion), do not indicate growing or declining prices (and allow the focus of sales on more lucrative segments.) While currency risk can be hedged or otherwise insured, they do not even show which part of the earning change is attributed to exchange rate movements and which is related to the market performance of their catalogue.
Ideally, the data collection should record volumes (number of streams), prices (value of individual stream), and exchange rates (conversion rate applied for the particular stream) for all significant markets of the artist, company or national organization. Currency translations must be consistent and meaningful.
2.4 Estimation strategies
Almost all recorded industry transactions have full transactional detail (use, licensing data, price) and to a lesser extent, live music performances have rich transactional data. The problem with this data is that it usually transactional, and it is linked to various data that is connected to the protection of personal data (both rightsholder and often user data) and to business confidentiality.
2.4.1 Data integration approach
The problem with the transactional data of the live and recorded segments of the industry is that it is fragmented and protected by business confidentiality. Our approach in the Digital Music Industry had been since 2014 to locate these data assets, provide a metadata map for the potential use of the data, and eventually, when there is an interest from the data owners to use the data, integrate the data for analysis. This way we are avare of the location, structure of much data related to the UK music business, but in the absence of a mandate to use the data, we cannot make copies of this data and use them in analysis.
2.4.2 Representative surveys
The ESSNet-Culture methodology gives guidance on the application of the ICET model on creating music user surveys which can be analyzed jointly with transactional data. The relevant ICET surveys are the Cultural Access and Participation surveys, which capture both the market-oriented (paid) and non-market (liturgical, amateur, home copying, pirated) uses of music. Representative surveying of the audience is possible, but it is getting increasingly expensive as people are more and more reluctant to give face to face interviews. In our experience, a well-designed CAP survey is not well suited for CATI telephone interviews or online surveys. However, plenty of data can be reused via “survey recycling” or new surveys can be compared with “retrospective harmonization” with data from other countries or earlier surveys.
2.4.3 Balanced surveys
The population of music enterprises and music professionals is unknown. Microenterpries usually register under one NACE classification, but carry out many different task (a small enterpreneur may create recordings for both the music and the film industry, and work in the staging of theatrical prodcutions, and even be engaged in completely different economic activities.) This makes the traditional, business or personal demography based sampling impossible. In my experience, a good quality of rebalancing or post-stratification weighting is possible when there is some information available from an authoritative source. The best source for this is collectively managed income, which is avialable with full transactional detail. For the making of the first Hungarian music industry report, and the Slovak music industry report, we compared the self-reported copyright and neighboring right revenues of musicians with an anonymous, full payout distribution, and we made sure that sample populations self-reported income distribution correlates almost perfectly with the anonymous, full distribution. Generally, we omitted the 5-10 largest payouts, which are not representative for the distribuiton, but could be guessed on the basis of the popularity of the artist.
2.4.4 Survey design and harmonization
Because surveying musicians and rightsholders is very difficult, we use various techniques to improve our results. As mentioned earlier, we can use known payouts (such as various collectively managed income paid to living British rightsholders) to create post-stratification weights for responses. Another possibility that the Digital Music Observatory uses ex ante and ex post survey harmonization. Asking the same questions with the same methodology used in large, representative, national or Pan-European surveys allows us to understand the various demographic biases of musician surveys.
For example, in the Central European Music Industry Report, we used relative income and life satisfaction questions that have been asked for decades by Eurobarometer, and compared results. In some countries, we could create a representaitve sample of musicians; we could observe that everywhere musicians’ earnigs were more precarious than that of the local population. A major advantage provided by survey harmonization is that we can compare musicians’ earnings with various occupational groups, too.
Our suggested approach to the MCE project was the use of survey harmonization and post-stratification, but this was turned down by the Steering Group.
When we are surveying rightsholders, which I have done so far in 12 countries, we must keep in mind time and again that respondents are typically drawn from a very small minority of the rightsholders. Music careers are not always linear or exponentially growing, many rightsholders are inactive for years or decades when they often receive very little or no earnings, and do not participate in any surveys. Heirs almost always decline to participate in any music earnings survey. Statistically valid surveying is possible but challenging – in our surveys we use ex ante and ex post survey harmonization with national and international standardized surveys (which are available in the United Kingdom, too) and we compare surveyed, anonymously reported earnings with anonymised known earnings. For example, in the CEEMID surveys conducted in Hungary and Slovakia, we compared the anonymously reported income from Artisjus and SOZA with the anonymised, full payout of actual Artisjus and SOZA royalties to living rightsholders. Our successfully distributed surveys very accurately resembled the known, true payouts. Because the distribution of the earnings was very similar in 12 analysed countries, and in two we could very carefully compare the reported earning distribution with the true distribution, it is safe to assume that British payouts have a similar form.
In any empirical analysis, it is important to note that usually we can observe a minority of the works or recordings, and a minority of the rightsholders. If we observe actual streams or payouts, most of the works/recordings are not used in any given period, and the second largest group is so scarcely used that the payout is carried over to a next period. In any given period, most of the works/royalty payments are not observed; additionally, the end of the long tail is also not visible.
2.4.5 Timeframe of analysis
As we have experienced during the creation of the CI-CEEMID indices, most recordings’ typical (median) earning over a single royalty payment cycle (a calendar month) is zero. But it is certainly not zero during the entire lifespan of the recording, and it is usually not zero for the first or second year. The choice of the timeframe for empirical observation is critical.
Various licensed uses of music have three royatly payment cycles. Mechanical royalties are paid once in a lifetime. Public performance royalties are collected on an annual basis, and small earnings (that would be too costly to pay out during the year) are accumulated over two or more years. Streaming earnings are paid out monthly, but similarly to public performance royalties, small amounts are accumulated over two or more periods. Observing royalty statements alone cannot enable meaningful comparisons, because the same payouts on a given date–for example, 30 April 2020–refer to different earning periods.
Periodical earnings often do not reflect accrued but unpaid royalties, which may affect a very large number of rightsholders, given that the earnings in the long tail are smaller than the accounting and bank transfer cost. These accrued earnings are not always lost, and often carried over to the next period when the payment, after payment costs, is practical. It is very easy to significantly underestimate the payouts in the long tail, because their payment frequency is lower than among successful artists.
The different timeframes of various royalty collections has another impact on the analysis: different royalty cash flows are subject to different currency exchange rates. Two, seemingly equal royalty payments may veil a different quantity or price by an offsetting currency movement.
2.4.6 Sampling the unknown universe of works and recordings
Many people are familiar with the work of Sixto Rodriguez, a Detroit-born songwriter, whose works were not licensed for more than 30 years after they were recorded between 1967 and 1971, until 1997, when it was famously discovered that he is a best-selling songwriter in South Africa. After the documentary Searching for Sugarman won the Academy Prize, his recordings, 40 years after their release, became golden and platinum records in many countries.
Many artists are active in their twenties, and in the 21st century they can expect to live 60-70 years longer, and their heirs enjoy copyright protection for further decades. Like Rodríguez’s recordings, there may be many protected works and recordings that have not been observed in distribution schemes, media-, and streaming platforms, yet become successful in the future. In the absence of a global, compulsory registry of copyrights, we do not even know how many works and recordings enjoy copyright protection in 2021.
Because there is no global database of copyright (and neighbouring right) protected music works and recordings, we do not know precisely how many rightsholders and how many works or recordings are protected by the law. From a strict mathematical perspective the average (mean) and typical (median) rightsholder earnings do not exist, because their calculation would involve division by, or ranking of an unknown number of rightsholders.
Most rightsholders create works and recordings that have a limited lifespan: in a few years they lose their audience, and in the absence of their use, they no longer create income. But, as the case of Sixto Rodríguez shows, they may in the future, even decades, or a hundred years later. There are many rightsholders who create something, but never become full-time creators, or eventually they go back to part-time or occassional creators. Creating a representative survey of rightsholders is very challenging, becaues the rightsholder population has many inactive members.
Similarly to rightsholders, most of the works and recordings are not in use, or hardly in use at any given time period. Creating a representative sample of recordings or works for empirical observation is very difficult, because we simply do not know how many works or recordings exist.
Since 2014, we perfected sampling both rightsholder populations and recordings to create better and better representations of the creators and their creations. We created special surveys among music professionals to establish their typical earning levels and composition, and we created special samples of recordings to observe their typical revenues.
The way we sample or weigh musicians’ self-reported earnings in the Digital Music Observatory’s surveys is based on the known, but anonymised payouts of author’s societies, and comparing it to the self-reported earnings from author’s societies. The closer the self-reported median, quartile, or average payouts get to the know, but anonymous true payouts, the better the sample. In Hungary and Slovakia we were able to create highly representative music professional surveys for the active musician population. On active musician population we meant those living artists, who recieve some collectively managed royalties in a given year, i.e., their works are played on radio, television, directly identified public performance, collectively managed streaming services, on concerts, or other uses where author’s societies record use and collect revenues. We believe that this is the best target population for investigating music creator’s earnings. This target population is usually rather different from the payrolls of author’s societies, not to mention neighboring rights soceities, becasue those contain heirs of many deceised artists. This target population does not include new artist who do not have yet a registered work or commercially released sound recording, or whose works are not exploited in these channels. Generally, this is an appropriate target, though in some genres they may not be representative. (For example, old and classical musicians play music that has no author royalties, and some subgenres of hip-hop are not radio friendly and avoid any registration.)
The approach we used for the Central European Music Industry Report, which is so far the most comprehensive European music industry report, was a systematic sampling of a very large portfolio managed by Consolidated Independent (in the state51 music group.) With the help of state51 music group’s engineers, we pooled the royalty statements of several million recordings, and chose to examine further those which were used in one of the four services (Spotify, Deezer, Apple, YouTube) in 20 select markets, including the United Kingdom. We did not believe that the Consolidated Independent portfolio was representative of the entire European or each of its national markets. We used a sampling technique reminiscent of some bond and stock market indexes to select recordings that were good candidates to represent the earnigs of a typical recording in a given market and time period. Our ‘index basket’ chose the median use of more than 300,000 recordings for 20 distinct national copyright jurisdictions and months. The index basket was ‘rebalanced’ every month: songs that were used more or less than the median value were replaced with the median used (played) songs of the month. We wil introduce these indice in the next chapter.
Our indexing was deliberately experimental, and connected to the creation of the report, not the Music Creators’ Earning project. We believe that more sophisticated and even largely different sampling techniques could be used for understanding earnings in streaming channels. Our selection method tried to find the ‘typical’ song that was played in a given months, which is very different from the ‘typical’ recording. In any given month, in the CI portfolio, as in any large portfolio, the ‘typical’ recording was not played, and the median (typical) earning was zero. We avoided this problem with selecting every month the ‘typical played’ song, but if we would like to characterize the ‘typical’ rightsholder, we would have to give up on the use of both the arithmetic mean (which is cannot be calculated as we do not know the number of protected works/recording) and the median (because it non-descriptively zero.) We still believe that the CI-CEEMID index is the probably the best descriptive statistic that can characterize British and European individual rightholder earnings.
2.4.7 Big data samples & synthetic datasets
Various streaming services provide a more or less open API to their data, which is not suitable to harness a representative survey of uses, but nevertheless, provides a very large data access. For example, the SpotiyR R statistical software packge provides a programatic access to the Spotify Web API and allows the request of relatively large amounts of data, potentially tens of thousands of data points per day. Music distributors, or rightsholder organizations also have access to vast amounts of data.
Big data sources provide an unusually large amount of individually immaterial data. To translate this to far more limited but material, useful data, rather advanced statistical methods are needed. The Law of Large Numbers proves that we can create useful, representative statistical indicators from such source.
For analysing markets, this is not a specific problem, or rather opportunity, for the music industry or music streaming. The market of ar offers a university of investible financial products that is probably impossible to monitor entirely. All over the world, government and corporate entities are issuing bonds, common stocks, mortage papers, and various hybrid forms of equity or securtized loan products every day, often with very short maturities. Short maturity securities often have a life span of only a month. Financial service providers, like the Dow Jones Company, the Standard & Poors, have been creating statistically valid, representative samples (“baskets”) of securities to understand the representative price and volume metrics, and total market sizes (“capitalization”) for more than a hundred years.
Recorded music is entirely a copyright-based industry. It means that the value of phonogram is only related to the value of its copyright (publishing side) and neighboring right (recording side, which may be divided between producers and performers.) The value of the recordings is derived from the cash flows, i.e. royalty payments to both sides (or in some jurisdictions, where performers are entitled to separate remuneration, all three sides.) This makes the approach used by stock and bond indexes particularly useful for analysing the music industry. Recordings are similar to stock-like securities in the sense that they do not have a fixed income — in the music streaming services, the volume of their use and the price of their use fluctuates in every single copyright jursdiction every months. They are similar to bonds in the sense that they have a fixed lifespan. The term of the copyright and neighboring right protection eventually expires, though, in our experience, most recordings are similar to short-term fixed income assets because they only yield a material income in the first one or two years of their protected term.
We used an analogy with the securities market in the Central European Music Report, where we compared volumes and prices of many European markets, among others, with the United Kingdom as a mature market. This analysis was extended within the Music Creators’ Earnings project [Antal (2020); ]. The Digital Music Observatory is currently teaming up with statisticians to improve our sampling methodology.
Another approach, which is a bit reminiscent to our music profesionnal surveys’, is the creation of synthetic datasets. This is the approach suggested by the Finnish policy brief, and partly, by our research collaboration on the modernization of copyright data and metadata (Osimo et al. 2019; Senftleben et al. 2021). In this approach, the dataset owner is not revealing data that is protected by personal data protection or business confidentiality, but provides a randomized dataset that has exactly the same pre-defined statistical properties as the true dataset. In simple terms, the data owner is not releasing any real data, and any personal information, just a random(ized) dataset where the distribution of the analyzed information (such as the distribution of income, average and median income, standard deviation of the income) is very closely resembles the “true” data. Such techniques are often employed by statistical agencies, too, when they provide researchers with datasets from official statistical surveys.
2.5 The CEEMID-CI Indexes
The average and median (typical) earnings cannot be calculated arithmetically—the average does not exist if we do not know the number of works, and the median cannot be observerd if we do not know how many works, recordings, or rightholders to rank. The mean values are also rather useless because of the large concentration of revenues in a greatest hits segment.
This problem is not specific for the music industry. Copyrights and neighbouring rights are claims for future cash flows, just like stocks and bonds. Many of the aspects mentioned above are present in the empirical analysis of stock and bond securities markets, where the global universe of investible securities is vast and changing daily, particularly in the bond market. Bonds often have only 30- or 90-days lifespan, they “expire” and new bonds are re-issued. Most copyrights and neighbouring rights only earn revenues in the first 1-2 years of their lifespan but remain in the copyright-protected universe potentially for 100-200 years.
The fact that copyrights and neighboring rights are fixed in term (even though often they are fixed relative to an unknown event, the death of the composer), they are similar to fixed income (bond markets). Eventually all rights will enter the public domain and stop paying royalties.
The fact that each music work and recording has a variable number of streams at a monthly varying price in varying territories makes streaming similar to dividend-paying stocks. Streaming payouts, like stock dividends or prices, are autocorrelated (March offers an insight into April earnings) but unpredictable in the long-run. Many songs enter a similar phase where they stop earning any royalties—much like the stocks of companies that cease to exist.
The reports of IPFI and CISAC, in a financial market analogy, are describing the annual growth of market capitalization, but not the individual or even typical performance of assets. Market capitalization can grow as the number of investible stocks, bonds, or streamable songs grows, even if representative sales volumes and prices fall. These are not useful for the characterization of the economic situation of investors or rightsholders. Instead, we pioneered the creation of indexes that represent the market view of a particular rightsholder.
The CEEMID-CI indexes slightly resemble the approach of the Standard&Poor, Dow Jones, or iBoxx indexes. Indexing is complicated—the formula and know-how to create these well-known indexes is protected intellectual property, and we only used some of their methods to create our proprietary streaming indexes. Our indexes were not created for the MCE report, and we would have used a different indexing for this purpose, but we believe that our “typical” indexes are very useful for the aims of the MCE project.
Our indexes paint a more rosy picture than that view from a ‘typical’ rightsholder. Our index shows what happens with the typical song of the month. Every month, we selected those songs that were performing better than half of the songs in a particular country, and worse than the other half in the CI portfolio. This is the median value of the songs that were listened to by anybody in a particular country, and not a measure of all possible songs. The median (typical) value for all songs is zero, because there are so many more copyright and neighbouring right protected works and recordings available every given month than the amount which are played at least once.
In the period of 2015-2019, the typically exploited song in a large, independent portfolio exhibited the following characteristics:
- The audience expressed in the volume (quantity) of the the monthly streams was generally increasing.
- The monthly revenues were flat.
- The prices of a single stream were declining.
- The variability (risk) of the earnings from month to month was greater in emerging markets.
- The greater risk in emerging markets was compensated with higher volume growth, and higher revenue growth, just like in financial markets.
- The revenues of poor markets were high relative to the household cultural spending differences. Often we felt that the streaming providers made pricing mistakes, or inconsistent pricing in national markets. The likely cause is that neither the streaming providers nor the rightsholders made adequate market research into sales and price planning in the smaller markets, and naturally this increased the riskiness of international revenues.
2.5.1 CEEMID-CI Streaming Value Index
The revenues for the typical used song were roughly flat in almost all markets. This means a decline in streaming value in the case of flat volumes. Our finding is not contradicting the IFPI reports of rising total value. Streaming services are lincensed to more and more markets, and more and more recordings (from the newer territories, from the back catalogue, and new releases) are becoming available to a growing global audience. Because of the internationally competitive British repertoire, this means growth for the UK industry, at least in the domestic perspective, though in the national perspective the growing international competition may lead to loss of market share for invidiual rightsholders who are UK nationals.
Figure 2.1: Total Monthly Streams of a Typical Song in the United Kingdom and 19 European Markets
2.5.2 CEEMID-CI Streaming Volume Index
The streaming volumes in the earlier saturating, advanced markets of the United Kingdom and Germany showed slightly increasing or flat volumes for the typical used song. This is likely to mean a decline for the typical song, but the typical (median) song is not played in any given period, i.e. month. Emerging markets, where the service is later introduced, show initial growth.
Figure 2.2: Total Monthly Streams of a Typical Song in the United Kingdom and 19 European Markets
2.5.3 CEEMID-CI Streaming Price Index
When volumes are rising and revenues are flat then prices must be falling. We have seen lowering prices per stream in almost all territories, expressed in GBP terms. Germany’s flat price in GBP is the result of a falling euro price offset by the devaluation of the British pound against the euro.
Streaming services logically target first the upper middle class users, and then offer cheaper subscriptions for families or students. The falling unit price follows the profit optimisation of streaming providers.
At first sight we may be surprised by relative high earning potential of poor markets like Albania. We attributed the relatively high remuneration in these countries to a low level of competition (few subscribers, fewer exploited songs) and the effect of various minimum licensing fees.
Figure 2.3: Total Monthly Streams of a Typical Song in the United Kingdom and 19 European Markets
In fact, we can see the very same effect within the United Kingdom, where the relatively new Apple Music behaves like an “emerging market” provider. In our report we hypothesized two explanations.
- Rightsholders set a minimum licensing fee at the beginning of a new license. In the first months, when there are only a small number of subscribers who stream a small number of songs, the average revenue on each song is relatively high.
- There is a price discrimination present. Each service provider sells for the most interested consumer groups first, then offers various sales, and then rolls out family packages, where the subscription fee per household member is very low.
Figure 2.4: Quarterly Moving Average Price of a Typical Song in the United Kingdom
While it is tempting to single out Spotify as a lowest value provider, Spotify usually generates the highest volumes. The streaming service providers are rolling out subscriptions to different parts of society, and at different costs. For Spotify, the largest marketing cost is access to Apple’s in-app purchase system. Spotify has challenged Apple on similar related cases in the UK, in the Netherlands and in EU jurisdictions (Chee 2021).
In this period, we did not see a very strong logic in how the subscription fees were set by the various streaming providers, though generally speaking, they set higher subscription fees in richer markets. Richer markets had higher subscriber bases, higher subscription fees, but also a higher level of repertoire competition—as larger parts of society could afford to buy electronic appliances and pay for subscription fees, they were interested in a broader repertoire of music. In small markets, where usually upper-middle class subscribers were present, the repertoire used was more homogeneous. In the 19 markets, we saw very different levels of interest for domestic and foreign music.
The index values for the United Kingdom can be used as a benchmark for domestic revenues for the UK Music Industry, particulary for smaller rightsholders who are British nationals or British taxpayers. But because of the international competitiveness of the UK repertoire, even these artists are earning foreign revenues. For benchmarking the domestic (as opposed to national) revenues, an internationally diversified portfolio is a better point of reference. The UK music industry often represents foreign artists, too, who may have their main audience outside the UK.
One measure of competitiveness for a recording is its ability to draw revenues for many jurisdictions. Music is a predominantly local business: most musicians have a smaller fan base than a country. Some bands may have followers only in Cardiff, or in Wales; some Welsh bands may be well-known in the entire United Kingdom but not abroad. And some of them are streamed globally. Any artists who can draw revenues from more lucrative markets than the United Kingdom may be better remunerated than the ‘typical’ British artist. And an artist who has many followers in relatively less lucrative markets, for example, in Eastern Europe, may receive less remuneration on the same number of streams.
2.5.4 Publishing side
We do not have direct information about the publishing side, and we believe that this is more difficult to observe. Some rightsholders, particularly large publishers, are licensing their songs directly to YouTube, Apple Music, Spotify and Deezer, and we do not have access to the terms of these agreements. Other rightsholders are collectively licensed. These licenses are not public.
Generally, the revenues of the publishing side should move parallel to the recording side, but in our experience, because of the lower value of the publishing side, and the hybrid licensing model (direct or collective), far less care has been exerted to correctly document works. Our experience outside the UK shows that uncollected revenues may be significantly higher for the publishing side.
2.6 Unpaid Revenues
In the streaming services, a very significant number of uses are not matched with rightsholders. There are many reasons for this: - There are billions of transactions that have a very low value; the financial motivation to fix small administrative errors is minimal. - Because of the very low expected income for many works and recordings, often the present value of the proper documentation of the work and recording exceeds the likely present value of future revenues. The documentation is patchy. - There are no harmonized metadata standards, and righsholders cannot keep up with the documenting needs of billions of transactions.
The problem of these uses is that they use up a large part of the total revenues (that are financing both royalty payments and operating costs, including identification costs). The cycle of these unpaid royalties is seemingly long, but due to the complexity of the problem, very often the problem is never solved on time. These revenues, after cost deductions, will eventually land in some form of a ‘fund’ that will re-distribute these revenues to other rightsholders. This effect is a redistribution from less documented and typically small catalogues to large catalogues.
In emerging markets, the problems are so persistent, that often they are not solved over several resolution cycles. We have all reason to believe that the problem is more severe on the publishing side, at least partly because the value of the publishing side is smaller in streaming, and the licensing model is more complex—there are more things that can go wrong, and the revenue pot to solve the problem is smaller (Senftleben et al. 2021).
We do not want to suggest that all factors always move against niche genres and independent labels. This example tries to explain why a very fragmented industry like the UK music industry can face so different revenues and economic landscapes. An empirical examination of this problem is even more difficult than observing the average or typical rightsholder revenues, because the magnitude of the problem is unlikely to be independent of firm size, language, and other characteristics of the repertoire documentation.