New release: an empirical analysis of VHS/DVD rental success.

Introduction

Prior to the introduction of the video cassette recorder (VCR) in 1980, movie consumers wishing to view a film had no choice but to attend the theater. Movie showings existed in limited quantities within bounded time periods. Consumer decisions had to be made quickly, and a decision to not see a film could not be later amended. Once a movie was pulled from the screens, it was gone forever, in most cases.

According to the Video Software Dealers Association (1996), there was considerable growth in the home video industry during the 1980s, which made it the largest domestic revenue source for movie studios. Specifically, in 1999, the $16 billion home video industry represented 55% of studios' domestic revenues, while box office revenues were 22%, and the remaining 23% came from all other forms of media, including sales of pay-per-view, cable, and broadcast television rights (Video Software Dealers Association 2000).

In 2001, motion pictures grossed a new all-time domestic box office high of $8.4 billion (MPAA 2002), an increase of nearly 10% over the year 2000. While such a figure is impressive, film makers enjoy additional revenues once they release films for rental. In 2001, rental revenues surpassed box office receipts, with Video Home System (VHS) rentals totaling $7.02 billion and Digital Versatile Disc (DVD) rentals grossing over $1.4 billion (Dretzka 2003). During the same year, Americans spent an estimated average of $109.60 and 56 h on home video entertainment per person. Given these numbers, it is clear that the demand for home theater is robust and should not be ignored.

A consumer may view a movie in a theater for a relatively low marginal cost and no associated fixed cost; however, while movie rentals may be purchased at an even lower marginal cost, complementary technological player requirements have, historically, held back the growth of the rental market. Nevertheless, improvements in technology coupled with greater competition in the VCR and, more recently, the DVD player markets have resulted in falling prices of VCRs and DVD players in recent years, making the fixed cost of acquiring this technology much less burdensome to consumers. Data indicates near-total penetration of VCRs into television households was achieved by the end of 2001, since 91.2% of television owners also owned a VCR (MPAA 2002). Consumers subsequently adopted DVD player technology relatively quickly. Bakalis (2003) points out, by 2003, roughly 50 million Americans had purchased a DVD player, since its introduction in 1997; it took a decade for VCR sales to reach the same level. Furthermore, DVD rentals overtook VHS rentals in June 2003 for the first time.

Despite the obvious popularity of this form of entertainment and the fact that the majority of film studios' domestic revenues are generated by home video viewing, a serious dearth of studies analyzing the success of films in the video rental market exists. The present research addresses this gap through the construction of an empirical model analyzing rental revenue for both VHS and DVD media, as measured by gross rental revenue, based on a number of available indicators. In particular, this study includes an explicit measure of word-of-mouth dissemination of information across consumers, proxy variables measuring the impact of movie star popularity, movie critic reviews, and movie awards, control variables for genre, Motion Picture Association of America (MPAA) rating, production budget, and economic variables. Given that research regarding the rental market remains limited, much of the present paper provides comparisons to the comparatively more developed body of literature about the movie industry and box office revenue.

Literature Review

The literature on movies seen in theaters is extensive and identifies several variables important in predicting the box office success of a film. The presence of popular actors in films is examined by Prag and Casavant (1994), Bagella and Becchetti (1999), Simonoff and Sparrow (2000), Kelwick (2002), and De Vany (2004). Production budgets are studied by Simonoff and Sparrow (2000) and De Vany (2004). The importance of award nominations and wins is investigated by Prag and Casavant (1994), Kelwick (2002), and De Vany (2004). Prag and Casavant (1994) and Ravid (2004) emphasize critic reviews in determining box office revenue. Kelwick (2002) and De Vany (2004) address the role of word-of-mouth information flow in the success or failure of a movie. Many of these variables also hold promise for explaining the performance of movies in the rental market.

Studies of the video rental market primarily have focused on characteristics of the industry itself rather than on the actual movie rental rates. Dana and Spier (2001), and Cachon and Lariviere (2005) focus on revenue-sharing contracts in the video rental industry. Lehmann and Weinberg (2000) examine the sequential distribution channels characteristic for movies and videos. Dana (2001) studies the relationship between price and availability of video rentals using Cournot and Bertrand models. However, there is an obvious gap in the literature when it comes to understanding the rental patterns of films, which is noted by Lehmann and Weinberg (2000).

Data

This study analyzes a cross section of the top 100 grossing domestic films in the USA for the year 2001 (Nash 2002). A lack of availability of data for rental revenue and word-of-mouth variables limits the sample to this year. While this study centers on the rental market, competition between films begins in the box office. Consequently, the sample originates from box office rankings and not rental rankings. All films re-released after an initial release in a previous year are excluded from the sample. Image Maximum (IMAX) films also are excluded from the data set, because IMAX screen constraints have prevented their widespread availability. In addition, films with missing data observations are omitted from the sample.

We recognize that using a censored sample like the one described can bias our regression coefficients. Nevertheless this practice has precedent in the movie literature. Smith and Smith (1986), De Vany and Walls (1997), and Lehmann and Weinberg (2000) all use samples of similarly high-performing films in their studies.

Dependent Variables

Two dependent variables are reported: revenue earned through rentals of VHS copies of each film and revenue earned through rentals of DVD copies of each film. (2) Using the top rental charts released weekly by BoxOfficeMojo.com, each film is tracked from its first appearance on the charts through its last appearance. The film's cumulative rental revenue at this point is then recorded. Thus, the revenue values in our data set represent each film's rental revenue captured throughout its respective strongest earning period; Lehmann and Weinberg (2000) indicate most revenues are earned in the first year of release. An alternate approach of recording revenue earned as of a specified date is not pursued, because this introduces bias attributable to different rental market release dates for the movies in the sample. All VHS revenue comes from a weekly top 40 chart, whereas the DVD revenue is released in a weekly top 25 list.

All rental revenues are recorded in 2001 dollars. Any rental revenue earned in 2002 is first removed from total rental revenue and then deflated appropriately using the Bureau of Labor Statistics (BLS) inflation calculator. The deflated 2002 revenue is then added back to the 2001 revenue to create the appropriate total rental revenue measured in 2001 dollars.

Independent Variables and Hypotheses

The independent variable, GROSS, represents each film's final domestic box office gross revenue in millions of real US dollars accumulated during the initial run in theaters. Revenue generated by sales or rentals of VHS or DVD copies of the film is not included. Films with greater box office gross revenue may have higher rental revenue for several reasons. Strong box office performance may signal higher film quality to renters. A film with a high box office gross likely also has had a relatively lengthy run in theaters during which time the film remains the subject of both media coverage and word-of-mouth discussion. Potential renters may more readily choose a film based on past name recognition.

The BUDGET variable is the total millions of real US dollars used to produce a film. The variable is vital because of its ability to proxy other hard-to-find variables. The production budget enfolds wages paid to actors and directors along with costs associated with special effects or exotic film locations. Unfortunately, expenditure on film promotional advertising is not included in our data set, since advertising is budgeted as a separate line item by studios and is considered proprietary. Reported production budgets also do not reflect any contractual performance-based wages paid to actors after a film's release. We expect BUDGET to have a positive relationship with each of the rental revenue dependent variables.

US personal income for the month of a film's rental release, INCOME, is included in the model. Renting movies is assumed to be a normal leisure activity, so a positive relationship between 1NCOME and rental revenue is expected.

A variable measuring price is included in the model. A Consumer Price Index (CPI) for rentals of videocassettes and discs is available from BLS for the month of each movie's release. Thus, our CPIRENT variable proxies the price of movie rentals. A priori, we expect that the CPIRENT variable will have a negative coefficient.

Critic reviews are taken from a popular website, RottenTomatoes.com (2004). The site reports the percentage of good reviews a film receives from a number of respected nationwide critics. The CRITIC variable is the percentage reported. Critic reviews inform consumers about a film's quality and content, so CRITIC is expected to have a positive relationship with the dependent variables. The CRITIC variable included in this model is similar to variables in the work of both Simonoff and Sparrow (2000) and Prag and Casavant (1994), which focused on Roger Ebert and a compilation of critics, respectively.

The variable STARPWR captures the drawing power of prominent Hollywood stars. Other researchers such as De Vany and Walls (1999) have used lists of Hollywood's "Most Powerful People" published by Premiere Magazine and James Ulmer's list of A/A + people. These lists are lengthy and include some individuals who, in our estimation, ought to have substantially less drawing power than others on the list. As a result, we have chosen another approach.

Our STARPWR variable is based on the annual Top Ten Favorite Movie Star lists published by the Harris Poll and the annual People's Choice Award nominees for four categories: Motion Picture Actor, Motion Picture Actress, Male Television Performer, and Female Television Performer. There are three People's Choice Award nominees for each category. A nationwide poll conducted by the Gallup Organization determines both the nominees and winners. Each poll involves a sample size between 1,000-2,225 respondents, and responses are weighted to reflect the average demographics of the US population. We have compiled a master list (3) across the top ten Harris Poll lists for the years 1995-2002 and the People's Choice nominations for the broadcast years 1997-2002. While our movie rental data is from the year 2001, we believe backdating and postdating allows for lagged and emerging popularity effects. Due to overlaps from year-to-year and across the Harris Poll and People's Choice nominations, our list consists of 66 high profile stars. (4) The stars in our master list all had previous success before the year in which their names appeared in the polls. The STARPWR variable is a count variable representing the number of individuals from the master list who appear in a given film.

The genres included in the model are Science Fiction, Comedy Drama, Drama, Comedy, Action Adventure, Horror, and Animation. Each enters the model as a dummy variable, and each is expected to have an ambiguous effect on each dependent variable. The Animation genre is the omitted condition.

MPAA Ratings included in the model are General Audiences (G), Parental Guidance Suggested (PG), Parents Strongly Cautioned (PG-13), and Restricted (R). Each is set as a dummy variable, and each is expected to have an ambiguous effect on each dependent variable. The G-rating is the omitted condition.

Kelwick (2002) and De Vany (2004) stress the importance of word-of-mouth recommendations as a factor in film performance because they have the potential to spawn bandwagon effects. We implement word-of-mouth information transfer in our model by using the poll service Cinemascore. Each Friday, Cinemascore conducts exit polls of 400-500 moviegoers as they leave the theater after viewing a film in 14 or 15 sites that the company routinely surveys. The viewers are asked to assign a letter grade to the film they just watched. Cinemascore then aggregates the national scores and posts the results on its website. We have translated the letter grades into numeric data (A + = 15, A = 14, A - = 13 ... F - = 1). The variable CINEMASCORE is expected to positively affect the dependent variables.

Major award nominations also are tested for relation to rental revenue. Similar categories are observed across four prominent awards: The Academy Awards, The Golden Globes, The (Orange) British Academy of Film Awards, and the Screen Actors Guild Awards. The award nominations studied for the model are Best Picture, Best Actor, Best Actress, Best Supporting Actor, Best Supporting Actress, Best Director, and Best Original Screenplay. The award nominations are released and the award shows are broadcast at different times during the year; each award nomination announcement and award show broadcast garners considerable media attention. An initial approach in our study involved creating separate dummy variables for each type of award, but the award dummies are collinear with one another. Hence a single award count variable, TOTFILMNOM, is created to account for each time a film was nominated or contained a nominee for one of the above award categories. The count range for this variable is 0 - 22. We expect that TOTFILMNOM positively affects rental revenue.

Descriptive Statistics

Table 1 presents descriptive statistics for the sample. In descending order, the three highest rental revenue films among VHS rentals are Don't Say a Word, Legally Blonde, and The Wedding Planner. The three lowest earning VHS rentals in our sample in descending order are In the Bedroom, Two Can Play That Game, and Amelie. The highest grossing VHS rental earned revenues of $54.35 million while the lowest grossing rental earned $2.65 million. The standard deviation for gross VHS rental revenue is $12.01 million. In descending order, the three highest performing films among DVD rentals are Ocean's Eleven, Training Day, and Don't Say a Word. The three lowest earning DVD rentals in our sample in descending order are The Brothers, Recess: School's Out, and See Spot Run. The highest grossing DVD rental earned revenues of $32.74 million. The standard deviation for gross DVD rental revenue is $6.54 million.

The average box office gross for a film in the sample is $73.42 million, and the average production budget is $46.69 million. This indicates that many of the films in the sample enter the rental market as successes. The average VHS rental revenue for the sample is $26.78 million, while the average DVD rental revenue for the sample is $11.94 million.

Films receive an average of only 48.5% positive critical reviews, but this sentiment is not echoed by theatergoers, who respond with a B+ average when polled by Cinemascore. 85% of the films in the sample are rated either PG-13 or R. 65% fall into the genres of comedy and drama. The average time spent on the VHS chart is about 16 weeks, while the average time spent on the DVD chart is about ten-and-a-half weeks. A film is released in the rental market an average of five-and-a-half months after its theatrical run begins.

Model

The final estimated models use a double-log functional form and are regressed using Ordinary Least Squares (OLS). According to Fernandez Blanco and Banos Pino (1997), the double-log specification is very common when estimating demand functions for unrelated goods. The Log(VHSGROSS) and Log(DVDGROSS) dependent variables are logged due to their long fight-tailed nature (Simonoff and Sparrow 2000). Using the log of the rental revenues solves the problem of the dependent variables' skewed distributions. The same logic is applied to the GROSS, BUDGET, and INCOME variables, so they enter the equations as logarithms.

The inclusion of the MPAA rating variables in the regressions is confirmed by F-tests at the 5% level (F-test statistics are 5.63 for VHS and 9.05 for DVD). The inclusion of the genre variables in the VHS is confirmed by an F-test at the 5% level (F-test statistic is 2.81); however, the inclusion of the genre variable group for the DVD regressions is rejected at the 5% level (F-test statistic is 1.91). Nevertheless, due to the individual statistical significance of certain genres and their theoretical relevance, we do include the genre variables in the DVD regressions.

The regression models are re-specified to account for possible simultaneity bias between the different rental revenue dependent variables (VHSGROSS, DVDGROSS) and the GROSS independent variable using the procedure developed by Hausman (1978). The variable, Log(GROSS), is determined to be a function of the logarithm of peak screens the movie played on during its initial release in theaters, Log(INCOME), the logarithm of the box office gross of any prequels in a series of films, a CPI measure for movie ticket price, CRITIC, STARPWR, TOTFILMNOM, Genres, MPAA Ratings, Summer and Holiday release date dummies, and an indicator of a preexisting audience because of a book, comic, television show, etc. The use of the logarithm of peak screens as a determinant of Log(GROSS) instead of the logarithm of production budget is supported by De Vany (2004) and other researchers.

A reduced-form equation for Log(GROSS) is obtained using the equations for both types of rental revenue and is estimated using OLS. The residual value for Log(GROSS) is calculated and named SCREENRESID, and it is entered directly into the OLS regressions for the rental revenue measures. Notably, the t-statistics on the SCREENRESID variable in the two regressions indicate that the regressor is not statistically different from zero at any conventional level of significance. Thus, we conclude that there is no simultaneity problem. The regressions reported as Model 1 for each dependent variable in Table 2 do not account for the potential simultaneity bias, while the regressions reported as Model 2 do include SCREENRESID.

Results

As seen in Table 2, the regression results across Models 1 and 2 for each dependent variable are uniformly consistent in sign and almost always are consistent in significance level. Hence, we choose to focus on Model 1 results in this section.

VHS Revenue Results

The VHS results presented in Table 2 have the logged gross revenue of VHS rentals as the dependent variable. The adjusted [R.sup.2] for the regression is 0.582. The estimate on the logged box office gross revenue variable is positive and significant, as expected, at the 1% level. All else equal, for every 1% increase in a film's box office gross, an average increase in VHS rental revenue of 0.54% is expected. This result indicates a notable carry-over effect from theaters to video stores, but one characterized by diminishing returns. The estimated coefficient on Log(BUDGET) is not statistically significant in this regression.

The estimate on the first of the two economic variables, Log(INCOME), is negative and insignificant. The negative sign paints VHS cassettes as inferior goods. This is logical, because viewing technology has advanced beyond VCRs to DVD players. People with higher incomes choose the more advanced technology. The other economic variable, CPIRENT, has a negative and insignificant coefficient and this negative sign is consistent with the basic law of demand.

The sign on the CRITIC estimate is unexpected. The fact that it is negative and insignificant indicates that while consumers may be more likely to take a chance renting a lower-quality film, perhaps due to its low price, it is not a consistent preference. The statistical insignificance of this variable provides support for the Prag and Casavant (1994) notion that the opinions of professional movie critics do not necessarily mirror those of the layman.

CINEMASCORE has a similar negative and insignificant coefficient as CRITIC. This insignificance indicates that consumers may not use the same word-of-mouth information passed on from theater-goers when making rental decisions. Due to the relatively low financial risk associated with renting, consumers may be less inclined to seek and/or heed the advice of others.

The coefficient on STARPWR is positive and significant at the 5% significance level in accordance with the findings of De Vany (2004) and other researchers. Ravid (2004) finds an insignificant result for stardom. For each additional star, VHS rental revenue increases 6% on average.

TOTFILMNOM has a negative and insignificant effect, which indicates that film nominations have little influential power with respect to renters. The coefficients of DRAMA and HORROR are both positive and significant, which suggests that dramas and horror films perform better than animated pictures, on average, in the rental market. Specifically, horror films and dramas garner an average of 17-19% higher revenue than animated films.

Of the MPAA ratings, both R and PG-13 have positive and significant coefficients. These results suggest that PG-13 and R-rated movies perform about 20-23% better, on average, than G-rated films. There is obviously a preference for more mature subject matter among renters.

DVD Revenue Results

The DVD results in Table 2 have the logged gross revenue of DVD rentals as the dependent variable. The adjusted [R.sup.2] for the regression is 0.659. As in the VHS regressions, the impact of Log(GROSS) is positive and significant at the 1% significance level. The elasticity between DVD rental revenue and box office gross is 0.5. This further validates the link between box office and rental success. Log (BUDGET) has a significant elasticity of 0.2 in this model. One possible interpretation of this result is that special effects, exotic filming locations, and other expensive production elements afforded by films with large production budgets translate better to the high resolution picture of DVDs and/or to the special feature add-ons on DVDs.

Log(INCOME) has a positive coefficient which is significant at the 1% level, an opposite result from the VHS regressions. This result suggests that DVD rentals are normal goods. The size of this increase is considerable based on the estimated elasticity of 27. (5) CPIRENT has an unexplained positive sign, though it is insignificant.

CRITIC is once again negative and statistically insignificant, providing further evidence of the lack of influence that critics have on renters. STARPWR retains a positive coefficient that is significant at the 5% level, similar to its coefficient in the VHS regressions. TOTFILMNOM and CINEMASCORE both have negative and insignificant coefficients like they do in the VHS regressions. Of the genres, DRAMA and COMEDY are statistically significant, and their positive coefficients are comparable. While action-adventure pictures are more likely to have special effects which benefit from the better resolution of DVDs, they do not perform better than in the VHS regressions. MPAA ratings replicate the results of the VHS model, but with much larger impacts. PG-13 and R-rated movies, on average, earn rental revenues that are approximately 52% larger than G-rated films.

Comparing and Contrasting VHS and DVD Results

Box office gross revenue exerts a statistically significant positive influence on rental revenue throughout the analysis. The estimated elasticities range from 0.54 to 0.57. This finding reveals a correlation between financial success in theaters and financial success in video stores. The statistical insignificance of CPIRENT throughout the regression models may be due to a measurement error in the creation of the CPI.

The estimated coefficients on Log(INCOME) are among the most interesting results. In particular, the negative sign on the Log(INCOME) coefficient in the VHS regressions suggests that renting VHS movies is an inferior good. Given the degree of VCR saturation in 2001 and the competition that VHS faces from the increasing diffusion of the technologically advanced DVD format, it is reasonable that VHS video rentals have become less desirable to those renters in the upper income brackets. Additionally, the large, positive, and significant estimates on Log (INCOME) in the DVD regressions support the vigorous emergence of DVD movie rentals as a normal good.

For the foregoing reasons, both of the economic variables are retained in the final specifications of the model. Bagella and Becchetti (1999) find statistically insignificant impacts of their price and income variables as this study does for the VHS models. The prevalent lack of significance of the CPIRENT variable across all models in this study most likely reflects inadequacies in the variable's measurement more than a true lack of pertinence.

The results for the included genres are mixed across the regressions. Across all models, only dramas consistently exert a positive and significant impact on rental revenues relative to animated films. Interestingly, HORROR, which is significant at the 5% level in each of the VHS regressions, actually performs much worse in the DVD regressions. This outcome may be consistent with our findings that VHS rentals are inferior goods and DVD rentals are normal goods, if consumers with higher income levels are less likely to find HORROR films appealing.

Of the MPAA ratings, only the R and PG-13 estimates are positive and significant relative to G. According to Prag and Casavant (1994), the conventional wisdom in Hollywood is that R and PG-13 films yield higher revenues thanks to a mature audience who can afford movie ticket prices. In the case of movie rentals, it seems reasonable that consumers will gravitate toward renting films with mature content to view in the privacy of their own homes. The consistently large coefficients on the R and PG-13 variables testify to the strength of this effect.

The negative and statistically insignificant impact of CINEMASCORE is a notable result. It questions the role of word-of-mouth information transfer in the success or failure of a movie in the rental market. Perhaps, this is a function of the lower price of movie rental compared to theater admission or evidence that diminishing returns to word-of-mouth recommendations have long since set in by the time a film is released on video. The consistent lack of significant estimates on CRITIC and TOTFILMNOM suggests that acclaim is similarly impotent in the rental market. Smith and Smith (1986) and Prag and Casavant (1994) obtain a positive and significant impact of the total number of Academy Awards received by a film on its box office success. What appears to matter more to renters is the appearance of a favorite star in a film, captured by our STARPWR variable. Throughout the regressions, each additional star from our list in a movie leads to a 5-6% average increase in rental revenue.

The Log(BUDGET) variable only achieves significance in the DVD regression, where it is positive with a coefficient of 0.2. Such a result indicates decreasing returns on the rental market to a studio's investment in a film, and that DVD rental revenue is inelastic relative to changes in the budget variable. Of particular relevance to DVD rentals, De Vany (2004) points out that larger budgets are not necessarily obvious on the picture screen in terms of better production value and that outcome becomes more likely as budgets grow larger and larger. This observation may be even more pertinent to the display of a film on a small television screen, where the diminishing returns of costly special effects may set in far sooner.

The adjusted [R.sup.2] values obtained in our study lie at the high end of the ranges reported by studies of box office performance. For those researchers reporting either unadjusted or adjusted [R.sup.2] measures for similar regression analyses, most of the values range from a low of roughly 0.09 to a high of about 0.7, with a concentration of values in the vicinity of 0.3 to 0.4. Our results range between 0.58 and 0.66.

Conclusions

Our regression results indicate that box office gross revenue, stardom, and the mature MPAA ratings of R and PG-13 exert positive and statistically significant effects on film rental success. Dramas, and occasionally other genres, are also positive and significant determinants of rental performance.

A few interesting observations and key differences emerge when comparing VHS and DVD formats. Based on the results, it appears that only DVD rentals are normal goods. The magnitude of the impact of personal income on DVD rental revenues is very large and an important finding. In addition, the size of a film's production budget is a significant determinant of DVD rental success and not VHS rental success, which may be linked to higher resolution and/or add-on special features associated with DVDs.

Perhaps what is most noteworthy about movie rentals relative to cinema features is the lack of importance placed on award nominations, the opinions of critics, and word-of-mouth praise in the financial success of rental films. It appears that indications of film quality and approbation from the motion picture industry, as well as from the average moviegoer, have no appreciable impact on rental gross revenue.

Acknowledgements The authors gratefully acknowledge the superb research assistance of Stephanie Perkovich. This study has benefited from suggestions offered by David Eaton.

This paper was presented at the annual meeting of the Southern Economic Association, New Orleans, LA, November 2004. The usual disclaimer applies.

References

Bagella, M., & Becchetti, L. (1999). The determinants of motion picture box office performance: Evidence from movies produced in Italy. Journal of Cultural Economics, 23(4), 237-256.

Bakalis, A. (2003). It's unreel: DVD rentals overtake videocassettes. Retrieved June from http://www. washtimes.com/business/20030620-113258-1104r.htm.

Cachon, G., & Lariviere, M. (2005). Supply chain coordination with revenue-sharing contracts: Strengths and limitations. Management Science, 51(1), 30 44.

Dana, J. (2001). Competition in price and availability when availability is unobservable. RAND Journal of Economics, 32(4), 497-513.

Dana, J., & Spier, K. (2001). Revenue sharing and vertical control in the video rental industry. Journal of Industrial Economics, 49(3), 223-245.

De Vany, A. (2004). Hollywood economics: How extreme uncertainty shapes the film industry pp. 71-138. New York, NY: Routledge.

De Vany, A., & Walls, D. (1997). The market for motion pictures: Rank, revenue, and survival. Economic Inquiry, 35(4), 783-797.

De Vany, A., & Walls, D. (1999). Uncertainty in the movie industry: Does star power reduce the terror of the box-office? Journal of Cultural Economics, 23(4), 285-318.

Dretzka, G. (2003). Digital nation at movie city news. Retrieved March from http://www.moviecitynews. com/columnists/dretzka/2003/030114.html.

Fernandez Blanco, V., & Banos Pino, J. (1997). Cinema demand in Spain: A cointegration analysis. Journal of Cultural Economics, 21(1), 57-75.

Hausman, J. A. (1978). Specification tests in econometrics. Econometrica, 46(6), 1251-1271.

Kelwick, J. (2002). The usher Just one opinion. Retrieved September from http://www.kelwick.karoo. net/index.htm.

Lehmann, D., & Weinberg, C. (2000). Sales through sequential distribution channels: An application to movies and videos. Journal of Marketing, 64(3), 18-33.

MPAA (2002). U.S. entertainment industry: 2002 MPA market statistics. Retrieved September from http:// www.mpaa.org.

Nash, B. (2002). The numbers: Box office data, movie stars, idle speculation. Retrieved October from http://www.the-numbers.com.

Prag, J., & Casavant, J. (1994). An empirical study of the determinants of revenues and marketing expenditures in the motion picture industry. Journal of Cultural Economics, 18(3), 217-235.

Ravid, S. A. (2004). Are they all crazy or just risk averse? Some movie puzzles and possible solutions. In V. A. Ginsburgh (Ed.) Contributions to economic analysis (pp. 33-47). New York, NY: Elsevier.

Rotten Tomatoes (2004). Retrieved March 22, 2004, from http://www.rottentomatoes.com.

Simonoff, J. S., & Sparrow, I. R. (2000). Predicting movie grosses: Winners and losers, blockbusters and sleepers. Chance, 13(3), 15-24.

Smith, S. P., & Smith, V. K. (1986). Successful movies: A preliminary empirical analysis. Applied Economics, 18(5), 501-507.

Video Software Dealers Association (1996). An annual report on the home video market 1996. Northridge, CA: Corporate Creativity.

Video Software Dealers Association (2000). An annual report on the home video market 2000. Northridge, CA: Corporate Creativity.

J. J. Jozefowicz ([mail]) * J. M. Kelley * S. M. Brewer Department of Economics, Indiana University of Pennsylvania, Indiana, PA 15705, USA e-mail: James.Jozefowicz@iup.edu

J. M. Kelley

e-mail: J.M.Kelley@iup.edu

S. M. Brewer

e-mail: Stephanie.Jozefowicz@iup.edu

(1) A complete list of film titles included in the sample is available from the authors upon request.

(2) A third dependent variable for the combined VHS and DVD rental revenue has been examined. The results closely mirror those reported for VHS rentals, which is not surprising since VCRs still had much deeper penetration for the sample year 2001.

(3) A complete list of the stars to create STARPWR is available from the authors upon request.

(4) While causality between star popularity and blockbuster hit movies potentially is an issue, we believe that by pooling the polls across years we have minimized this effect. Additionally, we believe causality is a bigger issue when a star is previously unknown and is involved in a blockbuster hit that causes his/her popularity to explode.

(5) It is conceivable that this result is biased upward, because the sample is censored and does not reflect films outside the top 100 list that may ultimately find success in the rental market.

Table 1 Descriptive statistics

Variable                  Mean        SD        Max       Min

VHSGROSS               $26.78M    $12.01M    $54.35M    $2.65M
DVDGROSS               $11.94M     $6.54M    $32.74M    $0
GROSS                  $73.42M    $64.24M   $318M      $21.93M
BUDGET                 $46.69M    $32.39M   $153M       $1.7M
INCOME                  $8.79T    $87.1B      $8.92T    $8.68T
SCREENS               2519.67     708.72    3715       303
CPIRENT                 93.89       2.14      96.40     87.10
CRITIC                  48.51      24.17      95         5
STARPWR                  0.28       0.62       4         0
CINEMASCORE             11.38       1.86      15         4
TOTFILMNOM               1.10       2.83      15         0
MONTHS SINCE RELEASE     5.57       1.39      10         3
ACTADV                   0.12       0.33       1         0
ANIM                     0.06       0.24       1         0
COMDRA                   0.11       0.31       1         0
COMEDY                   0.29       0.46       1         0
DRAMA                    0.36       0.48       1         0
HORROR                   0.04       0.20       1         0
SCIFI                    0.02       0.14       1         0
G                        0.04       0.20       1         0
PG                       0.11       0.31       1         0
PG-13                    0.45       0.50       1         0
R                        0.40       0.49       1         0

Table 2 VHS and DVD rental revenue (ordinary least squares results)

Independent Variable       Log(VHS Revenue)

                            Model l (a)             Model 2 (a)

LOG(GROSS)                0.5439 (e) (8.111)      0.5724 (e) (5.332)
LOG(BUDGET)              -0.0462 (-0.731)        -0.0565 (-0.804)
LOG(INCOME)              -5.3128 (-1.309)        -5.0673 (-1.223)
CPIRENT                  -0.0031 (-0.373)        -0.0028 (-0.341)
CRITIC                   -0.0012 (-1.498)        -0.0014 (-1.488)
CINEMASCORE              -0.0022 (-0.245)        -0.0029 (-0.313)
STARPWR                   0.0600 (d) (2.414)      0.0598 (d) (2.390)
TOTFILMNOM               -0.0074 (-1.026)        -0.0079 (-1.068)
ACTADV                    0.0186 (0.257)          0.0177 (0.243)
DRAMA                     0.17024 (2.535)         0.1752 (d) (2.535)
COMDRA                    0.1237 (1.535)          0.1285 (1.562)
COMEDY                    0.1044 (1.619)          0.1058 (1.628)
HORROR                    0.1919 (d) (2.081)      0.1908 (d) (2.056)
R                         0.1919 (d) (2.031)      0.1980 (d) (2.033)
PG                        0.0378 (0.412)          0.0388 (0.419)
PG-13                     0.2335 (d) (2.457)      0.2330 (d) (2.436)
SCREENRESID                                      -0.0445 (-0.342)
CONSTANT                 72.3506 (1.386)         69.0202 (1.292)
Adjusted [R.sup.2]        0.582                   0.577
N                        93                      93

Independent Variable   Log(DVD Revenue)

                            Model l (b)             Model 2 (b)

LOG(GROSS)                0.5669 (e) (6.734)      0.6417 (d) (4.961)
LOG(BUDGET)               0.2043 (c) (1.952)      0.17671 (1.563)
LOG(INCOME)              27.4321 (e) (4.798)     28.0680 (e) (4.850)
CPIRENT                   0.0006 (0.055)          0.0012 (0.102)
CRITIC                   -0.0001 (-0.076)        -0.0004 (-0.416)
CINEMASCORE              -0.0101 (-0.716)        -0.0119 (-0.800)
STARPWR                   0.0543 (1.599)          0.0538 (1.595)
TOTFILMNOM               -0.0134 (-1.255)        -0.0147 (-1.391)
ACTADV                    0.1334 (1.192)          0.1312 (1.138)
DRAMA                     0.2232 (c) (1.907)      0.2364 (d) (2.060)
COMDRA                    0.0796 (0.542)          0.0920 (0.638)
COMEDY                    0.2004 (c) (1.700)      0.2042 (1.722)
HORROR                    0.2333 (1.550)          0.2306 (1.502)
R                         0.5285 (d) (2.137)      0.5324 (d) (2.082)
PG                        0.1902 (0.833)          0.1935 (0.816)
PG-13                     0.5261 (d) (2.225)      0.5247 (d) (2.137)
SCREENRESID                                      -0.1167 (-0.705)
CONSTANT               -354.6051 (-4.808)      -363.2301 (-4.866)
Adjusted [R.sup.2]        0.659                   0.656
N                        92                      92

(a) t-statistics appear in parentheses.

(b) t-statistics in parentheses are based on White
heteroskedasticity-consistent standard errors.

(c) Significant at 10% level

(d) significant at 5% level

(e) Significant at 1% level