13.2.1 Industry Background

We present the key characteristics of the concert industry that are relevant for this chapter. A more detailed review is available in Connolly and Krueger (2006) and Waddell et al. (2007). The modern touring industry was born in the late 1960s when a few bands such as the Rolling Stones and Led Zeppelin regularly started touring a variety of arenas and stadiums, using their own experienced crew to take care of the sound, staging, and lighting. In the 1980s, advances in technology allowed bands to offer even more ambitious stage shows that were louder and brighter, and available to ever-larger audiences. By 2007, the North American concert industry had grown to $4 billion in revenue and 100 million in attendance.³

Although some artists give single concerts, the dominant model in the industry is that of tours. In brief, a concert tour is typically organized by an artist represented by his or her manager, a (booking) agent, and a promoter. The artist and the agent agree on an act and a tour plan. The agent then looks for promoters to organize the event in each city. The artist comes to an agreement with each promoter on a pricing policy and on a revenue sharing rule. Promoters are in charge of organizing the events. This involves booking venues, advertising, and collecting revenues. There are some variations on the theme. Most artists use the same set of promoters to be in charge of the tour, but some also use local promoters in certain cities to tap into the local expertise so crucial for success. A few artists even do everything in-house and contact the venues directly. Although there are different types of tours (e.g. promotional tours of new releases, seasonal tours, festival tours), all of the concerts in a single tour usually include a common set of songs and similar staging, and are marketed together.

13.2.2 What Is Specific About the Pricing of Live Events? A Review of the Literature

Ticket prices of concerts are typically set jointly by the artist and the promoter(s) when the tour is announced, and remain unchanged thereafter. Each event is unique and there is no set formula for pricing a concert. There is no second chance if one gets the wrong number of seating categories or prices. Events are sometimes added or cancelled, but prices or category allocations typically remain the same.

The problem of pricing tickets for live events shares much in common with selling perishable products such as tickets for air travel, booking hotel rooms, or handling restaurant reservations. At the heart of the problem is the issue that the seller has a fixed capacity, faces much demand uncertainty, and has a limited amount of time to sell tickets. Many industries dealing with perishable products use techniques known as revenue management, dynamic pricing, or responsive pricing (Courty and Pagliero, 2008) to handle these problems. However, the live event industry does not think about pricing a seat for a concert in the same way that a revenue manager thinks about pricing a seat for a flight or a hotel room. The concert industry is unique in its lack of sophistication. Although we have seen more experimentation with revenue management in recent years, it is still rare, and one has to ask why the concert industry does things differently.

13.2.3 Summary and Questions to be Addressed

The pricing of tickets offers an ideal case study to investigate standard questions in industrial organization (monopoly pricing, price discrimination), but with several twists due to the emotional nature of the product (musical performance), the special relationship between buyer and supplier (fan-idol), and the role played by the media in influencing the demand for top artists (celebrity status). The following questions are open:

The rest of this chapter presents a detailed analysis of price discrimination and rationing. We identify several puzzling features of the data and propose a unified framework based on the concept of artist pricing styles to explain these puzzles.

13.3.1 Data

Our data identifies the main parties involved in organizing a concert (artists, venue, and promoter), with the exception of the agent, whose role is limited to putting artists and promoters in touch. For each concert defined by the date, venue, and artist(s), the Billboard dataset reports the promoter in charge, ticket prices, venue capacity, attendance, and the revenue realized. One main shortcoming of this dataset is that it does not have information on tours. We gathered that information from band and fan websites. In addition, we gathered information on the characteristics of the bands from music websites, artist websites, and the Rolling Stone Encyclopedia of Rock and Roll.

Our resulting panel data is thus three-dimensional. The first dimension describes the product (i.e. a concert) and can be aggregated by music genre, artist, or tour. The second dimension describes the local demand and can be aggregated at the level of city or state. In addition, knowledge of the venue where the concert takes place provides information about both product (venue characteristics) and demand (through location) characteristics. The third dimension is time.

There are several differences with respect to the Connolly and Krueger (2006) dataset. In terms of depth, our data is richer in several dimensions: (i) we observe all of the prices for each concert, rather than just the highest and lowest, and (ii) we know whether a concert is part of a tour and, if so, what tour it belongs to. This additional information allows us to provide a much more complete picture of the pricing strategies across seating categories and also across venues by comparing only concerts that belong to the same tour (with the same product offered in different local markets). In terms of breadth, our dataset covers fewer artists and fewer years. Still, we cover a large fraction of the industry measured in value terms for the years in our sample.

13.3.2 Scope and Representativeness

Our sample includes all concerts collected by Billboard given by the top 100 grossing artists over the period 1992–2005. Billboard collects data on most concerts offered by our sample of artists in North America. We checked this by sampling a few tours, for which we collected the exact tour schedule from the artist’s website and matched it with the concerts reported in our database. In terms of breadth, our sample represents the majority of the industry in value terms. If we increased the sample to include the top 500 grossing artists over the same period, for example, the top 100 artists would represent 70% of the total revenue. Obviously, the sample covers only a small fraction of all performing artists. For our purpose, however, the pricing policies in our sample are representative, in value terms, of the average ticket sold in North America. That being said, our selection rule draws only from the superstars. The industry distinguishes between new performers and established artists. Established artists have more bargaining power over their promoters. They also probably have more market power to set prices.

A few entries in our sample include multiple artists who often tour together (e.g. Billy Joel and Elton John, Bob Dylan and Paul Simon). We treat each of these artists as one artist when they tour alone and as another when they tour together. Hence, we have a total of 122 different artists. In the rest of this chapter, the term artist (or act) may refer to an individual, a band, or a set of these systematically touring together.

Table 13.1 presents some descriptive statistics. Our sample contains 122 artists, 779 tours, and 20 362 concerts. There are 1561 concerts given on average each year.⁴ Most concerts in our sample were given as part of a tour. The average number of artists performing in a given year is 57 and this number does not vary much across years (the minimum is 42 and the maximum 67). The average artist gives seven tours and 167 concerts in our sample period with respective medians of 5 and 151. The majority (75%) of artists give at least two tours. The average tour has 24 concerts with a median of 18 and a standard deviation (SD) of 22. There is variability in the number of concerts per tour, but half the tours have between 8 and 34 concerts.

Concerts are given in 579 different cities throughout the United States. For half these cities, all the concerts are hosted in the same venue. For the other cities, there is much variation in the number of venues used. The overall average number of venues per city is 2.8 and the maximum is 25.

The tours in our sample are large multi-million-dollar operations. Each concert is associated with a promoter. There are 464 promoters in our sample. Table 13.1 presents the distribution of the number of concerts organized by each promoter. The median promoter organizes two concerts and there is much variation across promoters. Clear Channel Entertainment dominates the market (it organizes a bit more than a quarter of the concerts in our sample), but it has many competitors. About 46 promoters organize 67 concerts or more.

13.5.1 Second-Degree Price Discrimination

We identify concerts that use multiple seating categories with a dummy variable:

(13.1)

where p_i^H (p_i^L) is the highest (lowest) price for a seat in concert i. Table 13.2 presents summary statistics on price discrimination. In our sample, second-degree price discrimination is used in 75% of the concerts. The dummy variable d_i measures the existence of price discrimination, but does not take into account the number of seating categories or the price difference between seating categories. The number of seating categories per concert ranges from 1 to 4 with an average of 1.99. Overall, 56% of the concerts offer two price categories, 25% one, 15% three, and the remaining 4% four categories.

We next report some statistics on the intensive margin of second-degree price discrimination. We measure the maximum differences in price for seats in the same concert. The average price range (p_i^H – p_i^L) is about $25.74. After normalizing by the low price, (p_i^H – p_i^L)/p_i^L, we get an average of 0.99. Top seats cost on average twice more than the worst ones. This figure, however, hides much heterogeneity. As reported above, p_i^H – p_i^L is equal to 0 for 25% of the concerts. The price range p_i^H – p_i^L grows to $34.43 for concerts in which p_i^H ≠ p_i^L. In addition, the quality premium is extremely high for a few concerts.

The three measures of price discrimination (price discrimination dummy, number of prices, relative price range) are positively correlated with very low p-values. In the rest of this chapter, we will often conduct the empirical work using the price discrimination dummy because it is simpler to manipulate (e.g. than the number of prices) and easier to interpret (e.g. than the price range, which has an arbitrary component to the extent that concerts may use different venue splits). However, the results still hold using alternative measures of price discrimination.

13.5.2 Third-Degree Price Discrimination

We measure third-degree price discrimination at the tour level, since concerts in a tour are virtually identical (same stage, musicians, and set of songs). Rental and labor costs can vary from one city to the other. The largest fraction of these costs, however, is highly inflexible at the venue level since the only choice variable that is costly to adjust is the number of shows offered in a given venue. However, most tours offer a single show in most cities visited. For the sake of conciseness, we do not discuss in detail the case of multiple concerts given in the same city.⁵

Conditional on visiting a city, the price of tickets should depend only on demand factors (local public) and on venue characteristics, in particular total capacity. If price discrimination takes place, we would expect prices to vary from city to city as long as there are important variations in public demand across cities. The only reason for a lack of variation in prices is the implausible scenario that differences in audiences are exactly compensated for by differences in venue characteristics.

To measure third-degree price discrimination, we define a concert pricing policy as the number of seating categories and the price for each seating category. For each tour, we record the pricing policy used in each city. We say that uniform pricing is used for a set of cities if the pricing policy does not vary across the cities in that set. The reader should keep in mind that the terminology ‘uniform pricing’ means different things for second- and third-degree price discrimination. The correct interpretation, however, will be clear from the context.

There is no single method of measuring uniform pricing at the tour level. We propose two measures. The first computes the fraction of concerts within a tour that use the modal pricing policy, which is the pricing policy most frequently used within a tour. On average, 22% of the concerts use modal pricing (Table 13.2). This is the average across all tours, of the proportion of concerts that use the same pricing policy as the tour modal policy. This high figure could be driven by tours with few concerts. For these tours, a high proportion of concerts may use the modal policy even though the actual number of concerts with identical policies is low. This is not the case. For example, the proportion of concerts that use the tour modal pricing policies does not decrease when we restrict the sample to tours with at least 10 concerts.

The median number of concerts per tour is 18. If each concert within a tour were priced differently (a different number of seating categories or different price for at least one seating category) the fraction of concerts using the tour mode would be 5.5%. The much higher figure of 22% suggests that uniform pricing across cities plays a large role in the concert industry.

Our second measure computes the Gini–Simpson homogeneity (or concentration) index for the set of pricing policies in a tour. This is the probability that two concerts drawn randomly from a tour use the same pricing policy. It can be written as:

where t denotes a tour, i denotes a pricing policy within a tour, n_i,t the number of concerts in tour t using pricing policy i, and N_t the number of concerts in tour t. Let N denote the total number of concerts in our sample. On average across all tours, the probability that two concerts in a tour use the same pricing policy, G = Σ_t(N_t/N)G_t, is 7.4% (Table 13.3). If all concerts in a tour sharing the same pricing policy used the modal policy, we would expect the Gini index to be around 4.8% (0.22 squared). The fact that it is much higher says that uniform pricing at the tour level is not just due to modal pricing.

One concern with our measures of price discrimination is that some pricing policies may just happen to be the same by chance. A second concern is that identical pricing policies may be associated with venues or with promoters rather than tours. Table 13.3 reports the Gini–Simpson index for different partitions of our sample. Note that the Gini–Simpson index is at least three times higher for tour partitions than for any other partition (venue, artist, year, city, or promoter). This indicates that uniform pricing occurs mainly at the tour level, confirming the validity of our measure of third-degree price discrimination.

There are many different measures of the extent of third-degree price discrimination. We compute the interquartile range of prices within a tour for the lowest, mean, and highest price. To illustrate these concepts, assume for the sake of argument that tours use a single seating category. The interquartile price range in tour t is (p_t⁷⁵ – p_t²⁵), where p_t⁷⁵ is the price that corresponds to the 75th percentile of prices in tour t and similarly for p_t²⁵. This measure provides information on how much prices vary across cities within a tour. The interquartile range provides a more robust measure of variability of prices than SD, for example, because there are outliers. On average across all tours, the interquartile range of the lowest price is $7.5, of the mean price $8.3, and of the highest price $9.4. Table 13.2 shows that the interquartile range of the mean price is about 23% of the average price within a tour.⁶

Our measures of third-degree price discrimination are correlated, and the correlation is statistically significant at conventional levels. Most interestingly, the tours that use less modal pricing also vary prices less across cities. There is no clear reason why this should be the case. Something common to all concerts in a tour probably influences several pricing decisions. In Section 13.9, we will see that artist pricing styles can rationalize these correlations.

13.5.3 Summary

We have defined two sets of measures for second- and third-degree price discrimination. The first set measures the existence of price discrimination. The second set measures the intensive margin of price discrimination (differences in price). We find that price discrimination is often but not always used. Uniform pricing is also common, although not as common as price discrimination. The next section shows that there is considerable variation in the use of price discrimination across artists.

• There is essentially only one seating category (on average 1.05) in the Bruce Springsteen tour (Fig. 13.1), while there are typically multiple seating categories (on average 2.37) in the Michael Bolton tour (Fig. 13.2), with significant variability in price within a venue (the highest price in a concert can be up to 200% higher than the lowest price). Bruce Springsteen rarely uses second-degree price discrimination while Michael Bolton often does so.

• Most prices are equal to one of two values ($30 or $33) across locations for Bruce Springsteen’s tour, while they vary greatly for Michael Bolton’s tour. The fraction of concerts that use the modal pricing policy is 44% and 0.6%, respectively (Table 13.4). The differences in pricing patterns for these two tours are remarkable.

13.6.1 Second-Degree Price Discrimination

Table 13.5 reproduces Table 13.2, but at the artist level. To illustrate the difference between these two tables, consider our measure of second-degree price discrimination. Here, the unit of observation is an artist. Denote E(d_i|a) the mean value of d_i across all concerts offered by artist a where d_i is defined by Eq. (13.1). This is a measure of an artist’s propensity to use second-degree price discrimination. Table 13.5 presents summary statistics of the variable E(d_i|a). On average, artists use second-degree price discrimination 77% of the time. This figure is similar to the same figure for the entire sample of concerts (Table 13.2). The new information is found in the next columns of Table 13.5 which report statistics on the variability across artists. These statistics differ greatly from Table 13.2, which reported statistics for the entire sample.

There is a large SD (26%) in artists’ average use of price discrimination. The range across artists is also very large. Billy Joel uses price discrimination in 4% of his concerts, Garth Brooks in 8%, and KORN in 22%. However, Madonna, the Eagles, and Pink Floyd almost always price discriminate. Figure 13.3 plots the distribution of E(d_i|a) for our sample of 122 artists. The height of the histogram corresponding to x on the horizontal axis, for example, measures the fraction of artists who use uniform pricing about x% of the time. The spread of the density mass is distributed across the two extremes of 0 (never use second-degree price discrimination) and 1 (always use it). This confirms that there is much variation across artists in the use of price discrimination.

Going back to Table 13.5, 10% of the artists use price discrimination in at most 38% of their concerts. At the other extreme, a quarter of the artists almost always use second-degree price discrimination (in 97% of their concerts or more). The same holds if we look at the average difference between the highest and lowest priced seats. Ten percent of the artists set an average price premium of 15% or less. At the other extreme 10% of the artists set an average price premium of 214% or more.

13.6.2 Third-Degree Price Discrimination

Table 13.5 also reports statistics on our measures of third-degree price discrimination averaged at the artist level. Again, the means do not change much. For example, artists use the modal pricing policy on average for 22% of their concerts (no change in the mean relative to Table 13.2). What is relevant for us are the statistics on the distribution across artists. The SD across artists in the use of modal pricing is 15%. There are on average 80 tours per artist. If modal pricing were random across artists, the use of modal pricing would average out at the artist level around the sample value of 22% and we would expect to observe little variation across artists in the use of modal pricing. This is not the case.

This is confirmed by Fig. 13.4, which reproduces Fig. 13.3 for modal pricing. Again the density mass is spread across the 0–1 interval. About 25% of the artists use modal pricing on average in less than 11% of concerts, while 10% use modal pricing in 47% of concerts or more. Consider the distribution of Gini–Simpson coefficients across artists. There is a great deal of heterogeneity across artists in the chance that any two concerts in a tour are equally priced. The SD across artists in the Gini–Simpson coefficients is 0.10 (recall that the average Gini–Simpson coefficient across all tours was 0.074). For 10% of the artists, the probability that two concerts in a tour use the same prices is 20% or higher. At the other extreme, 10% of artists never set the same price for any two concerts in a tour.

The same conclusion holds when we look at the intensive measures of third-degree price discrimination. Ten percent of the artists have an interquartile range of the average price that is 11% of their average tour price; while at the other extreme, 10% of the artists have an interquartile range that is 41% of the average tour price. The amount of price variation across cities within a tour varies greatly across artists.

13.6.3 Summary

There is much heterogeneity in the extent to which artists use second- and third-degree price discrimination. This is true for our binary indicator for uniform pricing and also for the measures of price discrimination that take account of differences in prices.⁷ This confirms that the difference between Bruce Springsteen and Michael Bolton is not specific to these two artists. In the next section, we investigate possible explanations for the observed differences in pricing across artists.

13.7.1 Theoretical Framework: When Should Artists Use Price Discrimination?

Assume an artist sells tickets to two different audiences. The tickets could be for the same concert in two different venues or for two different seats for the same concert. Accordingly, the public could live in two different towns or buy two different types of seating categories. In this latter interpretation, we make the simplifying assumption that consumers are interested in only one seating category. Allowing for the possibility of substitution across seating categories adds realism, but does not change our main conclusions.

The inverse demand by consumer c = 1,2 for seat category c and for artist a is P(q|c,a) = α_c,a – βq. The marginal cost is χ (typically small or zero in the concert industry). We assume that differences across consumers and artists can only influence the intercept α_c,a. This is to establish a benchmark; later we will revisit this assumption.

Under price discrimination, the artist chooses prices in order to maximize q(α_c,a – βq – χ) in each market. Profits from audience c are (1/4β)(α_c,a – χ)². Under uniform pricing, overall profits are (1/8β)(α_1,a + α_2,a – 2χ)². The increase in profits, or the return from price discrimination, is:

where F is the fixed cost of implementing price discrimination. Consider the benchmark case where the demand intercept for a concert performed by artist a in front of audience c is additively separable.

Additivity assumption: α_c,a = α_a + α_c.

The net profits from price discriminating simplify to R = (1/8β)(α₁ – α₂)² – F.

Proposition 1 is important for two reasons. (i) Artists are expected to use price discrimination when there is enough difference across audiences. For example, they should use second-degree discrimination when seating categories are perceived to be sufficiently different. This could stem from physical differences in seating categories within a venue or heterogeneity in willingness to pay for seats of different quality. Similarly, they should use third-degree price discrimination if the local audiences where the tour stops are sufficiently different or if the venues are sufficiently different. (ii) The decision to price discriminate does not depend on the characteristics of the artist that equally affect all consumers. Proposition 1 says that we should control for demand shifters that influence quality differences or difference in willingness to pay for quality. After controlling for product and demand shifters, the decision to price discriminate should not depend on the artist’s identity as long as the additivity assumption holds.

13.7.2 Summary

Proposition 1 helps interpret the results presented in the previous section. For example, the differences across artists in the use of second-degree price discrimination could be rationalized if artists perform in front of different audiences with different willingness to pay for seating quality. The differences in the use of third-degree price discrimination could be rationalized if artists tour different subsets of cities. Coming back to Figs. 13.1 and 13.2, it could be that Michael Bolton visits very different cities and performs in venues with very heterogeneous seating experiences, while Bruce Springsteen tours similar cities and books venues where all seats are similar.

The next section initially assumes that the additivity assumption holds, and investigates whether the variations in the use of price discrimination can be explained by demand and product characteristics. In the rest of the section, we relax the additivity assumption and consider a number of other explanations for the variations in artist pricing practices.

13.8.1 Second-Degree Price Discrimination

We propose to explain the decision to price discriminate with controls for demand, product heterogeneity, and artist fixed effects. Assuming that the additivity assumption holds, we follow the empirical methodology proposed by Bertrand and Schoar (2003) to identify the existence of managing styles. In a nutshell, we estimate artist fixed effects θ_artist from model:

(13.2)

where θ_city denotes city fixed effects that control for differences in local audiences and for differences in venue characteristics for all the cities where there is a single venue (more than half the cities in our sample), θ_venue denotes venue fixed effects that control for venue characteristics more precisely than city fixed effects do, θ_year denotes year fixed effects that control for changes over time in public taste, public preferences for seating quality, or in the cost of implementing price discrimination, Popularity_a,y controls for heterogeneity in artist popularity as we will explain shortly. City and year fixed effects control for unobserved differences in the level of competition across cities and over time.

We estimate Eq. (13.2) using a linear probability model, instead of a non-linear model (logit or probit). This is without loss of generality in a saturated model where all right-hand side variables are dummies or categorical. The fitted probabilities are simply the conditional probability of using second-degree price discrimination within each cell defined by the different values of the dummy variables.⁹

Like Bertrand and Schoar (2003), we look at three sets of statistics: changes in adjusted R² associated with the artist fixed effects, F-tests that the artist fixed effects are equal to 0, and summary statistics on the distribution of the artist fixed effects.

We can answer several questions:

Table 13.6 reports the results. Column (1) shows that artist fixed effects explain 27% of the variations in the use of second-degree price discrimination. Column (2) shows that year and city fixed effects explain about 18% of the variations in the use of price discrimination. The adjusted R², however, goes from 18% to 40% as we add artist fixed effects (move from column (2) to column (3)). This shows that the variations explained by city and year fixed effects are to a large measure orthogonal to the variations explained by artist fixed effects.

Note that artist fixed effects are economically highly significant in the sense that they explain a large fraction of the variations in price discrimination. This result will remain in all our specifications. In contrast, manager fixed effects in Bertrand and Schoar (2003) explain only 4% of the variations in corporate behavior.

The bottom of Table 13.6 shows statistics on the distribution of the artist fixed effects. The SD of estimated artist fixed effect is 0.25, which is very close to the 0.26 figure in Table 13.5, as can be expected. The SD corresponding to Table 13.6, column (3) is only slightly lower than in Table 13.6, column (1). The percentile estimates change very little.

We repeat the same exercise in Table 13.6, columns (4) and (5), with venue fixed effects instead of city fixed effects. The conclusion remains the same. The adjusted R² increases from 28% to 46% when we add artist fixed effects (compare columns (4) and (5)). Interestingly, year and venue fixed effects explain about 10% more of the variations than year and city fixed effects (compare columns (2) and (4)). This suggests that venue fixed effects capture some variations in product characteristics.

We conclude that local market characteristics and venue characteristics explain a large portion of the variations in the decision to second-degree price discriminate – a finding consistent with Proposition 1. Still, even after controlling for these potential sources of unobserved heterogeneity, the proportion of variation in the use of price discrimination explained by artist fixed effects does not decrease much. In fact, Fig. 13.5 reproduces Fig. 13.3, but using the estimated fixed effects of the specification controlling for venue and year fixed effects. If local market and venue characteristics explained much of the variability across artists in the decision to price discriminate, then heterogeneity across artists, captured by the range of the distributions, would decrease after controlling for venue and year fixed effects. This is not the case. Heterogeneity in pricing styles still seems to play an important role.¹⁰

13.8.2 Third-Degree Price Discrimination

In the case of third-degree price discrimination, Proposition 1 says that we should control for the fact that different tours stop in different subsets of markets with possibly different venue characteristics and local audiences. Holding the set of cities within a tour constant should go a long way toward controlling for the mix of audience and venue characteristics. However, there are 579 cities in our sample and no two tours visit the same set of cities. One option would be to focus on the set of most visited cities. However, doing so would still leave many differences in the set of cities visited across tours.

We cannot use our measure of third-degree price discrimination computed at the tour level and also hold the set of cities visited constant. Therefore, we instead leverage the fact that there are many individual cities that are visited by a large fraction of tours. The nuance is that each city is visited by a slightly different subset of tours. Instead of measuring price discrimination at the tour level, we consider pairs of cities. For each pair, we can identify those artists who use the same pricing policy in the two cities and those who do not. We then aggregate this information across all pairs of cities and compute differences in pricing style across artists after holding city-pairs constant.

The exact procedure is as follows. We first select the top 10 cities most visited and form the 45 possible city-pair combinations. For each pair, we construct an observation for each tour that visits that pair of cities. We construct a variable that is equal to 1 if the two pricing policies for that tour are identical and equal to 0 otherwise. This produces a dummy variable describing uniform pricing that assumes the value of 0 or 1 each time one of the 779 tours in our sample stops in one of the 45 possible city-pairs. The dummy variable for uniform pricing is equal to 1 in 16% of these observations. In Table 13.7, we explain the variation in this dummy variable with artist fixed effects (column (1)), with city-pair fixed effects (column (2)), and both sets of fixed effects (column (3)). The adjusted R² with artist fixed effect alone is 0.18 (column (1)), and with city fixed effect alone it is 0.14 (column (2)). The first result is consistent with our earlier finding that there is much heterogeneity across artists in the use of third-degree price discrimination. In fact, the SD of the artist fixed effects is 0.19.¹¹ The second result is consistent with Proposition 1 stating that the use of third-degree price discrimination should depend on differences in local market characteristics. City-pair dummies control for differences in audience and venue characteristics. Price discrimination should be more likely to take place in pairs of heterogeneous cities. Most interestingly, the 18% figure is identical to the increase in adjusted R² when we add in column (3) the artist fixed effect to the city fixed effects (0.32 – 0.14 = 0.18).

Table 13.7 also presents summary statistics on the artist fixed effects. We find no decrease in the SD of the artist fixed effects after controlling for differences in local market characteristics. The SD of the artist fixed effect is 0.19 in columns (1) and (3). There is significant heterogeneity across artists. The probability that two concerts have the same pricing policy in the same pair of cities varies by 0.77 across all the artists in our sample. Figure 13.6 reproduces Fig. 13.4, but by using the estimated artist fixed effects from Table 13.7, column (3). The spread of the distribution does not change much relative to Fig. 13.4.

One possibility we have not discussed so far is that some artists may use a cost-based rule to set the price of tickets in each venue. They may charge a price equal to the rental and labor cost plus some fixed mark-up. Uniform pricing could result if subsets of venues have the same rental and labor costs. Table 13.7 rules out this possibility because we hold constant city-pairs. Uniform pricing cannot be due only to the fact that some venues have the same costs.

We have made important progress. We started from Proposition 1. It says that after controlling for demand and product characteristics, artists should make the same price discrimination decisions. Instead, we find large differences across artists even after controlling for demand and product characteristics and this holds both for second- and third-degree price discrimination. What is the explanation for this? It could be that we have missed some dimensions of demand and product heterogeneity. The demand specification of the model behind Proposition 1 or other features of that model may be too simplistic. One could extend the model in several directions. Although a full treatment of the issue is beyond the scope of this work, we can rule out three candidate rationales that are particularly relevant to our application.

13.8.3 Other Sources of Heterogeneity in Artist Demands

The model assumes that artists influence the demand for tickets only through an additive component to the intercept. It could be that artists influence the demand in more complex ways. For example, the artist could influence the slope of the demand β or the overall demand in a multiplicative way, as in P(q|c,a) = k_a(α_c – βq). The additivity assumption does not hold anymore. The return to price discrimination becomes R = (k_a/8β)(α₁ – α₂)² – F. It now depends on k_a. Both k_a here and α_a in the previous model are measures of artist popularity. The main difference is that they have a different impact on the return to price discrimination. Proposition 1 no longer holds when popularity is multiplicative. More popular artists, in the sense of an increase in k_a, are more likely to price discriminate. One could argue that some unobserved component of k_a explains the variations in pricing styles across artists. This possibility cannot be ruled out a priori.

The existence of non-additive demand heterogeneity across artists is not implausible. An increase in popularity then increases both the level of willingness to pay and also the difference in willingness to pay across audiences. In the case of second-degree price discrimination, an economic argument in support of this case goes as follows. The increase in willingness to pay to upgrade seating category is related to the level of willingness to pay if, for example, artist quality and seating quality are complements in the utility function. Consumers are willing to pay more to upgrade their seating category when they are willing to pay more for the concert. One can make a similar case for third-degree price discrimination.

The possibility of heterogeneous demands deserves serious consideration. Table 13.6 controls for venue fixed effects. This should take care of the above problem if artists sort across venues by demand type (e.g. more popular artists play in larger venues). Still, there may remain some heterogeneity in artist popularity that is unaccounted for.

We can make some progress toward showing that this is unlikely. We can measure many characteristics of the artists, but we can use as controls only those that vary over time. We check whether the variations in pricing style across artists remain after controlling for these time-varying characteristics. For those artist characteristics that do not vary over time, we follow a split-sample approach (see below). For the sake of conciseness, we conduct these robustness results only for second-degree price discrimination.

As our first control, we use a measure of artist popularity based on the success of musical recordings. We measure the number of albums and singles in the top charts up to a given year. If popularity is the main driving force of artist heterogeneity, then adding this variable should reduce the fraction of adjusted R² explained by adding the artist fixed effect. The last two columns of Table 13.6 control for artist popularity. The impact of popularity on the use of second-degree price discrimination is positive and significant. Having one additional top single or album increases the likelihood of using second-degree discrimination by 1%. This is consistent with the earlier interpretation of k_a as a multiplicative impact of popularity. However, the fraction of adjusted R² that is explained by artist fixed effects only decreases marginally. The distribution of artist fixed effects shows that the economic magnitude of differences in pricing styles does not change.

In the split-sample approach, we rank all the artists in our sample according to the average revenue per seat (average price of tickets sold) and then compute the mean across all artists. We split the sample into two categories – high and low average price artists – according to whether an artist’s average price is above or below the average across artists. This controls for artist popularity under the reasonable assumption that average ticket price is a proxy for popularity. This approach directly addresses the concern that price discrimination could be correlated with the level of ticket price. Under that scenario, the explanatory power of artist fixed effects should decrease within the subsamples of high and low average price artists. Table 13.8 reproduces selected specifications of Table 13.6 for our split sample.

We first discuss the evidence on the F-test for the artist fixed effects. In all cases, we reject the null that all artist fixed effects are equal to 0. Adding artist fixed effects increases the adjusted R² although the magnitudes are smaller than before. For high price artists, the adjusted R² increases by 18% (difference between columns (3) and (1) in Table 13.8) when we control only for the cumulated number of hits. It increases by 14% when we control for year and venue fixed effects as well. The figures are a bit lower for low price artists, but even in the lowest case, artist fixed effects still increase the adjusted R² by 10%. Again, the distribution of artist fixed effects does not change.

Heterogeneity in demand could also be due to differences in musical genres. There are several channels that could be at play. First, and most importantly, demand may vary across musical genres. The return to price discrimination may be higher for rock artists because, for example, they sing to more diverse audiences. Alternatively, one may argue that the community standards of fairness vary by musical genre. The rock audience may respond more strongly to unfair and exploitative pricing.

The main musical genre in our sample is rock music, which represents a bit more than half of the artists. The remaining artists cover a wide range of music including country, jazz, and rap. We split the sample by rock versus non-rock music. Our main interest is to investigate what happens to the artist fixed effects when we focus on the rock subsample. If musical genre explains pricing decisions, we would expect that artist heterogeneity should matter less for the subsample of rock artists. Table 13.9 reproduces selected specifications of Table 13.6 for rock and non-rock artists. For the rock subsample, the increase in R² associated with the inclusion of artist fixed effects remains high around 17–24%. Similarly the SD in artist fixed effect remains around 22%. Differences in musical styles do not explain the heterogeneity in the use of second-degree price discrimination across artists.

13.8.4 Cost of Implementing Price Discrimination

Artists may have different access to information regarding the benefit of implementing price discrimination. Coming back to the model, the fixed cost of implementing price discrimination, F, could include information costs that are artist dependent. There is a related version of this argument. The model assumes that artists know the demand and can compute the profit-maximizing prices. In practice, the return from price discrimination depends on the knowledge that an artist has about the demands for the differentiated product. Variations across artists in the use of price discrimination may be explained by differences in knowledge or expertise.

Many artists set prices jointly with promoters. If promoters have important information on how to set prices, we expect that promoter fixed effects should absorb some of the heterogeneity across artists in access to information. In Table 13.10, we add a set of fixed effects for promoters. The adjusted R² in column (1) in Table 13.10 is 0.28, which is about 0.10 higher than the adjusted R² with city and year fixed effects alone. This large increase in the adjusted R² is consistent with several interpretations. It may be due to access to information as argued above. Another explanation is that promoters may specialize in different types of music. Promoter dummies may control to some extent for unobserved musical style.

The important point for these additional results, however, is that even after controlling for promoter fixed effects, we still find that the adjusted R² increases by 16%. We also find that the distribution of artist fixed effects does not change even after controlling for promoter fixed effects. Again we find that access to information does explain some of the variations in the use of second-degree price discrimination, but these variations are largely orthogonal to the variations explained by artists fixed effects.

13.8.5 Learning

It may be difficult to find out whether it is profitable to adopt price discrimination in a city that was never visited before. However, even if they make mistakes early on, artists should learn over time, particularly when they repeatedly visit the same city or venue. This suggests controlling for the number of times an artist has performed in a given city or venue before.

We compute statistics on artist past experience for each concert in our sample. For a given concert, past experience is defined as the number of times an artist has previously given a concert in the same city. The value of past experience is equal to 0 for the early observations in our sample and increases up to 4 for the final years in our sample (the average experience is 1.26). Columns (3)–(6) in Table 13.10 report the regression results. The main result of this table is that the increase in adjusted R² explained by artist fixed effects does not change even when we control for past experience; this holds when we measure experience at the city or venue level.¹² In both cases, the increase in adjusted R² associated with artist fixed effects is around 0.19. The distribution of artist fixed effects does not change.

13.8.6 Summary: A Price Discrimination Puzzle?

We have considered a number of explanations based on standard economic theory. We found some evidence in support of these explanations. This demonstrates that it is important to control for local demand, product characteristics, artist popularity, and access to information. Taken together, all our controls explain a bit less than half of the variations in the decision to second-degree price discriminate. Although some of the variation in the use of price discrimination can be explained by differences in demand and product characteristics or access to information, doing so does not decrease the amount of variation that is explained by artist fixed effects. We are left with a puzzle. What explains the variations in pricing across artists?

One could still argue that the variation in pricing practices across artists is due to unobserved demand heterogeneity. The return from price discrimination for a concert by Bruce Springsteen may not be the same as for one by Michael Bolton. Returning to Proposition 1, it could be that there are unobserved demand differences (that are not related to the controls we have tried) or other variables that influence the return to price discrimination. As argued before, it is not possible to fully exclude this possibility. However, we think it is unlikely, because we would have expected that much of this heterogeneity should have been related to factors that vary across cities, venues, promoters, years, and our measures of popularity and musical styles. The finding that these controls did not reduce much the role of artist heterogeneity indicates that seeking new controls is unlikely to resolve the issue. Keeping in mind that unobserved demand heterogeneity may contribute to some of the heterogeneity in pricing practices, in the next section we consider an alternative hypothesis, based on the assumption that artist pricing styles exist, to explain the variations in pricing practices across artists.

13.9.1 Hypothesis: Willingness to Exploit Market Power

Assume artists differ in their willingness to exploit market power. At one extreme, the pro-social artist does not want to leverage his or her market power. They use uniform pricing for all tickets in a tour. They charge the same price for all seats in a venue and for all venues in a tour. At the other extreme, the profit-maximizing artist charges the market clearing prices for all seating categories and to all audiences.¹³ These are two archetypes, but artists may also take intermediate positions. The evidence from the previous sections that there is much heterogeneity across artists in the use of second- and third-degree price discrimination offers some support to our hypothesis. We derive two new implications.

Consider first the level of prices. Pro-social artists may keep prices low to make their concert affordable to all fans. This implies that some artists will systematically sell out. They will also refrain from responding to positive demand shocks by immediately increasing prices to the new equilibrium level. Instead, they may slowly increase prices and partially incorporate the demand shock. Krueger (2005) and others have documented dramatic increases in demand in our sample period. An implication is that pro-social artists should be more likely to sell out in this period. As artists do not adjust prices by the same margins to match demand, we expect rationing probabilities to vary across artists.

H1 investigates a new feature of pricing. It does not consider differences in prices across products, as we have done so far, but instead tries to back up information on the artist’s choice of the level of prices from evidence on rationing. Those artists who refrain from exploiting market power will price tickets below market price. If this is the case, we would expect that they should be more likely to sell out their concerts. While all the evidence presented so far has been on price discrimination, H1 has to do with the level of prices. The main shortcoming of H1 is that it is subject to the same reservations as those made in regard to our analysis of price discrimination. Artist pricing styles is consistent with variations in rationing probabilities. However, unobserved demand heterogeneity is an alternative candidate explanation.

The second implication presents a totally new way of looking at the evidence. It tackles the issue of exploitation of market power directly. If artists differ in their willingness to exploit market power, we should be able to predict how they will price tickets if we know where they stand between the two archetypes. Although we do not have this information, we can compare the different decisions each artist makes. Those artists who are willing to exploit market power should do so along all dimensions of pricing.

It is more difficult to argue that H2 could be explained by unobserved heterogeneity across artists. The unobserved demand or cost factors that are picked up by artist fixed effects would have to be correlated across second- and third-degree price discrimination. This puts a much more demanding requirement on the set of candidate unobserved factors. The same holds for why these two decisions should be correlated with the decision to ration.¹⁴ However, our simple framework based on exploitation of market power delivers a unique prediction about the relation between three decisions.

This new hypothesis is important for two reasons. Finding a non-zero correlation would add to the case that artist pricing styles matter. Bertrand and Schoar (2003) have argued that if fixed effects are caused by individual heterogeneity (instead of unobserved heterogeneity) one would hope that they are all caused by a common root factor. They argue that the correlation evidence is due to some overarching patterns in decision making. This conclusion rests on the implicit assumption that there is no theoretical argument for why any unobserved heterogeneity picked up by the fixed effects for different decisions should be correlated across decision makers. This is reasonable in our application. There is no economic theory that links second- and third-degree price discrimination (Stole, 2007). There is also no reason why the unobserved demand factors for second- and third-degree price discrimination should be correlated across artists.

However, the assumption that pricing styles are due to different willingness to exploit market power allows us to go one step further. We can sign the correlations. This provides a unique test of our new hypothesis that the differences in pricing practices are due to differences in willingness to exploit market power.

13.9.2 Artists’ Rationing

It is reasonable to assume that there is excess demand for those concerts that are sold out. Sold out is a coarse measure of rationing, however, because we do not know how much excess demand there is. Table 13.2 reveals that 43% of the concerts in our sample are sold out. The last line in Table 13.5 shows that there is much variation in rationing probabilities across artists. For example, the Allman Brothers never sell out. Janet Jackson, Styx, Bob Dylan, and Paul Simon sell out in less than 15% of the concerts. However, a quarter of the artists ration tickets in at least 57% of their concerts. For example, Madonna always sells out, while Billy Joel, Elton John, and Garth Brooks sell out in more than 85% of the cases. The interquartile difference across artists in rationing probability is 34%. This is a very large number considering that the sold out probabilities are fairly well estimated. In fact, the minimum number of observations per artist in our sample is 15 and the median across artists is 150. Figure 13.7 plots the distribution of sell out probabilities across artists. The range is striking. Even if we restrict the sample to artists with at least 100 concerts in the sample, the results are not much affected (Fig. 13.8).

The fact that some concerts are sold out is not surprising. After all, demand is uncertain and prices have to be set in advance. A random component of the demand is realized only after tickets are offered for sale. By increasing the price of tickets, the artist increases the revenue per seat sold, but also increases the risk of having unsold tickets. If demand is uncertain, the probability of rationing should be strictly positive. The profit-maximizing level of rationing depends on the elasticity of demand, the amount of uncertainty, and the venue capacity.

Several stylized facts from the concert industry are difficult to rationalize within this simple profit maximization framework. As mentioned above, a large subset of artists sell out most of their concerts and they do so tour after tour. This cannot be profit maximizing. There is a high suspicion that these artists systematically underprice tickets (as illustrated by the large literature reviewed earlier that points towards systematic underpricing). For example, according to our initial quotes, there seems to be a widespread belief in the industry that Bruce Springsteen, Pearl Jam, and Dave Matthews have never charged as much as they could. In our sample, they sell out 70%, 66%, and 56% of their concerts respectively. If all artists perform in similar demand conditions, theory predicts that artists’ rationing probabilities should be fairly close to one another. Returning to Fig. 13.7, we see that this is not the case. Although there is a peak centered on 40%, there are large tails on both sides.

One concern with Fig. 13.7 is that artists may not face similar demand conditions. We check that artist fixed effects are robust after controlling for demand and product heterogeneity by following a similar approach as we did for second-degree price discrimination. Under a profit maximization hypothesis, rationing probabilities should depend on venue and demand characteristics that influence the uncertainty of demand and the shape of the demand curve. Assume, for example, that the population of concert fans varies from city to city and that this influences the local demand elasticity and/or the level of demand uncertainty. This could be due to differences in income, racial composition, age composition, or other variables. We would expect that the rationing probabilities should differ from city to city.

As with second-degree price discrimination, demand and product characteristics may explain variations in the probability that a concert sells out. We use a similar empirical model as before to extract artist fixed effects. The dummy variable r_i equals 1 if concert i is sold out and 0 otherwise. We estimate the model:

(13.3)

where the control variables were defined in Section 13.8. Table 13.11 reports the results with various sets of controls. Artist fixed effects alone explain 18% of the variability in concert sell out. Most interestingly, the amount of variations explained by artist fixed effects does not decrease by a large amount after controlling for year, venue, or city fixed effects and artist popularity. Artist fixed effects increase the adjusted R² by 15% when we control for year, city fixed effects, and artist popularity. This figure remains the same if instead we control for venue and year fixed effects. The F-test corresponding to the hypothesis that artist fixed effects are jointly equal to 0 is rejected in all three specifications (columns (1), (3), and (5)).

Figure 13.9 reproduces Fig. 13.7 but using the artist fixed effect estimated in Table 13.11. The distribution is again strikingly spread out. Table 13.11 also presents statistics on the distribution of artist fixed effects. The SD does not change across the three specifications. The range of the distribution of artist fixed effects and the interquartile range are large. The finding that some artists almost never ration and others always do so holds even after controlling for a number of demand and product factors. Is this due to artist pricing styles?

In Section 13.2.2 we reviewed the main explanations for rationing that have been proposed in the literature. Most of these explanations (consumer demand for fairness, coordination game between consumers, publicity value of selling out, complementary products, and gift exchange) can explain the overall level of rationing in the concert industry but cannot on their own explain large variations across artists in rationing. To explain these variations, explanations would have to assume that some artists are subject to these forces while others are not. It is not clear, for example, why Bruce Springsteen’s fans care about fairness while Michael Bolton’s do not. In addition, specification (3) controls for a large number of variables associated with demand. Again, we doubt that the variations across artists in sell out probabilities are due only to unobserved demand or product heterogeneity across artists. Some of it is most likely caused by different pricing styles.

13.9.3 Artist Pricing Styles and Exploitation of Market Power

Under the assumption that artists differ in the propensity to exploit market power, we should find that the decisions to second- and third-degree price discriminate are positively correlated and that these two decisions are negatively correlated with artist sell out probability. We start with evidence from the raw data based on artist averages. Figures 13.10–13.12 present the raw plots of the three variables of interest, second- and third-degree price discrimination and sold out, taken two by two. Each point on the figures represents an artist. The three figures show correlations that are consistent with our hypotheses (and statistically significant at 1% confidence level). Artists who more frequently differentiate prices within a given venue are less likely to use the same pricing policy across concerts within a tour. Artists who are less likely to price discriminate are more likely to sell out concerts. These correlations are difficult to explain under the theory of price discrimination. However, they are consistent with our behavioral assumption that artists differ in their willingness to exploit market power.

We follow the approach of Bertrand and Schoar (2003) to address a shortcoming with these raw correlations. Each point of the graphs is computed by taking averages for an artist. However, artists perform in different cities, venues, and years. Hence, it may be possible that it is the characteristics of the cities, venues, and years in which concerts take place that determine the choice to price discriminate and to sell out. We check whether the correlations across decisions are still present after controlling for these characteristics.

Like Bertrand and Schoar (2003), we construct a new artist dataset. For each artist, we collect the estimated fixed effects and standard errors from regressions (1) and (2) for second-degree price discrimination and rationing, respectively (results in Tables 13.6 and 13.11), and from the regressions described in Table 13.7 for third-degree price discrimination. We then estimate the following equation:

(13.4)

where F.E.(y) and F.E.(z) are any two of our three fixed effects. We can then test if the estimated coefficient β is significantly different from 0 and has the predicted sign, keeping in mind that this coefficient has no causal interpretation. It is a measure of the association between the fixed effects. This is consistent with H2 which only says that the fixed effects should be associated in a systematic way. The right-hand side variable in Eq. (13.4) is itself an estimated coefficient which is noisy by definition. This will tend to bias the estimated coefficient β toward 0. Hence, the results will be biased towards rejecting the existence of pricing styles.

Since we know the precision with which the fixed effects were estimated, following Bertrand and Schoar we use a generalized least-squares (GLS) technique to account for the measurement error in the right-hand side variables. We weigh each observation by the inverse of the standard error of the independent variable.

Table 13.12 reports the results of these regressions. The average R² of these regressions is 0.1 with a maximum of 0.38 and a minimum of 0.014.¹⁵ Panel (A) in Table 13.12 computes the regression coefficients using the fixed effects from specifications that include only the artist fixed effects. All regression coefficients are significant and have the predicted sign. Fixed effects for second- and third-degree price discrimination are positively correlated, and both are negatively correlated with rationing.¹⁶

Panels (B) and (C) in Table 13.12 address the issue of unobserved product and demand characteristics. In both panels (B) and (C), we use the artist fixed effects for third-degree price discrimination from Table 13.7, column (3), which holds constant city-pairs. Panel (B) takes the artist fixed effects for second-degree price discrimination and rationing from a specification that controls for city, year fixed effects, and artist popularity. Panel (C) is similar, but controls for venue rather than city fixed effects. The signs of the regression coefficients remain the same, although their statistical significance decreases. This is likely due to the fact that artist fixed effects are less precisely estimated when more control variables are included in Eqs. (13.1) and (13.2). Similar results hold using alternative measures of price discrimination. For example, if we use the number of seating categories instead of the price discrimination dummy as a measure of second-degree price discrimination, the results are again significant at the 10% level.

Overall, accounting for measurement error as well as demand and product characteristics does not change the main findings of the analysis described in Table 13.12, panel (A). The evidence is consistent with the hypothesis that there are differences across artists in willingness to exploit market power.

13.9.4 Summary

The hypothesis that artists vary in their willingness to exploit market power is consistent with the observed heterogeneity across artists in the use of second- and third-degree price discrimination, and the large differences in their propensities to ration tickets. Most importantly, this hypothesis implies that the propensity towards second- and third-degree price discrimination should be positively related, and that both should be negatively related to the propensity to ration tickets. The empirical results are broadly consistent with these predictions.

• What ultimately differentiates artists? Sincere preferences, strategic motives, or something else? Is there a relationship between an artist’s ‘exploitation of market power’ and celebrity status? Are famous artists less likely to exploit market power? Are artist pricing styles fixed or is there a life cycle in pricing style?

• Our study can be generalized in several directions. Do less popular artists (non-top-100 artists) also have individual pricing styles? Do the results generalize to other pricing decisions (e.g. the speed of response to demand shocks)? Are pricing styles specific to live concerts for popular music? Do the results generalize to other performing arts? Is the notion of individual pricing style relevant in other markets where celebrity matters? How general is the notion of willingness to exploit market power?

• Some artists have started to experiment with more innovative pricing policies leveraging the distribution opportunities offered by the Internet (Halcoussis and Mathews, 2007). Our results suggest that not all artists may adopt these new opportunities to the same degree. Which ones are embracing these new possibilities?

• What is the welfare impact of heterogeneity in pricing style? Does consumer demand for fairness influence the overall level of prices?