Social Media, Traditional Media, and Music Sales: A Panel VAR - - PowerPoint PPT Presentation

Social Media, Traditional Media, and Music Sales: A Panel VAR Approach A Panel VAR Approach

Jui Ramaprasad McGill University Presented to: U i it f Mi t University of Minnesota MIS Workshop April 15, 2011

Music Consumption Music Consumption

4/15/2011 2

Music Consumption: Competing with Free Music Consumption: Competing with Free

• pre-napsterization: radio, live show, purchase CD

pre napsterization: radio, live show, purchase CD

• post-napsterization

– [sharing] free, robust music catalog built by users, [ g] , g y , portable format (napster) – [subscription/streaming] feels free (monthly charge), seamless ser interface rob st m sic catalog m sic seamless user interface, robust music catalog, music “on demand,” portable (rhapsody, the new napster, MOG, Thumbplay) – [sampling] free, music fans want to discover new music based on existing musical tastes, simple interface robust music catalog (pandora) interface, robust music catalog (pandora).

4/15/2011 3

Digitization of Music Consumption Digitization of Music Consumption

• Music is an information good

Music is an information good

– shareable: easy to access and make available

• nline

– free: viable alternative digital ways of consuming music without purchasing, esp. at the song level – unbundled: can now “consume” music as individual songs, not just albums

• M

i i i d

• Music is an experience good

– learning and discovering: from traditional media (radio play) and social media (radio play) and social media

4/15/2011 4

Social Media and Music Social Media and Music

“Amy Kuney's "All Downhill From Here" has been listened to nearly h i i f 250,000 times on her Myspace page. Kuney, who is 25, is part of a new class of musicians who are bushwhacking their way to success. Kuney isn't necessarily trying to use the old formula of getting signed to a record label which is becoming increasingly difficult as the business splinters — which is becoming increasingly difficult as the business splinters. Instead she's using a variety of online tools — from social media to YouTube.” Laura Sydell NPR

• -Laura Sydell, NPR

See also: The Real Value of 7 Million Facebook Fans

4/15/2011 5

Traditional vs Social Media Traditional vs. Social Media

Lorenz Curves: Sampling and Spins

0 6 0.7 0.8 0.9 1

spins/samples

0 2 0.3 0.4 0.5 0.6

lative share of s

0.1 0.2 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Cumu Cumulative share of songs from lowest to highest spins/sampling

Top 1% of Songs played on the radio  50% of radio play Top 10% of songs played on the radio  90% of radio play Top 1% of songs sampled  8% of total sampling Top 10% of songs sampled  40% of total sampling

Sampling Spins

4/15/2011 6

Top 10% of songs played on the radio  90% of radio play Top 10% of songs sampled  40% of total sampling

Research Questions Research Questions

• What is the relationship between social media

What is the relationship between social media activity, traditional media activity and sales?

How does the relationship between traditional – How does the relationship between traditional media and sales vs. social media and sales differ? – How does this relationship vary at the song-level How does this relationship vary at the song-level and the album-level? – How does this relationship vary for more How does this relationship vary for more “mainstream” vs. “niche” music?

4/15/2011 7

Decline in Music Purchasing Decline in Music Purchasing

• Global recorded music sales fell by almost $1.5

Global recorded music sales fell by almost $1.5 billion (8.4%) to $15.9 billion in 2010

– Physical sales fell by 14.2% to $10.4 billion – Digital sales grew by 5.3% to $4.6 billion – Rate of digital revenue growth has halved year on year

• United States:

– Overall shipments of recorded music in the United States fell 11% to $6 9 billion States fell 11% to $6.9 billion. – Growth in digital formats only partially offset a decline of 20% by value in physical formats.

4/15/2011 8

Decrease in Album Sales Decrease in Album Sales

4/15/2011 9

Source: http://mashable.com/2011/04/08/napster-never-existed/

Social Media and Music Consumption Social Media and Music Consumption

• Awareness

Awareness

– Artists can share their music more easily to a broader audience broader audience – Consumers are exposed to a wider variety of music

• Sharing
• Sharing

– Consumers have the ability to “sample” the music C h i i h – Consumer can consume the music without purchasing

4/15/2011 10

Literature Literature

• Impact of Napsterization

p p

– Piracy has led to a decrease in music purchases (Rob and Waldfogel 2004) Piracy has *not* led to a decrease in music sales – Piracy has *not* led to a decrease in music sales (Oberholzer-Gee and Strumpf 2005)

• Social Media & Music

– Sampling on MySpace has a positive relationship with music purchases (Chen and Chellappa 2009) music purchases (Chen and Chellappa 2009) – Impact of social media on sampling is different for mainstream vs. niche music (Dewan and Ramaprasad 2010) 2010)

4/15/2011 11

Conceptual Framework Conceptual Framework

Song Sales Song Buzz Album Sales Album Buzz Radio Play

4/15/2011 12

Panel Vector Auto-Regression (PVAR) Panel Vector Auto Regression (PVAR)

• VAR model with panel data

VAR model with panel data

• Dynamic relationship among a set of

endogenous variables endogenous variables

• Each variable is a linear function of:

– Past values of itself – Past values of all other variables – Error term

4/15/2011 13

Panel Vector Auto-regression Panel Vector Auto regression

• Finance:

– Financial Development and Investment Behavior (Love and Zicchino 2006) Financial Positions and Investment (Stanca and – Financial Positions and Investment (Stanca and Gallegati 1999)

• Marketing

– Marketing Investments on Sales (Dekimpe and Hanssens 1995) Hanssens 1995) – Differential impact of marketing-induced vs. WOM- induced customer acquisition (Villanueva et al. 2008)

4/15/2011 14

Model Model

t j t j t j

S S

  

       

11 12 13 , 21 22 23 , 1 t j t j t j t j t S t J t j t j t j t t j B t j j j j

S S B B        

     

                                 



31 32 33 , t j t j t j R t t t j

R R    

   

               

• J is the order of model (determined by Akaike’s

Information Criterion)

• S B and R and denote song sales song buzz and
• St, Bt, and Rt and denote song sales, song buzz and

radio play, respectively in time period t (t=1,…,T)

• Similar model estimated at the album level

4/15/2011 15

PVAR: Model & Estimation PVAR: Model & Estimation

• Lag Length Selection: AIC for each cross

Lag Length Selection: AIC for each cross section  6 lags

• Variable transformation: Natural Log &
• Variable transformation: Natural Log &

Forward mean differencing (Helmert Transformation) Transformation)

• Estimated using GMM (following Love and

Zi hi 2006) Zicchino 2006)

4/15/2011 16

Data Data

• 993 cross-sections

993 cross-sections

– Song-Level: Song Buzz, Song Sales, Radio Play Album Level: Album Buzz Album Sales – Album-Level: Album Buzz, Album Sales – Other: Record Label (Independent/Major), Artist Reputation Reputation

• 24 weeks

J 19 2006 D b 3 2006 – June 19, 2006 – December 3, 2006

4/15/2011 17

Summary Statistics: Mean (Std Dev ) Summary Statistics: Mean (Std. Dev.)

Full Sample Major Label Independent Label p j p Radio Play (# spins) 60.197 (487.842) 55.553 (323.731) 70.238 (725.193) Song Sales 359 477 322 685 439 075 Song Sales (# units) 359.477 (3295.174) 322.685 (2092.771) 439.075 (4986.696) Album Sales (# units) 1399.939 (5298.380) 1429.907 (5052.9) 1335.135 (5793.582 ) ( ) ( ) ( ) ( ) Song Buzz (# blog posts) (768.327) 7724.130 934.972 (9251.646) 407.970 (1844.706) Album Buzz 27.063 29.639 21.491 (# blog posts) (138.598) (135.183) (145.564) # of observations 23832 (993 songs) 16296 (679 songs) 7536 (314 songs)

4/15/2011 18

Summary Statistics: Mean (Std Dev ) Summary Statistics: Mean (Std. Dev.)

Full Sample High Artist Rep. Low Artist Rep. p g p p Radio Play (# spins) 60.197 (487.842) 226.512 (1157.973) 37.552 (289.138) Song Sales 359 477 1227 048 241 324 Song Sales (# units) 359.477 (3295.174) 1227.048 (8004.678) 241.324 (1870.009) Album Sales (# units) 1399.939 (5298.380) 3762.568 (9892.356 ) 1078.254 (4208.535) ( ) ( ) ( ) ( ) Song Buzz (# blog posts) (768.327) 7724.130 311.3789 (1512.064) (830.543) 8212.332 Album Buzz 27.063 45.564 24.544 (# blog posts) (138.598) (93.829) (143.436)

• No. of observations

23832 (993 songs) 2856 (119 songs) 20976 (874 songs)

4/15/2011 19

PVAR Regression Results: Full Sample PVAR Regression Results: Full Sample

Dependent Variable Song Sales Album Sales Buzzt-1

• 0.101(0.045)**

0.018(0.009) ** Buzzt-2

• 0.087(0.032) ***

0.022(0.007) *** Buzzt-3

• 0.616(0.023) ***
• 0.007(0.006)

Buzzt-4

• 0.048(0.018) ***

0.002(0.005) Buzzt-5

• 0.046(0.019) ***

0.019 (0.005) ***

*** ***

Buzzt-6

• 0.054(0.015) ***

0.014(0.005) *** Airplayt-1 0.069(0.020) *** 0.079(0.020) *** Airplayt-2 0.047 (0.014) *** 0.012(0.014) Airplay 0 007(0 012) 0 020(0 010) * Airplayt-3

• 0.007(0.012)
• 0.020(0.010)

Airplayt-4

• 0.006(0.011)

0.004(0.008) Airplayt-5 0.003(0.011) 0.000(0.007) Airplayt 6 0.004(0.010)

• 0.005(0.008)

4/15/2011 20

Airplayt-6 0.004(0.010) 0.005(0.008) Note: Standard Deviations are in parentheses.

PVAR Regression Results: Sample Splits PVAR Regression Results: Sample Splits

Dependent Variable Song Sales Album Sales Song Sales Album Sales Major Independent Major Independent Buzzt-1

• 0.010(0.040)
• 0.385(0.166)***

0.004(0.011) 0.0519(0.017) *** Buzzt 2

• 0 016(0 027)
• 0.322(0.127) ***

0 017(0 009) * 0.038(0.013) *** Buzzt-2 0.016(0.027) 0.322(0.127) 0.017(0.009) 0.038(0.013) Buzzt-3

• 0.010(0.195)
• .240(0.095) ***
• 0.015(0.008) *

0.008(0.012) Buzzt-4

• 0.177(0.154)
• 0.181(0.774) ***
• 0 .006(0.006)

0.019(0.011) * Buzzt-5

• 0.003(0.153)
• 0.209(0.809) ***

0.020(0.006) *** 0.017(0.010) *

t 5

Buzzt-6

• 0.022(0.012)
• 0.183(0.066) ***

0.004(0.006) 0.035(0.010) *** Airplayt-1 0.082(0.031) *** 0.254(0.089) *** 0.067(0.027) *** 0.100(0.028) *** Airplayt-2 0.050(0.019) *** 0.143(0.053) ***

• 0.003(0.019)

0.043(0.019) *** Airplayt-3

• 0.004(0.014)

0.010(0.307)

• 0.015(0.013)
• 0.019(0.016)

Airplayt-4 0.006(0.013)

• 0.023(0.025)

0.004(0.011) 0.007(0.013) Airplayt-5

• 0.014(0.013)

0.049(0.027) * 0.019(0.009) **

• 0.026(0.011) ***

*

Airplayt-6

• 0.011(0.012)

0.048(0.026) *

• 0.003(0.010)
• 0.005(0.011)

4/15/2011 21

Note: Standard Deviations are in parentheses.

PVAR Regression Results: Sample Splits PVAR Regression Results: Sample Splits

Dependent Variable Song Sales Album Sales High Artist Rep. Low Artist Rep. High Artist Rep. Low Artist Rep. Buzzt-1 0.023(0.119)

• 0.105(0.045) ***
• 0.086(0.225)

0.021(0.010) *** Buzzt-2

• 0.009(0.069)
• 0.089(0.033) ***
• 0.080(0.177)

0.026(0.008) *** Buzzt-3

• 0.006(0.040)
• 0.067(0.025) ***
• 0.087(0.140)
• 0.004(0.007)

Buzzt-4 0.002(0.030)

• 0.054(0.020) ***
• 0.071(0.152)

0.003(0.006) B 0 013(0 033) 0 051(0 020) *** 0 037(0 112) 0 021(0 006) *** Buzzt-5 0.013(0.033)

• 0.051(0.020) ***
• 0.037(0.112)

0.021(0.006) *** Buzzt-6

• 0.039(0.030)
• 0.053(0.016) ***
• 0.034(0.053)

0.017(0.006) *** Airplayt-1 0.075(0.081) 0.070(0.021) *** 0.334(0.674) 0.082(0.022) *** Airplay 0 034(0 048) 0 049(0 015) *** 0 179(0 429) 0 013(0 015) Airplayt-2 0.034(0.048) 0.049(0.015) 0.179(0.429) 0.013(0.015) Airplayt-3 0.037(0.052)

• 0.017(0.013)

0.050(0.179)

• 0.019(0.011) *

Airplayt-4

• 0.003(0.041)
• 0.007(0.012)

0.083(0.153) 0.001(0.009) Airplayt-5

• 0.010(0.036)

0.003(0.012) 0.038(0.107) 0.001(0.008)

4/15/2011 22

p yt-5 ( ) ( ) ( ) ( ) Airplayt-6 0.031(0.028) 0.002(0.011) 0.042(0.160)

• 0.002(0.008)

Note: Standard Deviations are in parentheses.

Impulse Response Functions Impulse Response Functions

• Shows the response of one variable (e g sales)

Shows the response of one variable (e.g. sales) to a shock in another variable (e.g. buzz), while holding all other shocks at zero while holding all other shocks at zero.

• Does the shock have a permanent or transitory

impact? impact?

4/15/2011 23

Impulse Response Functions (IRFs) Impulse Response Functions (IRFs)

(p 5) songbuzz_log songbuzz_log (p 95) songbuzz_log 0.5098 (p 5) songsales_log songsales_log (p 95) songsales_log 0.0152 (p 5) airplay_log airplay_log (p 95) airplay_log 0.0293

response of songbuzz log to songbuzz log shock

s 6 0.0000

response of songbuzz log to songsales log shock

s 6

• 0.0348

response of songbuzz log to airplay log shock

s 6

• 0.0125

response of songbuzz_log to songbuzz_log shock response of songbuzz_log to songsales_log shock response of songbuzz_log to airplay_log shock

(p 5) songbuzz_log songbuzz_log (p 95) songbuzz_log 0.0031 (p 5) songsales_log songsales_log (p 95) songsales_log 0.5141 (p 5) airplay_log airplay_log (p 95) airplay_log 0.0617

response of songsales_log to songbuzz_log shock

s 6

• 0.2105

response of songsales_log to songsales_log shock

s 6 0.0000

response of songsales_log to airplay_log shock

s 6 0.0000 (p 5) songbuzz_log songbuzz_log (p 95) songbuzz_log 0.0019 (p 5) songsales_log songsales_log (p 95) songsales_log 0.0579 (p 5) airplay_log airplay_log (p 95) airplay_log 0.4069

response of airplay log to songbuzz log shock

s 6

• 0.1401

response of airplay log to songsales log shock

s 6

• 0.0015

response of airplay log to airplay log shock

s 6 0.0000

4/15/2011 24 Errors are 5% on each side generated by Monte-Carlo with 500 reps

response of airplay_log to songbuzz_log shock response of airplay_log to songsales_log shock response of airplay_log to airplay_log shock

IRFs: Song Level IRFs: Song Level

Response of Song Sales to:

0.5 0.6

• ck

Airplay Song Buzz

0 2 0.3 0.4

g sales to sho

0.1 0.2 1 2 3 4 5 6

sponse of son

• 0.2
• 0.1

1 2 3 4 5 6

Res Number of weeks post-shock

4/15/2011 25

IRFs: Album Level IRFs: Album Level

Response of Album Sales to:

0 04 0.045 0.05

shock Album Buzz Airplay

0.025 0.03 0.035 0.04

bum sales to s

0 005 0.01 0.015 0.02

Response of alb

0.005 1 2 3 4 5 6

R Number of weeks post-shock

4/15/2011 26

IRFs: Album and Song Level Buzz IRFs: Album and Song Level Buzz

Response of Sales to Buzz

0.05

to shock Album Sales to Album Buzz Song Sales to Song Buzz

• 0.05

1 2 3 4 5 6

album sales t

• 0.1

nse of song &

• 0.15

Respon Number of weeks post-shock

4/15/2011 27

IRFs: Album and Song Level Airplay IRFs: Album and Song Level Airplay

Response of Sales to Airplay

0 45 0.55

shock Album Sales to Airplay Song Sales to Airplay

0.35 0.45

song sales to s

0.15 0.25

nse of album/s

• 0.05

0.05 1 2 3 4 5 6

Respon Number of weeks post-shock

4/15/2011 28

Results: Song Level Results: Song Level

• Results largely consistent for independent

label & low artist reputation

• No significant relationships for high artist

g p g reputation sample split

Buzz Sales Airplay Buzz Negative Negative Sales Positive Sales Positive Airplay Positive Positive

4/15/2011 29

Results: Album-Level Results: Album Level

• Results largely consistent for independent

g y p label & low artist reputation

• No significant relationships for high artist

No significant relationships for high artist reputation sample split

Buzz Sales Airplay Buzz Positive Positive Sales Positive Airplay Positive

4/15/2011 30

Interpretation of Results Interpretation of Results

• Main Findings:

g

– Song Buzz seems to decrease future song sales – Album Buzz seems to increase future album sales – Airplay increases both song and album sales Airplay increases both song and album sales instantaneously; this is larger for songs

• Explanation:

A h k i bl b i ht b l t d ith – A shock in blog buzz might be correlated with a contemporaneous shock in the supply of free shareable music Media sharing happens less at the album level – Media sharing happens less at the album level – Those who consume albums are different from those who consume songs in their propensity to pirate vs. purchase.

4/15/2011 31

Next steps Next steps

• Collect additional data to verify “media

Collect additional data to verify media sharing” explanation

• Examine relationships between song and
• Examine relationships between song and

album-level variables F h l li l d

• Further sample splits: release date, genre
• Consider other types of social media, e.g.

Tweets, Facebook Fans, etc.

4/15/2011 32

THANK YOU! THANK YOU!

4/15/2011 33