Introduction

The peculiarities of the recording industry system, such as fashion cycles, the hedonic nature of music, socionetwork effects, informal heuristics in the decision-making process of recording companies, and the opaque selection process of media gatekeepers have created uncertainty about the chart potential of musical products [Steininger & Gatzemeier, 2013]. Even listening to any popular radio station supports evidence to the notion that the commercial success of music performers is skewed: on the radio, we can hear only popular compositions. It is so because a relatively small number of performers dominate the music industry. [Cox, Felton & Chung, 1995]

The problem of the current study grows out from the following situation. First, structure of popular music market is oligopoly. Distribution of successful composition is not flat and not normal - it is highly skewed. Moreover, the skewness of `golden records' - indicator of the reward obtained by musicians from their work - is not explained only by talent of the artists and quality of their compositions. There are many other factors, some of which are external to the artist. In addition, the situation is escalating due to the mass nature of the market, and the Internet, which serves to accelerate communication and facilitate market entry, and hence, increase competition. All these theses are discussed in more details and evidences further.

Belinfante and Johnson [1982] give an evidence of an oligopoly market structure in a study of the company concentration in the U.S. recorded music industry. Rosen [1981] finds substantial differences in income between musicians of the first rank as opposed to those of the second rank. MacDonald [1988] proposes a theoretical model for the music industry where the distribution of rewards is such that relatively minor differences in talent can generate large differences in returns. Finally, Hamlen [1991] studies the relation between voice quality and success (while controlling for factors such as gender, race, type of music, and duration of career). The results show that, despite consumers appreciate quality, voice quality does not explain all the variation in dollar record sales. [Hamlen 1991]

Studies of chart success show that incidence of chart success is substantially skewed to the right, whether measured by total weeks spent per artist, average weeks spent per album, or the total number of albums listed per artist. [e.g., Strobl & Tucker, 2000] Rosen [1981] proposes two explanations for the existence of superstars. The first suggests that differences in talent imply that there is imperfect substitution between the output of different artists. Basically, consumers are unwilling to substitute a lesser talent for a greater talent, and the greater this unwillingness to substitute among talents, the larger are the rewards which accrue to the most talented artists relative to their lesser talented counterparts. Thus, small differences in talents can lead to large success differences. Rosen [1981] also suggests that the skewness is reinforced by technological developments in the media through which music is disseminated and the possibility of reproducing recordings repeatedly at the same fixed cost, so that the costs of services to larger audiences do not rise in proportion to the size of the market. Therefore, the most talented artists have a significant advantage over their rivals in terms of their ability to meet the larger demand for their output by exploiting scale economies. [Rosen, 1981]

Adler [1985] argues that a consumer's appreciation of a particular artist's output increases as their knowledge of the artist increases. Thus, the individual's consumption decision over time involves a learning process during which consumers accumulate their “consumption capital”. In music market learning process involves both listening to music and, more importantly, discussing it with other persons. This results in a snowballing of demand for the output of artists. [Adler 1985]

Important point is commercial success of recorded music depends on several influential factors, which are not necessary in the artists' locus of control. [Steininger & Gatzemeier, 2013] For example, the decision to select a potential talent is not based on the actual ascertainment of consumer taste but on the expectations of consumers' taste by media gatekeepers. Even though the creative output passes the media gatekeeper, it might fail to be appreciated by the audience [Hirsch, 1972].

Power and Hallencreutz [2007] show that music distribution channels and retail environments exist in a networked commodity chain dominated by a limited set of oligopolistic global firms. It is suggested that to understand success or failure in music sales and better develop regional production milieus, one must first understand those who control the distribution channels and retail environments through which sales are made. [Power and Hallencreutz, 2007]

While increasing music production and availability, and new forms of communication with the audience are emerging, musicians face with the problem that being good is not enough to be successful. Internet changed dramatically the commercial music industry landscape: the industry has evolved from the traditional model of the big music labels to the much criticized peer-to-peer music file trading and, today, to the legitimate online downloading model [Vaccaro, Cohn, 2004]. This last one is also changing, as there is a constant trend not to download music anymore, but to listen to it directly through social media such as the older MySpace or the actual very popular YouTube, Facebook, and other similar national or regional social networks [Mjos, 2011].

With the social media opportunities, as never before, international success is possible for artists from small commercial music markets, such the ones of the emerging countries, who become rapidly well-known. [Florina & Andrea, 2012] Therefore, despite digital technologies have altered the way music is produced, promoted, distributed and consumed, and this has individualized the majority of musicians by the weakening of the labels, managerial function is still in demand [Hracs, 2015]. This fact evidences that artistic talent does not necessarily lead to success. Musicians need to survive in a competitive market, and they (or their managers and producers) need to know what factors inflate commercial success.

Summing up, relatively small number of performers dominate the music industry. Opaque selection process of media gatekeepers and decision-makers lead to uncertainty. Considerable role of limited set of oligopolistic firms dominating in networked commodity chain of music distribution channels and retail environments increase uncertainty and inequality of musicians' chances. Raised Internet provides new channels of distribution and communication with the audience from the one hand, but these multiple channels inflate completeness and therefore competitiveness of the market from another hand. All of it lead to the situation when equally talented or slightly different in the level of talent musicians differ dramatically in gained success. Thus, highly skewed success of musical artists and uncertainty about the chart potential of musical products drive us to explore factors of commercial success in music industry.

Described situation is understood as problem of presented research. Subject of the research is commercial success. Object of the research is pop music compositions. There is a natural limitation on genre caused by structure of music industry. Specific of many genres [e.g. post-punk, noise, krautrock] not assume commercial success and popularity. Consequently, we will not find songs of these genres in top of charts. Thus, our object is constrained by popular music.

Empirical base of the study is the sample of songs, charted as popular by two different charts - Billboard and Spotify - in 2017, and the same number of randomly selected songs, released in 2017, and not charted by Billboard and Spotify. The data used in the study mostly collected from Billboard Magazine official website, Spotify streaming platform website, and Million Song Dataset project. Also, other open resources were used in data collection.

Key research question is why very similar music can be dramatically different in gained success. Therefore, main goal of the study is to model success depends on various factors and determine the key success factors. There are technical audio characteristics, vocabulary and semantic, release characteristics, live concerts characteristics, media image, outside collaborations. All these factors may affect commercial success of popular music compositions. There is a theoretical analysis and conceptualization of listed influencers in the first - theoretical - part of the study. There is a statistical analysis of some of them in the second - empirical - part of the study. Presented research is aimed to test the following hypotheses:

1) Commercial success of music compositions is partly determined by technical audio characteristics: loudness, tempo, key, mode, duration, instruments.

2) Commercial success of music compositions is partly determined by personnel: number of musicians, average age, gender structure.

3) Commercial success of music compositions is partly determined by musicians' career moment: album number, musician lifetime.

4) Commercial success of music compositions is partly determined by aggressive communication and promotion: video production, participation in tv show, using a composition in movie, other media usage, live performance.

5) Commercial success of music compositions is partly determined by record companies' characteristics: label power, number of labels.

The study requires to fulfill the following tasks:

1) To perform a theoretical analysis of the current state of the defined problem under study in the world science.

2) To develop a research program that allow to model success depends on various factors and determine the key success factors.

3) To collect the needed data on pop music compositions: their success indicator and influential characteristics according to formulated hypotheses.

4) To perform a statistical analysis of the data: data processing and modeling.

5) To describe and discuss the results of the research according to defined problem and formulated hypotheses.

Research methods and techniques have been chosen based on the need to build a model that can classify a song as popular or not popular based on some features of the song. Current study proposes defined problem be solved as classification machine learning problem. Pop music compositions are marked as “successful” and “not successful”. Catboost classifier is implemented for binary classification modeling.

The paper expands existing literature on music scenes and suggest a complex way for advancing research on success of musical products. It makes an important empirical contribution to musicology area. Due to practical orientation of the research question, relevance of the study goes beyond academy, and also can be addressed to music industry decision-makers such as managers, labels and producers.

Paper is organized as follows: introduction, theoretical chapter, empirical chapter, conclusion, limitations. Problem, goal, object, subject, hypotheses and tasks of the study are defined in the introduction.

`Theoretical model of the international success of popular music compositions' chapter contains six paragraphs. First is to give an idea of the world of music and delineate boundaries of the focus of current study. The second is devoted to success - key concept of presented research. Next paragraph gives on overview of what is done in the sphere of musicology and other fields, dealing with the question of commercial success of popular music. Then, literature review continues with scrutinizing attempts to predict commercial success in popular music industry, described from the literature. The last paragraph of the first chapter presents the theoretical model of popular music lifecycle. Some conclusions are given for summing up the results of theoretical chapter.

The second chapter is called `From Theoretical Model to Statistical: Empirical Chapter'. It describes research approach, results and discussion of the results. Research approach is described in 7 paragraphs. Each of them is devoted to one part of research program: operationalization of indicators and structural scheme, data, sample, data collection, methods and procedures, software, analysis pipeline with first results of exploratory analysis. Chapter continues with results and discussions. Some conclusions are also given for summing up empirical chapter.

In conclusion, answers to the key research questions and hypotheses are summarized. Paper ends with discussion of limitations of the study.

Theoretica l. Model of the International Success of Popular Music Compositions

music popular international success

There are several angles to think about music. First, music is an art form that reflects reality through sound organized in time. From the music theory perspective, important elements for analysis are pitch [which governs melody and harmony], rhythm [and its associated concepts tempo, meter, and articulation], dynamics [loudness and softness], and the sonic qualities of timbre and texture [which are sometimes termed the "color" of a musical sound]. Those are components of music structure, and we are interested in them as in objective characteristics of music itself. Music structure analysis help music semantics understanding which can aid music transcription, summarization, retrieval and streaming. [Maddage et al, 2004] Music is often seen by researchers as analog to language. Listeners perceive emotions and feelings from music partly universally, partly differently across cultures [Balkwill & Thompson, 1999; Egermann, 2015; Juslin et al, 2016].

Second, music is cultural and social activity. Music plays a key role in religious rituals, rite of passage ceremonies (graduation and marriage), social and activities (dancing) and cultural activities (community choir). [Lawson, 2011; Ferer, 2012] Moreover, music is a part of everyday life. People may make music as a hobby or just listen to it during everyday activities. There are number of studies devoted to effect of music listening or studying and performing on health, wellbeing and academic and work success. [MacDonald et al, 2013; Brabec, 2011]

Third, music is a service industry. The music industry includes the individuals who create new songs and musical pieces (such as songwriters and composers), individuals who perform music (which include orchestra, jazz band and rock band musicians, singers and conductors), individuals who record music (music producers and sound engineers), individuals who organize concert tours, and individuals who sell recordings, sheet music, and scores to customers. Besides music industry itself, there are related areas such as soundtracks for film production and musical accompaniment for advertising industry. In that sense, lifecycle of music products can be considered as business process and explored from economic efficiency point of view. Consequently, all players and stakeholders are important to be analyzed. [Strachan, 2007; Presenza & Iocca, 2012; Hracs, 2015; Lize, 2016] Our focus is mostly those who perform music regardless of what else they do or not do: write, record, distribute, etc. Consequently, music is understood in our study as products of musicians - performed and recorded compositions.

The main basis for the classification of music is genre. Genre is a conceptual tool most often used to classify varieties of cultural products, particularly in the fields of visual art, popular culture, video games, film, literature, and music. It describes a manner of expression that governs artists' work, their peer groups, and the audiences for their work [Becker 1982; Bourdieu 1993]. All music can be divided in 4 big groups: folk (national, traditional) music, religious (sacred) music, academic (art, classical music, canonic) music and popular music. Within each of these groups there is variety of genres. Scholars devote their studies to building taxonomies of musical genres [Li & Ogihara, 2005] as well as questioning evolution and development of musical genres and tastes [Lena, & Peterson, 2008].

Music of different genres is associated with different contexts. Each kind of music requires its own traditions and canons not only in musical structure, but also in the way it is performed, distributed and listened to. Academic music is playing in big ancient halls of the Conservatory, religious music is playing in churches and adjusting to the rituals performed. Different genres of popular music are also different in concert venues, forms of distribution and social situations of performing and listening. So, underground, grange, post-punk music is mostly (with rare exceptions) performing in cheap underground bars and pubs with tiny stage, distributing self-recorded songs through small editions of analog media or through the Internet. Such music can be called niche. Pop music is oriented to (but not necessarily achieve) bigger halls, more expensive tickets, larger circulations - it tends to win mass market. All channels of distribution and promotion are important for pop-music artists, and all channels of feedback too: professional critique, charts, comments from listeners at forums. We are interested in the analysis of those artists for whom the focus on mass market, higher profit from sells, higher positions in charts are relevant.

Summing up, music is a service industry, and we are interested in analysis of musical artists considering them as firms. As any firm music artist is aimed to reach profit, i.e. commercial success. In this study we understand musical artist as one who is represented to mass audience as performer of music, individual or in the band. Musical products are, in this case, all forms of presenting music to the audience - songs, singles, albums, videos, concerts, etc. Hereby, musical artist's or composition's success can be defined as attainment of legitimacy and measured, accordingly, by acceptance by an audience. [Baumann, S., 2007]

Since thinking of commercial success, we want to capture acceptance by an audience. Traditionally, success of an artist was evaluated based on performance of the chart sales, signaling of popularity. For example, Hamlen [1991] used information on record sales in the U.S. to determine whether small differences in artists' ability produced large differences in success. Chung, Felton and Cox [1995] use data on gold record awards in the U.S. to assess the probability distribution. Burke [1996] use chart listings from the British Phonographic Industry to examine the dynamics of product differentiation in the British record industry using. Ordanini [2006] use it to analyses strategies for cooperation with recording studios.

Steininger and Gatzemeier [2013] argue that being successful in the charts is the most important key performance indicator in the industry. Recording companies track peak positions and duration on the charts in order to get insights about the quality and lifecycle of their musical material. [Bhattacharjee, 2007] Strobl and Tucker [2000] state that having an album featured in the charts can be considered the primary goal of most popular music artists and their record labels. Not only does it indicate the quality of an artist's pre-recorded musical output, and hence the artist's talent, but it also reflects the considerable amount of earnings of artists and their record labels. Some researchers try to estimate earnings more precisely. [Strobl and Tucker, 2000] So, Chon et al. [2006], bi-weekly sales data from the Billboard magazine, particularly Top Jazz chart, to predict an album's success.

However, charts are not the only indicator of popularity, especially since the advent of streaming platforms. Araujo et al. [2017] specify the Spotify Popularity and the Billboard Units as best charts to measure commercial success. They claim is that Spotify represents the most popular type of music consumption today (audio streaming), while Billboard ranking still favors the old school market (physical albums). [Araujo et al. 2017]

Fisher et al. [2001; 2002] indicate multiple measures of musical artists' commercial success:

Several researchers have tried to accommodate the information need by developing econometric prediction models in order to find patterns and regularities of successful experiential goods based on their original value, stemming from the intrinsic (e.g., song features such as tempo, loudness) or extrinsic features (e. g., chart history of an artist). Intrinsic factors are interesting for analysis for its objectivity and simplicity of interpretation. However, extrinsic factors can help to answer main question of current study - why musicians with the same or almost the same music (both in structure and in quality) show dramatically different levels of success and popularity. [Bhattacharjee et al., 2007]

Study of statistical patterns in general public's taste of music shows that the higher the starting position of an album is, the longer it is likely to stay in chart. Araujo with colleagues studied comments posted on social networks during 30 days before the album release. They found out that the number of Positive Tweets can explain 95:5% of the variation in the Spotify Popularity with a simple linear model. On the other hand, they could not find statistical evidence that the volume of comments on Twitter correlates with the album success measured by the Billboard magazine. [Chon et al., 2006]

Strobl and Tucker [2000] remark that many artists become successful as much for the lifestyle they represent as for the quality of their music. In investigating possible determinants of chart survivability of albums authors find that soundtracks and greatest hits albums are more successful in generating enough sales to guarantee them chart survival than other types of albums. The results also suggest that one of the most important factors in guaranteeing chart survival is initial popularity. Apart from the reputation and popularity of the artist, audience exposure and substantial promotional campaigns are likely to play a key role in this result, and it is, of course, those albums signed under the label of the more powerful record companies that are likely to be those relatively more promoted. [Strobl and Tucker 2000]

Florina and Andreea [2012] demonstrate crucial points of good marketing strategy that help to achieve great mainstream success: first. orientation to win the international market from the very beginning (lyrics in English, international distribution, frequent concerts, paid or for free, in clubs, festivals and in open air, in many countries). Second is constancy: launch one hit after another, stressing on the right kind of songs. Third, image construction: sending to the press all the information that could find use in articles. Fourth, presence in the media should be chosen very carefully to be associated only with proper and favorable brands. Also, constant relationship with fans using Twitter and Facebook, including innovative ways of communicating, are important. [Florina and Andreea 2012]

A study on relationship between music group success and their external service provide [Fisher et al, 2002] show no statistically significant relationships between overall attitudes towards external service providers and success but reveal some interesting trends. Those with the most positive attitudes towards external service providers on average made the highest amount of money per gig, sold the lowest number of units, had the lowest number of bookings in the prior three months, and were less likely to have gone on tour during the prior year. Authors also explored business practices of musical groups [Fisher et al, 2001] and found that success of a musical group is related to use of outside professionals and using the Internet for promotion, and attitudes towards stakeholders. [Fisher et al, 2001, 2002]

Ordanini [2006] analyses two main approaches for artists' selection in the recording industry: the direct model in which major companies directly choose new artists from the supply market, and the agency model in which small independent labels realize the first choice and large organizations pick their new artists among those pre-selected by independents. The study reveals that artists selected through the agency model exhibit a longer presence on the chart due to repeated successes, while they are slower to reach heavy success once they have entered the chart. Conversely, the direct model leads to artists with a faster path to a strong success, but the same artists have a shorter presence on the chart due to the sporadic nature of their success. The profile of artists selected through these two models is also found to be different: big international soloist stars are more frequently selected through the direct model, while national bands are more frequently selected through the agency model. [Ordanini, 2006]

Hiller [2016] examines characteristics that are important to current commercially viable massive music festivals. The festival hires unknown bands which its consumers will enjoy but are not yet familiar with. These bands benefit a festival because they can enhance reputation and customer experience at a lower fee. The magnitude of this effect is best considered with a band that has played a festival before, where their probability of being hired in a subsequent year is decreased by about 75 %. However, bands without festival experience are more often hired. The most important measurable characteristic of these bands is their quality, specifically inclusion in quality measures. Commercial success, represented by a top album, increases the probability of hiring substantially in the first year as well. In fact, the impact is greater than a quality album. [Hiller, 2016]

Summarizing the above, we can highlight the main factors related to musical product's success, including not only driving factors, but also those which can be used on initial stages to predict further success:

We can construct a theoretical model, representing all parts of musical product lifecycle and those elements of them which might be related to the success. Obtained model represents traditional way of music production and distribution and traditional success. After describing a model, we will discuss the extent to which it can take into account modern scenario.

The first phase is songwriting. There are many things, influencing quality of the musical products from artists talent to global political, economic and other conditions and audience' preferences. Primogeniture is questionable: some believes that audience drives production, others claim that producers enforce consumption. [Florina & Andreea, 2012] A substantial literature is devoted to determining firm quality decisions and how they affect consumer choices. The early works of Lancaster [1971] and McFadden [1977] began the research into the importance of product characteristics. Wolinsky [1983] investigates the possibility that product quality can be used as a signal to consumers. Further empirical work has been done by Berry et al. [1995], Petrin [2002], and others that have explored empirical methods for estimating models of consumers with heterogeneous preferences and the impact of varied product characteristics. Mazzeo [2002], Chu [2010], and Matsa [2011] all explore the relationship between the level of competition and product quality.

Regardless of factors influencing quality of musical products, there are objective characteristics of songs, which can be measured and used in the analysis of success: perceptual qualities (pitch, timbre, meter), formal elements (prosody, melody, harmony, rhythm), performed activity (singing, drumming), genre (lullaby, song, dance).

Next step is record, and here come the managers - either hired professionals or artists themselves. There are a few big corporate players sharing the market, and all major labels promote quite standardized formulas for the artists they represent, depending on the music genre they sing, but the audiences seem to find delight in identifying these formulas, as Adorno believed [Adorno, 2005, p. 56]. As shown above, contract with powerful label not only provide high quality recording, but also inflates promotion. Therefore, artists need to go through a selection process by `gatekeepers' [Throsby, 1994; Caves, 2000] - critics, artistic directors. They evaluate talent's marketability, quality of music. However, there is no standard indicator for quality of hedonic experiential goods. [Steininger & Gatzemeier, 2013] Instead, gatekeepers' use surrogate measures to describe the quality level, such as the innovativeness, uniqueness and authenticity of a song [Zwaan and Ter Bogt, 2009]. In addition, they often anchor the talents to examples of similar and already established artists [Negus, 1992]. Furthermore, intuition is commonly applied in ascertaining the potential of talents. Finally, criteria with respect to the personal characteristics of the artist are considered, such as the artistic skill level, the working attitude, charisma and the visual appearance [Negus, 1992]. Managers involve communicating with recording companies' agents, connecting artists to them and establishing cooperation. In fact, role of managers is important in all stages - quite comprehensive it is described by Hracs [Hracs, 2015].

Fig. 2. Structural Scheme of commercial success of popular music compositions. Block diagram.

There are five parameters of commercial success of popular music composition accordingly to five research hypotheses. Each of the parameters is represented by some empirical indicators.

The first parameter is technical audio characteristics. There are:

1) Loudness

2) Tempo

3) Key

4) Mode

5) Duration

6) Instruments

These are what makes the sound what it is.

Loudness (i.e. amplitude) is one of the most important characteristics of the sound understanding the sound as a vibration of air.

Tempo is automatically associated with the heartbeat - that makes fast music suitable for intensive training and makes slow music appropriate for breaks.

Key and mode are two characteristics defining mood of the music. The symbolism of classical (academic) music has assigned a certain meaning to certain keys. So, Des-dur is true love, B-dur defines beautiful men, heroes, and e-moll is grief.

Duration is simple characteristics related to the time during which the listener's attention is occupied with a particular song.

The sound of a particular musical instrument affects a particular organ of the human body. There is a widely known opinion that string instruments (violin, guitar, harp and cello) have a healing effect on the cardiovascular system, evokes a sense of gratitude, compassion and sacrifice.

There was the first group of indicators - technical audio characteristics. Doubtless, there are more of them, and if even stay with this list, a combination of the characteristics matters - fortissimo E-moll differ in perception from subito piano
E-moll. However, we use these indicators separately in the model since we are interested in the importance of each of them individually.

The second parameter is personnel, presented in our research as gender and age composition of performers. We use three indicators in the model:

1) Number of musicians

2) Average age of musicians

3) Share of women among musicians [1 for female singer, 0 for male singer, 0.5 for mixed-sex duo, etc.]

The third group of indicators is communication and promotion strategies. This is represented as a bench of binary indicators:

1) Whether a music video was released

2) Whether a composition is used in movie/series/commercial/tele show as a soundtrack

3) Whether a composition was presented at some TV show [e.g. Britain's Got Talent or any Late-night talk show]

4) Whether a composition was used in any other media in any way

5) Whether a composition was presented at live concert (within the tour or not)