Blockbuster or bust? Silver screen effect and stock returns

Abstract

This study introduces a novel mood metric—blockbuster movie releases—and investigates its correlation with stock market dynamics. We document a significant positive correlation between blockbuster movie releases and US stock market returns in the subsequent week. This pattern remains robust across various robustness tests both in-sample and out-of-sample. The changes in weekly box office revenue and increased Internet searches for movie-related terms further affirm this relationship. Moreover, releases of blockbuster movies predict lower expected market volatility and risk aversion. The positive predictive effect on market returns is also evident in international markets.

      “The cinema is not a slice of life, but a piece of cake.”

                    —Alfred Hitchcock

1. Introduction

In the behavioral finance literature, mood measures are typically classified into two broad categories. The first category consists of endogenous indicators, where the underlying cause of mood shifts is difficult to discern. For example, sentiment metrics often represents a mix of causes and effects. The second category includes exogenous but unpredictable indicators, such as the outcomes of sporting events, where the winner is unknown in advance, and the resulting market reaction is therefore uncertain. In this study, we introduce a novel mood metric—blockbuster movie release—that is exogenous and demonstrates a strong capacity to predict mood changes.

The broad audience reach of blockbuster movies means that their capacity to influence mood is felt by a large portion of the population, thereby potentially impacting a significant number of investors. With their release schedules and cinema screenings determined well in advance, often months or even years before their debut, blockbuster movie releases act as exogenous shocks that influence mood, offering a clearer causal mechanism that distinguishes their effect from mood-congruent behavior and shifts in equity market sentiment in their release weeks.

The psychological literature suggests that movies can influence mood through primary two mechanisms. First, as a dominant form of media entertainment, movies generally provide enjoyment and escapism, contributing to mood enhancement regardless of their specific content. Second, movies influence mood through their genres and narratives—happy movies tend to uplift viewers, while sad movies may evoke melancholy, although the average movie sentiment is positive. Both mechanisms suggest that the release of blockbuster movies is likely to improve mood on average. In this article, we examine the impact of blockbuster movie releases on mood and the underlying mechanism within the context of a high-stakes environment—the equity market.

Employing US movie data from Box Office Mojo in 21st century before the outbreak of coronavirus disease 2019 (COVID-19), we start with initial validation tests. We find that a single blockbuster release is associated with a 24.2 percent increase in box office revenue during the contemporaneous week, suggesting that blockbuster films engage a broader segment of the population and provide more pronounced entertainment effects on average. Furthermore, blockbuster movie releases lead to significantly elevated levels of public happiness and investor optimism during the trading days of the subsequent week, confirming that blockbuster movie releases serve as an effective mood metric.

Our main analyses investigate the relation between blockbuster movie release and stock market returns. We find that the release of blockbuster movies presents a strong predictive effect on the market return in subsequent week, while it shows no significant correlation with contemporaneous market returns. This predictive effect is both statistically and economically significant. The debut of a single additional blockbuster movie within a week boosts subsequent weekly market returns by fifty basis points (an annualized increase of 26 percent). Remarkably, the positive predictive effects of blockbuster movie releases survive a series of robustness tests, both in-sample and out-of-sample (OOS).

Using the change in box office revenue as an alternative ex-post measure for robustness, we document consistent results. Economically, the positive predictive effect of blockbuster releases on market returns is significantly stronger than that of box office sales, affirming the superior entertainment impact of blockbuster movies. In addition to blockbuster movie releases and box office sales, we consider an indirect measure proxied by the Internet search volume for movie-related terms such as popular theater chains. We consistently find that heightened online searches for movie-related terms positively correlate with stock market returns in the subsequent week.

We then assess whether investors exhibit increased optimism in their risk assessments. Our findings indicate that the release of blockbuster movies correlates with a subsequent decrease in the Chicago Board Options Exchange’s (CBOE) Volatility Index (VIX), and the variance risk premium (VRP), implying that investors expect lower volatility and become less risk-averse after watching blockbuster movies.

We further investigate the primary channel through which blockbuster movies influence investors’ mood and future stock market returns. Tests on movie content positivity and ratings reject the alternative that the specific sentiment of the movie content drives investors’ mood, suggesting that the entertainment effect of blockbuster movies dominates in real-life settings and serves as the primary channel. Building on our initial findings, we broaden our analysis to encompass international markets with well-established movie industries. Consistent with our earlier findings, we document a similar pattern: an increase in box office sales typically precedes a corresponding rise in subsequent stock market returns.

Our study contributes to the literature on how investor mood influences stock market dynamics. Existing studies have identified numerous mood variables, either unpredictably exogenous or endogenous, ranging from the weather and sporting outcomes to aviation disasters and musical positivity (e.g., Kamstra, Kramer, and Levi 2000; Hirshleifer and Shumway 2003; Kamstra, Kramer, and Levi 2003; Edmans, Garcia, and Norli 2007; Kaplanski and Levy 2010; Goetzmann et al. 2015; Chen et al. 2020; Edmans et al. 2022). Our study introduces a novel approach by demonstrating that mood shifts can be predicted ahead of time based on exogenously determined blockbuster movie release schedules. These release dates are known in advance, making the mood effects foreseeable, with economically significant magnitudes. While traditional market efficiency theories assert that predictable effects should be fully arbitraged away, our findings suggest that this is not entirely the case, and we provide a psychological rationale to support our findings. Particularly, our article challenges the misconception that mood variables are insignificant drivers of returns, unpredictable or unhelpful for investments, or incapable of distinguishing cause from effect.

Our study also contributes to psychology research that investigates the effects of movies on mood (e.g., Katz and Foulkes 1962; Oliver 1993; Vorderer and Knobloch 2000; Zillmann 2000; Vorderer, Klimmt, and Ritterfeld 2004; Andrade and Cohen 2007; Bartsch and Viehoff 2010; Oliver and Bartsch 2010; Sjöberg and Engelberg 2010; Oliver and Raney 2011; Bartsch 2012; Niemiec and Wedding 2013; Hanich et al. 2014). Most of these studies are based on laboratory experiments and other settings with relatively small numbers of participants who are shown movies without their own active selection. We test the implication of these psychological studies in real-life settings where viewers actively choose when and which movies to watch in cinemas within the context of a very high-stakes field: the equity market. Our results suggest that the entertainment effect of blockbuster movies dominates.

The remainder of the article is structured as follows: Section 2 discusses the motivation for this study and reviews the relevant literature. Section 3 outlines the data. Section 4 presents the main empirical results, and Section 5 concludes.

2. Motivation and literature review

Exploring the effects of movies on stock returns provides an intriguing and unique perspective on how investor mood impacts financial market dynamics. Importantly, movies align with all three criteria for mood impactors as proposed by Edmans, Garcia, and Norli (2007). First, as a predominant form of media entertainment, movies exert a substantial and unambiguous influence on mood. People actively decide and select movies that are entertaining to them because of the enjoyment they anticipate (Raney and Bryant 2019). The enjoyment is achieved not only from the pleasure, laughter, and excitement through positive genres like comedy, adventure and action movies (hedonic gratifications; Zillmann 2000), but also from the thrill, fear, sadness, and thoughtfulness through negative genres like drama and sad movies, which audiences use for cognitive needs such as seeking life’s meaning, truths, and purposes (eudaimonic gratifications; e.g., Oliver 1993; Knobloch 2003b; Oliver and Raney 2011). Beyond specific genres, movies can also provide gratifications associated with perceiving deeper meaning, feeling moved, and being motivated to elaborate on thoughts and feelings inspired by the experience (appreciation; Oliver and Bartsch 2010). Second, movies’ broad audience reach means that they are capable of affecting the mood of a large portion of the population, potentially influencing a significant number of investors. According to the Motion Picture Association and National Association of Theatre Owners, the US box office grossed approximately $11.4 billion from about 1.24 billion movie tickets sold in 2019—nearly four times the US population of 328 million for that year. Approximately 67 percent of the population attending at least one movie in a theater.¹ Thirdly, the influence of movies is generally consistent across the majority of individuals within a country, as evidenced by the widespread preference for attending movies during their opening weekends.

Using blockbuster movie release provides several advantages. First, blockbuster movies are likely to exhibit greater entertainment effects. Their immense appeal often captures public attention well before they hit the screens. Distributors invest heavily in commercials and trailers, ensuring these films are prominently showcased across various media platforms. This extensive promotion, combined with the inherent allure of blockbuster movies, means they have a heightened capacity to influence public mood and behaviors. Their pervasive presence also makes them more likely to affect a broader segment of the population. Second, blockbuster movie releases are exogenous. The release schedule and cinema screenings for blockbuster films are determined well in advance, often months or even years before their debut. This mitigates concerns that blockbuster releases could be influenced by stock market movements or a demand for mood regulation through cinema in the release week. For example, the release date of Avengers: Endgame—Friday, April 26, 2019—was publicly announced on December 7, 2018. On the day of its release, US daily box office sales surged by 2,989.5 percent, jumping from $5.47 million to $169.2 million. This extraordinary increase is unlikely to be attributed to a sudden, simultaneous emotional shift across a large segment of the population or abrupt changes in market sentiment. Instead, this substantial spike in revenue on release day reflects the anticipated enjoyment and entertainment value of the film, affirming the entertainment effect of blockbuster movies. In light of this, blockbuster releases act as exogenous shocks that influence mood, differentiating their effects from behavior driven by mood-regulation requirements and fluctuations in stock market sentiment during their release week.²

The complex interplay between public mood and financial market movements has been extensively examined. This field has utilized a broad spectrum of mood indicators, each with its own strengths and limitations. Prior research has predominantly focused on the effects of sudden mood shifts triggered by events such as international sports outcomes (Edmans, Garcia, and Norli 2007), aviation disasters (Kaplanski and Levy 2010), terrorist attacks (Chen et al. 2020), and changes to daylight saving time (Kamstra, Kramer, and Levi 2000). However, the highly infrequent nature of these events limits their utility to capturing mood fluctuations throughout the year. More continuous factors like weather conditions—cloud cover (Hirshleifer and Shumway 2003; Goetzmann et al. 2015) and daylight duration (Kamstra, Kramer, and Levi 2003)—have been explored, though they might not precisely reflect the strength of mood’s impact on investors. More recent innovations include using music valence on Spotify as a global mood indicator (Edmans et al. 2022), although such measures remain endogenous to market sentiment.

In this literature, mood measures are generally divided into two main categories. The first category consists of endogenous indicators, such as Spotify song choices or emoji usage (Edmans et al. 2022; Gu, Hong Teoh, and Wu 2023), where the causality of mood is ambiguous, often blending cause and effect. For instance, it is not completely clear whether individuals listen to happy music because they are already in a positive mood or if the music itself induces happiness. The second category includes exogenous but unpredictable indicators, such as the outcomes of sporting events (Edmans, Garcia, and Norli 2007). While the mood of the victor’s supporters may improve and that of the loser’s decline, predicting which team will win and how the market will respond is challenging. This study presents a novel mood metric, showing that mood fluctuations can be anticipated in advance using exogenously predetermined blockbuster movie release schedules.

Another relevant stream of literature involves psychology research investigating the effects of movies on mood. Vorderer, Klimmt, and Ritterfeld (2004) show that movies provide enjoyment, a core component of media entertainment that encompasses physiological, affective, and cognitive dimensions. Such enjoyment can manifest in various ways, including serenity, exhilaration, and laughter through comedy (Zillmann 2000); thrill, fear, and relief through drama (e.g., Vorderer and Knobloch 2000; Knobloch 2003b); and sadness and thoughtfulness through melodrama (e.g., Oliver 1993; Schramm and Wirth 2010; Hanich et al. 2014). Oliver and Raney (2011) further highlight that movies serve as powerful conduits for both hedonic and eudaimonic gratifications, providing pleasures and the satisfaction of social and cognitive needs (Bartsch and Viehoff 2010). Additionally, movies can provide appreciation, a type of gratification associated with perceiving deeper meaning, feeling moved, and being motivated to reflect on thoughts and emotions inspired by the experience, which coexists with hedonic and eudaimonic gratifications and is not confined to specific genres (Oliver and Bartsch 2010). Furthermore, movies could enhance mood through escapism, that offers a mental break by immersing viewers in fictional worlds, helping relieve stress (Katz and Foulkes 1962). Viewers also choose films to regulate their emotions, aligning their choices with their current or desired mood to amplify positive feelings or reduce negative ones (Zillmann 1988; Knobloch 2003a). Moreover, cinema viewing within a cinematic environment—darkened room, large screen, and surround sound—enhances enjoyment through emotional engagement, desire activation, and immersion in a symbolic reality (Flisfeder 2012; Mulvey 2013).

While studies mentioned above generally suggest that movies as media entertainment can improve moods by providing enjoyment and other mood-enhancing impacts, another stream of literature indicates that different genres and content can have varied effects on mood. For instance, Gross and Levenson (1995) found that sadness-inducing scenes can dominate the emotional experience, particularly when resonating with the audience. Andrade and Cohen (2007) observed a significant rise in negative emotions after viewing horror movies, especially among those who avoid fear-inducing content. Sjöberg and Engelberg (2010) noted that horror and disaster movies often induce negative moods, while comedies tend to elicit positive emotions, which can differentially influence viewers’ perceptions of risk. Niemiec and Wedding (2013) highlight that specific films can inspire optimism when viewers identify with the main character. Bartsch (2012) suggests that while sad movies can foster positive outcomes like emotional engagement and contemplation, they may also evoke empathic sadness, negatively affecting mood. These studies suggest another possible channel through which movies could affect mood, in addition to the primary entertainment effect: movie content and sentiment. It is important to note that most psychological studies rely on laboratory experiments or controlled settings, often involving relatively small participant samples who are exposed to preselected movies. In contrast, our study examines the implications of these psychological studies within the real-life context of the equity market, providing evidence that movies enhance mood irrespective of their contents or genres.

3. Data

3.1 Blockbuster movie release

We source our movie box office data from Box Office Mojo, focusing primarily on the US market. We choose a sample period starting from 2000 to ensure data accuracy; prior to this year, comprehensive movie data are sporadically available. To avoid the systemic shocks introduced by the COVID-19 pandemic, which largely impacted global movie markets, our sample concludes at the end of 2019.³ During our sample period, a total of 5,217 movies were released and shown in US cinemas.

Box Office Mojo offers daily box office data for every movie released in the US market, amassing 343,425 movie-day observations over our sample period. Each movie’s daily gross sales and release days are reported. Supplementary Appendix IA.4 illustrates how these daily data are displayed on the Box Office Mojo website. Additionally, more granular details about each movie, such as its genres and distributors, are available (see Supplementary Appendix IA.5). Supplementary Appendix IA.1 lists the highest-grossing movies for each year within our sample period.

From these data, we collect the number of blockbuster movie releases (Release) within a given week, from Monday to Sunday. This yields a time series of 1,043 weekly observations spanning two decades, from 2000 to 2019. Our analysis is on a weekly basis due to the cyclical nature of cinema visits—predominantly over weekends, as delineated in Supplementary Appendix IA.2. To identify blockbuster movies, we adopt a release threshold criterion of over 4,000 theaters on the release dates, a benchmark that delineates 149 titles as blockbuster movies within our dataset spanning January 2000 to December 2019. This equates to an average of roughly seven blockbuster movies per annum—a frequency that aligns with our expectation that blockbuster movies are relatively rare events, designed to garner significant audience attention.⁴ Notably, blockbuster movies account for about 26.1 percent of the total box office sales during our sample period.

As shown in the last column in Supplementary Appendix IA.2, the majority of movies premiere on Fridays, taking advantage of the larger available audience and the appeal of a new release heading into the weekend.

3.2 Box office sales

In addition to blockbuster movie release, we employ changes in weekly box office sales as a validation and an ex-post alternative measure for robustness. We collect daily box office data from Box Office Mojo. Then we compute the weekly box office sales for the entire movie market by summing up the daily gross sales of all movies screened in cinemas for each week, from Monday to Sunday.

To account for the effects of public holidays and other potential seasonal factors on box office performance, we deseasonalize the weekly time series to remove monthly seasonality. Figure 1 displays the deseasonalized historical dollar values of weekly box office sales throughout our sample period. As evident in Figure 1, the time series for total weekly box office sales exhibits an overall upward trend, suggesting non-stationarity. This trend aligns with expectations: as the economy grows and incomes rise, people naturally consume more on services, including entertainment. To obtain a more accurate measure for validation test, we apply the deseasonalized percentage change in weekly box office sales for all movies (ΔMovie), which is stationary.⁵

Figure 1.

This displays the deseasonalized historical dollar values of weekly box office sales over the sample period January 2000 to December 2019.

Figure 2 depicts the weekly per-movie average box office sales by calendar week, starting from the release week, for all movies in Panel A and blockbuster movies in Panel B. Movies typically peak in box office performance during their first calendar week. Despite most movies being released on Fridays, the initial three days (Friday to Sunday) often generate higher sales than subsequent full weeks. This further supports the rationale of using blockbuster movie releases as an effective mood metric.

Figure 2.

This figure presents the weekly per-movie average box office sales (categorized by calendar week) from the release week onwards for all movies in Panel (A) and blockbuster movies in Panel (B) over the sample period January 2000 to December 2019.

3.3 Internet search for movies

In addition to direct movie-related measures like blockbuster movie releases and box office sales, we introduce an indirect measure by focusing on search terms related to movies that capture public attention and interest. We argue that when individuals plan to visit cinemas, they are likely to search movie-related terms such as specific theater chains to find showtimes. Therefore, a higher frequency of Internet searches for movie-related terms may suggest that more individuals will visit cinemas to partake in the entertainment experience provided. In addition to simply utilizing general terms like “movies,” “films,” or “cinemas,” we consider the names of popular theater chains in the USA as our search queries. In our analysis, we focus on forty-six search terms that encompass the leading theater chains in the USA, such as “AMC” and “Regal,” along with general movie-related terms like “movies” and “films.” We collect search volume data for these terms from Google Trends available from January 2004.⁶

Specifically, we first obtain the daily Search Volume Index (SVI) data for each search term, covering the period from January 2004 to December 2019. Next, we convert the daily SVIs into weekly ones by aggregating the SVIs from Monday to Sunday for each week.⁷ Following the methodology outlined by Da, Engelberg, and Gao (2015), we normalize the weekly SVIs for each search term “i” by first calculating the logged difference as:

$$\Delta SV{I}_{i,t}=\mathit{ln}\left(\frac{SV{I}_{i,t}}{SV{I}_{i,t-1}}\right).$$

Then, we employ winsorization by trimming extreme values at the 5 percent level (2.5 percent in each tail) for each of the forty-six logged differences to address concerns regarding outliers. Next, to remove seasonality effects, we regress $\Delta SV{I}_{i,t}$ on month dummies and retain the residuals, $\Delta \mathit{ASV}{I}_{i,t}$⁠. To ensure comparability and address heteroskedasticity, we standardize the residuals by scaling each one by its standard deviation.

Then, to identify the most relevant m search terms for movie’s entertainment effect, we perform backward rolling regressions of $\Delta \mathit{ASV}{I}_{i,t}$ on weekly change in box office sales every 52 weeks. Lastly, we construct the index of Internet search for movies (ISM) as follows:

$${\mathit{ISM}}_t={\sum }_{j=1}^m{\mathit{Rank}}_{j,t}(\Delta \mathit{ASV}{I}_{i,t}),$$

where ${\mathit{Rank}}_{j,t}(\Delta \mathit{ASV}{I}_{i,t})$ is the $\Delta \mathit{ASV}{I}_{i,t}$ for the search terms that had a t-statistic rank of j from the period January 2004 through the most recent year end, where the ranks run from smallest (j = 1) to largest (j = 46). As our main measure for Internet search, we select the ten most positive search terms related to movie box office performance (i.e., m = 10).⁸ Our analysis suggests that the most relevant terms are those leading theater chains, namely “Regal,” “AMC,” and “Cinemark.”

3.4 International markets

For robustness, we expand our analysis to international markets. Specifically, we focus on the largest movie markets outside the USA, depending on their local box office earnings (in USD) as of the end of our sample period. Our emphasis on major movie markets is due to the nature of movies as consumable services that typically require payment for viewing in cinemas. This contrasts with music, which can often be accessed freely. For example, on platforms like Spotify, users can enjoy a vast array of songs at no cost, provided they are amenable to periodic commercials. Thus, in countries with underdeveloped film markets where movies are not as popular, the entertainment effects of films are likely to be negligible.

While Box Office Mojo offers weekly, country-level data for international markets, it comes with a limitation. Particularly, the platform primarily features records for English movies or those whose original or translated titles (if there are any) can be represented within the English alphabet. This poses challenges for countries with established local movie industries that predominantly produce films in their native languages, such as China and South Korea. Therefore, relying solely on English movies from Box Office Mojo in these markets could skew our estimations.

To address this limitation, we turn to another data source, The Numbers, which furnishes more accurate box office data for local movies in these markets.⁹ For illustrative purposes, Figure 3 displays the international box office earnings for the top-seller in each market during the weekend of June 7th, 2019. To highlight the disparity in data accuracy between these two sources, we compare the data from the Chinese box office, the second-largest box office market after the USA, for the weekend of October 6th–8th, 2023, using both The Numbers and Box Office Mojo. This comparison, presented in the Supplementary Appendices IA.7 and IA.8, reveals that Box Office Mojo only includes data for two English movies (i.e., “Oppenheimer” and “A Haunting in Venice”) and misses out on local Chinese films that significantly outperformed these two English releases.

Figure 3.

This figure displays the international box office earnings for the top-seller movies in different markets during the weekend of June 7, 2019.

Source: The numbers.

Taking this into consideration, we set specific criteria for our international sample selection: the movie market must be sizable, ranking among the top ten all around the world by 2019; for countries where English is not the primary language, data must be available on The Numbers, ensuring comprehensive coverage of local movie box office performances. By adhering to these criteria, we identify seven markets: Australia, China, Italy, Mexico, Russia, South Korea, and the UK. Specifically, we source weekly box office data for Australia and the UK from Box Office Mojo, while data for China, Italy, Mexico, Russia, and South Korea is collected from The Numbers. Supplementary Appendix IA.10 presents a summary of key information, including the data source, sample periods for those markets and the three movies with the highest local box office revenues during those periods.

However, for international movie markets, it is worth noting that both Box Office Mojo and The Numbers only predominantly provide weekend box office sales data. These sources do not disclose detailed information regarding the number of theatres operational on the dates of movie releases or accurate debut dates. This limitation in data necessitates a focus on the change in box office performance as the principal metric for our international analysis. Following Edmans, Garcia, and Norli (2007) and Edmans et al. (2022), we incorporate the MSCI world market return in our panel regression analyses.

3.5 Other data

For the validation tests, we use established mood and investor sentiment measures. Specifically, we employ the Hedonometer Happiness Index, which tracks public happiness on a daily basis using online expressions from Twitter. Constructed using the hedonometer algorithm by Dodds et al. (2011), the index ranges from one to nine, with one indicating extremely negative, five neutral, and nine extremely positive.¹⁰ To better capture the impact of blockbuster movie releases on the happiness of investors, we exclude weekends and calculate weekly happiness as the average of daily happiness scores over the trading days of each week. We then deseasonalize this weekly index to account for any seasonal effects and use changes in this deseasonalized weekly happiness (Happiness) as our final measure for the validation tests. In addition, we utilize the well-established investor sentiment measures of Obaid and Pukthuanthong (2022), specifically Photo Pessimism and Text Pessimism, which are based on news photos and articles from The Wall Street Journal. These data, available from September 2008 to December 2019, are collected from Kuntara Pukthuanthong’s website. Similarly, we define weekly Photo Pessimism (Photo_pes) and Text Pessimism (Text_pes) as the average of the daily pessimism scores over the trading days within a given week. Higher values of these pessimism indices indicate lower investor sentiment.

Considering that stock market return predictability may be driven by information associated with economic fundamentals and business cycles, we include six widely studied economic variables in our predictive regressions for comparison: price–earnings ratio, PE, defined as the difference between the log of prices and the log of earnings on the S&P 500 index; dividend–payout ratio, DE, defined as the difference between the log of dividends and the log of earnings on the S&P 500 index; term spread, TMS, calculated as the long-term yield minus the T-bill rate; default yield spread, DFY, defined as the difference between BAA- and AAA-rated bond yields; consumption–wealth ratio, CAY, as defined in Lettau and Ludvigson (2001); cyclically adjusted price-to-earnings ratio, CAPE, developed by Robert Shiller. The data for calculating weekly PE and DE are sourced from Bloomberg, while data for TMS and DFY are obtained from the Federal Reserve Bank of St Louis. CAY and CAPE data are available from Martin Lettau’s website and Shiller’s Data website, respectively.

To ensure the robustness of our findings, we compare our movie measures with three high-frequency predictors. The first two are market sentiment indices: the AAII Sentiment Survey (SENT^AAII), which we obtained from the American Association of Individual Investors’ (AAII) website (https://www.aaii.com/sentimentsurvey/sent_results), and the Financial Advisors’ Sentiment (SENT^advisor), sourced from Investors Intelligence.

Additionally, we consider weather conditions as another comparable weekly predictor, acknowledging the prior literature that suggests cloud cover impacts mood (Hirshleifer and Shumway 2003; Goetzmann et al. 2015). Climatological data, inclusive of hourly weather observations since 2000, were retrieved from the National Oceanic and Atmospheric Administration’s (NOAA) website.¹¹ In line with Edmans et al. (2022), we focus on hourly sky conditions reported by various weather stations across the USA. Each weather station reports cloud cover on a scale from 0 (clear sky) to 8 (overcast sky). Following Goetzmann et al. (2015), we calculate the average daily cloud cover by aggregating hourly values from 6 am to 12 pm. To remove seasonal effects, we adjust the daily cloud cover by subtracting the weekly average as in Hirshleifer and Shumway (2003). The average daily change in deseasonalized cloud cover within a week (ΔDCC) serves as our proxy for weekly weather conditions.

We use the S&P 500 index returns as a representative measure for stock market returns. In line with the methodology of Da, Engelberg, and Gao (2015), we incorporate several control variables including the CBOE VIX, and changes in the Economic Policy Uncertainty (EPU) index as defined by Baker, Bloom, and Davis (2016). Additionally, we employ the Aruoba–Diebold–Scotti (ADS) business conditions index sourced from the Federal Reserve Bank of Philadelphia. Accordingly, the ADS index comprises a range of macroeconomic variables including jobless claims, payroll employment, industrial production, personal income less transfer payments, manufacturing and trade sales, and real gross domestic product (GDP). Following Da, Engelberg, and Gao (2015) and Edmans et al. (2022), we utilize the changes in the ADS index to gauge innovations driven by macroeconomic conditions.¹²

We also include a holiday dummy variable to address concerns that blockbuster releases may coincide with public holidays, and that the predictability could be driven by holidays as mood boosters rather than by the entertainment effects of the movies. The dummy variable equals one if any federal holidays (e.g., Christmas Day, Columbus Day, Independence Day, Labor Day, Martin Luther King Jr. Day, Memorial Day, New Year’s Day, Thanksgiving Day, Veterans Day, or Washington’s Birthday) occur within a given week, and zero otherwise.¹³ To account for potential seasonality in stock market returns, as noted in previous studies (e.g., Heston and Sadka 2008; Keloharju, Linnainmaa, and Nyberg 2016, 2021), we consider month dummies.  Appendix A presents the definition of all the variables used in this study. Table 1 reports the summary statistics and the correlation matrix for the key variables.

Table 1.

Summary statistics.

This table reports the summary statistics and correlation matrix for the key variables in Panels A and B, respectively. Ret represents the weekly stock market returns proxied by S&P 500 index. Release is the number of blockbuster movie releases in a given week. ΔMovie is the recursively deseasonalized percentage change in weekly box office sales. ΔMovie_PCE is the alternative measure of box office change defined as the change in deseasonalized dollar values of box office sales divided by the total personal consumption expenditure from the previous quarter. ISM is the index of Internet search for movies. SENT^AAII is the sentiment index based on the survey of the AAII. SENT^advisor is the sentiment of financial advisors reported by Investor Intelligence. ΔDCC is the weekly change in cloudiness. VIX is the CBOE VIX, EPU is the changes in a news-based measure of EPU index. ADS is the changes in the ADS business conditions index.

Panel A
Variable	N	Mean	Std	Median	p5	p95
Ret(%)	1,043	0.1061	2.3818	0.2029	−3.8978	3.5445
Release	1,043	0.1429	0.3688	0.0000	0.0000	1.0000
ΔMovie	1,043	0.0000	0.4064	−0.0454	−0.3783	0.5475
ΔMovie_PCE (*10−6)	1,043	0.0054	5.5010	−0.3280	−8.0394	9.2791
ISM	782	0.0017	0.9995	0.0173	−1.5746	1.7034
SENTAAII	1,043	7.3696	17.984	6.9400	−21.370	40.200
SENTadvisor	1,043	23.387	14.384	25.200	−4.1000	42.300
ΔDCC	1,043	−0.0005	0.1246	0.0011	−0.2025	0.1980
VIX	1,043	19.298	8.5064	17.100	11.010	34.740
EPU	1,043	0.0295	41.375	−0.9340	−58.026	62.035
ADS	1,043	−0.0006	0.1091	−0.0066	−0.1593	0.1717

Panel B
	ΔMovie	Release	ΔMovie_PCE	ISM	SENTAAII	SENTadvisor	ΔDCC	VIX	EPU
Release	0.1983
ΔMovie_PCE	0.6098	0.2489
ISM	0.8327	0.2800	0.8584
SENTAAII	0.0390	−0.1090	0.0303	0.0479
SENTadvisor	−0.0251	0.1167	0.0139	0.0334	0.4672
ΔDCC	−0.0417	−0.0443	0.0206	−0.0147	−0.0292	−0.0046
VIX	0.0288	−0.1239	0.0049	−0.0132	−0.2564	−0.6733	0.0143
EPU	−0.0066	0.0104	0.0090	−0.0089	−0.0080	−0.0153	0.0455	0.0143
ADS	−0.0206	0.0420	−0.0187	0.0083	0.0010	−0.0044	−0.0034	0.0238	−0.0040

4. Empirical results

4.1 Validation

We begin our empirical analysis by validating blockbuster movie releases as effective mood enhancers, capable of reaching a large audience and promoting increased public emotion and investor optimism. We first examine whether blockbuster movies, due to their extensive promotion and pervasive presence, generate greater entertainment effects by engaging more people. To do this, we analyze the correlation between blockbuster movie releases and contemporaneous deseasonalized weekly changes in box office sales.

The results, presented in Columns (1) and (2) of Table 2, show a significant positive correlation between blockbuster movie releases and contemporaneous box office changes. The coefficient estimates suggest that a single blockbuster release is associated with a 24.2 percent increase in box office revenue during the contemporaneous week. These findings confirm the broader capacity of blockbuster movies to engage and entertain a wider audience.

Table 2.

Validation of blockbuster movie releases.

This table reports the results for validation tests. The dependent variable in Columns (1) and (2) is the contemporaneous change in ΔMovie, defined as the deseasonalized percentage change in weekly box office sales. In Columns (3) and (4), the dependent variable is the one-week-ahead change in deseasonalized weekly trading-day happiness based on Twitter (Happiness). Columns (5) and (6) use the one-week-ahead Photo Pessimism (Photo_pes), derived from news photos in the Wall Street Journal, as in Obaid and Pukthuanthong (2022). Similarly, the dependent variable in Columns (7) and (8) is the one-week-ahead Text Pessimism (Text_pes), also from Obaid and Pukthuanthong (2022). Control variables include lagged dependent variable, a dummy for federal holidays, lagged market returns up to five lags, changes in a news-based measure of EPU, the CBOE VIX, and changes in the ADS business conditions index. The sample period for Columns (1) and (2) spans from January 2000 to December 2019, while the sample period for the remaining columns is from September 2008 to December 2019 due to data availability. Newey–West standard t-statistics are reported. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	ΔMovie(t)	ΔMovie(t)	Happiness(t + 1)	Happiness(t + 1)	Photo_pes(t + 1)	Photo_pes(t + 1)	Text_pes(t + 1)	Text_pes(t + 1)
Release	0.218***	0.242***	0.009***	0.008***	−0.126**	−0.123**	−0.194***	−0.088**
	(6.58)	(6.99)	(3.26)	(2.77)	(−2.56)	(−2.51)	(−2.92)	(−2.24)
Controls	No	Yes	No	Yes	No	Yes	No	Yes
Month dummies	No	Yes	No	Yes	No	Yes	No	Yes
Adj. R2	0.0384	0.0624	0.0221	0.1895	0.0089	0.1035	0.0098	0.7005
N	1,043	1,038	590	589	594	593	594	593

Then, we draw on prior psychology literature suggesting that movies, as a form of media entertainment, enhance mood by providing enjoyment. To test whether blockbuster movie releases influence public mood, we apply the Hedonometer Happiness Index. Specifically, we regress changes in deseasonalized weekly happiness, based on trading days, on the lagged number of blockbuster movie releases. The results, presented in Columns (3) and (4) of Table 2, show that the coefficient estimates are statistically significant and positive, indicating that the public, on average, experiences greater happiness in the week following a blockbuster release. These effects remain robust after controlling for holidays and other variables. Economically, the coefficient estimate of Release in Column (4), 0.008, suggests that a single blockbuster movie release increases happiness by an amount equivalent to 32 percent of the standard deviation of weekly happiness changes (0.025). These findings align with previous psychological studies.

Additionally, we utilize the well-established endogenous investor sentiment measures—Photo Pessimism and Text Pessimism, based on news photos and articles from The Wall Street Journal—from Obaid and Pukthuanthong (2022) to test whether investors entertained by blockbuster movies exhibit greater optimism. We regress these two sentiment measures on lagged blockbuster movie releases and report the results in Columns (5)–(8). The findings consistently show that blockbuster movie releases lead to significantly lower photo and text pessimism. The coefficient estimates in Columns (6) and (8) suggest that each additional blockbuster movie release decreases the Wall Street Journal’s photo and text pessimism by amounts equivalent to 22.6 percent and 10.9 percent of their respective standard deviations. These results align with Hirshleifer and Shumway (2003), who suggest that individuals in a positive mood tend to display greater optimism. Overall, these validation results confirm that blockbuster movie releases serve as an effective mood metric.

4.2 Blockbuster movie and US stock market returns

In our main analysis, we investigate the relation between movie and stock market returns. We estimate the following baseline regression:

$${\mathit{Ret}}_{t+k}=\alpha +\beta {\mathit{moive}}_t+\sum _n{\gamma }_n{\mathit{Control}}_{n,t}+{\epsilon }_{t+k},$$

(1)

where Ret_t + k denotes the weekly stock market return in week t + k, represented by the S&P 500 index. The variable of interest is ${\mathit{moive}}_t$⁠, which could be either the number of blockbuster movie release (Release) or secondary measures such as change in box office sales (ΔMovie) for robustness. Our control variables comprise lagged market returns up to five lags, changes in a news-driven measure of EPU, the CBOE VIX, changes in the ADS business conditions index and a holiday dummy, which equals one if any federal holidays occur within a given week, and zero otherwise. Month dummies are controlled to account for potential seasonality in stock market returns. For all our findings, we calculate Newey–West standard errors and report the corresponding t-statistics.

Table 3 presents the results of our tests for Equation (1), with Release serving as the principal independent variable. Our primary interest lies in examining the correlation between Release and subsequent weekly stock market returns. As reported in Columns (4), (5), and (6), we observe positive and statistically significant coefficients for Release, both with and without the holiday dummy, month dummies, and other control variables. The economic significance of Release is also noteworthy. Based on our coefficient estimates from Column (6), the debut of just one additional blockbuster movie within a week can boost weekly market returns by around fifty basis points (an annualized rate of 26 percent). The baseline univariate regressions exhibit relatively low R² values (0.32 percent). Given that the predictive regressions span a period of two decades with high-frequency data and that weekly returns inherently possess a substantial unpredictable component, relatively modest R² statistics are anticipated. Nevertheless, prior research indicates that an R² of approximately 0.5 percent constitutes a significant degree of monthly return predictability (e.g., Campbell and Thompson 2008; Rapach, Ringgenberg, and Zhou 2016). Given that our analysis employs a higher-frequency weekly dataset, the R² values presented in Table 3 are sufficiently robust to affirm an economically meaningful degree of predictability in returns.¹⁴

Table 3.

Baseline regressions.

This table illustrates the estimation results for the following regression:

$${\mathit{Ret}}_{t+k}=\alpha +\beta {\mathit{Release}}_t+\sum _n{\gamma }_n{\mathit{Control}}_{n,t}+{\epsilon }_{t+k},$$

where the dependent variable represents the weekly stock market returns at time t + k (k = 0,1). The variable of interest, Release, is the number of blockbuster movies released in a given week. Control variables include lagged returns (up to five lags), changes in a news-based measure of EPU, the CBOE VIX, changes in the ADS business conditions index and a holiday dummy, which equals one if there is a federal holiday in a given week. Month dummies are included to control for potential seasonality in stock returns. Newey–West standard t-statistics are reported. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

	(1)	(2)	(3)	(4)	(5)	(6)
	Ret(t)	Ret(t)	Ret(t)	Ret(t + 1)	Ret(t + 1)	Ret(t + 1)
Release	−0.032	−0.253	−0.233	0.416***	0.451***	0.500***
	(−0.17)	(−1.52)	(−1.32)	(2.75)	(2.71)	(2.98)
VIX		−0.099***	−0.101***		0.006	0.004
		(−5.54)	(−5.64)		(0.42)	(0.26)
EPU		−0.001	−0.001		−0.003	−0.003
		(−0.29)	(−0.26)		(−0.91)	(−0.87)
ADS		−0.014	−0.193		−0.889	−1.073
		(−0.01)	(−0.16)		(−0.63)	(−0.75)
Holiday		0.214	0.172		−0.103	−0.111
		(1.24)	(0.97)		(−0.57)	(−0.58)
Ret(t)					−0.069	−0.078
					(−1.40)	(−1.56)
Ret(t−1)		−0.155***	−0.161***		0.040	0.036
		(−3.21)	(−3.34)		(0.90)	(0.79)
Ret(t−2)		−0.049	−0.052		−0.074*	−0.075*
		(−1.10)	(−1.18)		(−1.77)	(−1.81)
Ret(t−3)		−0.145***	−0.151***		−0.043	−0.046
		(−3.03)	(−3.17)		(−0.90)	(−0.97)
Ret(t−4)		−0.113**	−0.118**		0.031	0.028
		(−2.21)	(−2.34)		(0.67)	(0.60)
Ret(t−5)		−0.045	−0.046		0.053	0.052
		(−1.01)	(−1.03)		(1.05)	(1.01)
Constant	0.111	2.059***	1.959***	0.047	−0.049	−0.089
	(1.42)	(6.49)	(4.85)	(0.60)	(−0.17)	(−0.24)
Month dummies	No	No	Yes	No	No	Yes
Adj. R2	−0.0009	0.1081	0.1075	0.0032	0.0168	0.0149
N	1043	1038	1038	1042	1037	1037

In Supplementary Appendix IA.9, we report the estimate results for ex-post measures based on changes in box office sales. The results are consistent. Economically, a one-standard-deviation increase in deseasonalized changes in box office sales (ΔMovie) is associated with a rise of twenty-six basis points in the following week’s stock market returns (an annualized return of 13.5 percent). Similarly, alternative box office change measures, adjusted for economic conditions (ΔMovie_PCE) and consumer confidence (ΔMovie^⊥), yield consistent findings, showing positive coefficients with statistically significant t-statistics.

On the other hand, our results do not indicate any significant correlation between blockbuster movie releases and contemporaneous stock market returns (see Columns (1)–(3)). Such findings imply that blockbuster movie releases, unlike music positivity as studied by Edmans et al. (2022), do not mirror endogenous investor sentiment. Instead, the positive predictive effects appear to stem primarily from the entertainment effect of blockbuster movies, which enhances investors’ mood. This interpretation is further affirmed by the exogenous nature of blockbuster movie releases.¹⁵

4.3 Additional tests

To address the concern that market return predictability may stem from information related to economic fundamentals and business cycles, we compare the predictive power of blockbuster movie releases with that of six standard economic variables, namely the price–earnings ratio (PE), dividend-payout ratio (DE), term spread (TMS), default yield spread (DFY), consumption–wealth ratio (CAY), and the cyclically adjusted price-to-earnings ratio (CAPE). We begin by running univariate predictive regressions with each economic predictor individually and present the results in Panel A of Table 4. During the sample period, only CAY exhibits marginally significant predictive effects on market returns at a 10 percent significance level.¹⁶ Next, we test whether the forecasting power of blockbuster movie releases remains significant after controlling for these economic predictors in bivariate predictive regressions. Panel B of Table 4 shows that the coefficient estimates for Release are all statistically significant and economically substantial with a positive sign, confirming our earlier results from Table 3. The adjusted R² in Panel B Table 4 are also substantially greater than those reported in Panel A based on those typical economic predictors alone. Moreover, when Release is incorporated into the predictive regression, none of the economic variables remain significant. These findings suggest that the return predictability associated with blockbuster movie releases is not driven by information captured by the economic variables.¹⁷

Table 4.

Comparison with economic variables.

This table reports the estimation results for univariate predictive regressions in Panel A, and bivariate predictive regressions that include both Release and a typical economic predictor in Panel B. The dependent variable is the one-week-ahead equity market return proxied by S&P 500 index. Release is the number of blockbuster movies released in a given week. PE is the price-earnings ratio of S&P 500 index; DE is the dividend–payout ratio; TMS is the term spread; DFY is the default yield spread; CAY is the consumption–wealth ratio, as defined in Lettau and Ludvigson (2001); CAPE is Shiller’s cyclically adjusted price-to-earnings ratio. Weekly observations on CAY (CAPE) are defined the most recently available quarterly (monthly) observations. Newey–West standard t-statistics are reported. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

	Panel A: Univariate regressions			Panel B: Bivariate regressions
	${\mathit{Ret}}_{t+1}=\alpha +\phi Z_t^k+{\epsilon }_{t+1}$			${\mathit{Ret}}_{t+1}=\alpha +\beta {\mathit{Release}}_t+\phi Z_t^k+{\epsilon }_{t+1}$
	$\phi$	t-stat	Adj. R2	$\beta$	t-stat	$\phi$	t-stat	Adj. R2
PE	−0.844	(−1.62)	0.0040	0.421***	(2.79)	−0.853	(−1.63)	0.0073
DE	0.010	(0.54)	−0.0007	0.409***	(2.63)	0.006	(0.29)	0.0023
TMS	−0.034	(−0.56)	−0.0007	0.409***	(2.65)	−0.019	(−0.30)	0.0023
DFY	−0.042	(−0.14)	−0.0009	0.415***	(2.73)	−0.031	(−0.10)	0.0023
CAY	−0.092*	(−1.83)	0.0023	0.332**	(1.99)	−0.066	(−1.21)	0.0037
CAPE	−0.024	(−1.40)	0.0020	0.424***	(2.80)	−0.024	(−1.44)	0.0054

In assessing the predictive capacity of blockbuster movie releases on stock market returns, we aim to distinguish the entertainment influence from other prevalent high-frequency predictors. To this end, we compare blockbuster movie releases with investor sentiment indicators and meteorological changes, which have historically been reported to correlate with market fluctuations. Specifically, our analysis incorporates weekly sentiment indices—the AAII (SENT^AAII) and financial advisor sentiment (SENT^advisor)—alongside weekly variations in US cloudiness levels (ΔDCC). Table 5 reports the comparative results. Notably, the influence of blockbuster film releases (Release) on market returns remains statistically significant with the inclusion of those variables, substantiating the robustness of our initial findings. Additionally, an interesting observation emerges from the meteorological data: an increase in cloudiness (ΔDCC) exhibits a positive correlation with stock market returns in the subsequent week, albeit with marginal significance at the 10 percent level. This pattern suggests a potential rebound in market performance following weeks of reduced sunshine.¹⁸ Regarding investor sentiment, both SENT^AAII and SENT^advisor exhibit negative coefficients, consistent with prior research indicating an inverse relationship between sentiment and future market returns. However, these coefficients do not exhibit any statistical significance. Overall, these results reinforce the distinctive predictive power of blockbuster movie releases on stock market returns, independent of market sentiment and weather conditions.

Table 5.

Comparison with other high-frequency predictors.

This table illustrates the estimation results for predictive regressions where other high-frequency predictors are included for comparison. The dependent variable is the stock market return over the subsequent week proxied by S&P 500 index. Release is the number of blockbuster movie releases in a given week. SENT^AAII is the sentiment index based on the survey of the AAII. SENT^advisor is the sentiment of financial advisors reported by Investor Intelligence. ΔDCC is the weekly change in cloudiness. Control variables include lagged returns (up to five lags), changes in a news-based measure of EPU, the CBOE VIX, changes in the ADS business conditions index and a holiday dummy, which equals one if there is a federal holiday in a given week. Month dummies are included to control for potential seasonality in stock returns. Newey–West standard t-statistics are reported. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Release	0.376**	0.459***	0.444***	0.517***	0.431***	0.521***	0.431**	0.503***
	(2.38)	(2.62)	(2.88)	(3.09)	(2.84)	(3.10)	(2.50)	(2.87)
SENTAAII	−0.007	−0.006					−0.007	−0.005
	(−1.46)	(−1.21)					(−1.34)	(−0.88)
SENTadvisor			−0.006	−0.010			0.002	−0.007
			(−0.83)	(−1.22)			(0.24)	(−0.94)
ΔDCC					0.970	1.104*	0.987	1.090*
					(1.60)	(1.88)	(1.63)	(1.85)
Month dummies	No	Yes	No	Yes	No	Yes	No	Yes
Controls	No	Yes	No	Yes	No	Yes	No	Yes
Adj. R2	0.0054	0.0157	0.0036	0.0155	0.0047	0.0172	0.0080	0.0176
N	1,042	1,037	1,042	1,037	1,042	1,037	1,042	1,037

To clarify the economic significance of movies on prospective stock market returns, we explore the predictive power of blockbuster movie releases in relation to tradable index securities. We consider four ETFs for this analysis: the SPDR S&P 500 (NYSEARCA: SPY), the PowerShares QQQ Trust (NASDAQ: QQQ), the Russell 1000 Index ETF (NYSEARCA: IWB), and the Russell 2000 Index ETF (NYSEARCA: IWM). As reported in Table 6, our results indicate significant correlation across all these ETFs. Furthermore, the regression coefficients align closely with those outlined in Table 3.

Table 6.

Tradable market ETFs.

This table illustrates the estimation results for tests based on the returns in the subsequent week for various exchange traded funds including the S&P Exchange Traded Fund (SPY), the Russell 1000 Exchange Traded Fund (IWB), the Russell 2000 Exchange Traded Fund (IWM), and the NASDAQ Exchange Traded Fund (QQQ). Release is the number of blockbuster movie releases in a given week. Control variables include lagged returns up to five lags, changes in a news-based measure of EPU, the CBOE VIX, changes in the ADS business conditions index and a holiday dummy, which equals one if there is a federal holiday in a given week. Month dummies are included to control for potential seasonality in stock returns. Newey–West standard t-statistics are reported. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

	SPY(t + 1)		Russell1000(t + 1)		Russell2000(t + 1)		QQQ(t + 1)
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Release	0.403***	0.488***	0.399***	0.500***	0.402*	0.548**	0.495**	0.618***
	(2.63)	(2.85)	(2.62)	(2.95)	(1.88)	(2.31)	(2.55)	(2.96)
Month dummies	No	Yes	No	Yes	No	Yes	No	Yes
Controls	No	Yes	No	Yes	No	Yes	No	Yes
Adj. R2	0.0029	0.0139	0.0029	0.0132	0.0014	0.0038	0.0019	0.0004
N	1,042	1,037	1,042	1,037	1,042	1,037	1,042	1,037

4.4 OOS forecast

To address potential overfitting and finite sample biases, as well as concerns that most in-sample time series predictors fail to forecast market returns OOS (Welch and Goyal 2008), we examine the OOS predictive power of blockbuster movie releases as well as box office revenue changes. In doing so, we employ an expanding window approach for coefficient estimation, which capitalizes on the maximum available data. For robustness, we also test with rolling in-sample training windows. To guarantee a sufficient number of weekly observations for our initial in-sample training—and to ensure an adequate OOS period for forecasting evaluation—we initiate our coefficient estimation with a baseline of 520 weeks (10 years) of data for the expanding window approach. Similarly, for the rolling training windows, the in-sample training duration is maintained at 520 weeks.

Following Welch and Goyal (2008), our OOS forecasts at time t are predicated solely on information available up to that point. To be specific, for blockbuster movie releases (Release), given its exogenous nature and the fact that it is not derived from deseasonalized data, we retain it in its original form. For changes in box office sales, when forecasting the return for week t + 1 at time t, we eliminate the seasonality in the original weekly box office using data available up to t and then recursively recalculate ΔMovie. Subsequently, we compute the OOS R² as:

$$R_{\mathit{OOS}}^2=1-\frac{{\sum }_{t=p}^{T-1}{\left(r_{t+1}-{\widehat{r}}_{t+1}\right)}^2}{{\sum }_{t=p}^{T-1}{\left(r_{t+1}-{\overline{r}}_{t+1}\right)}^2},\mathrm{ }$$

(2)

where${\widehat{r}}_{t+1}$ is the forecasted one-week-ahead stock market return, which is estimated using information up to time t. ${\overline{r}}_{t+1}$ symbolizes the historical mean of one-week-ahead stock market return up to time t, and T represents the sample size. If a variable proves to be a robust predictor, we expect to secure an OOS R² that is statistically greater than zero. To further validate the statistical significance of the OOS R², we apply the mean squared forecast error (MSFE)-adjusted statistic of Clark and West (2007).¹⁹

Table 7 presents the results for the OOS forecasts using expanding and rolling training windows. Notably, both Release and ΔMovie exhibit positive OOS predictive power, with OOS R² values ranging from 0.61 percent to 0.94 percent, comparable to their in-sample R²s. This reinforces the strength of Release as a reliable predictor of subsequent stock market returns. Furthermore, the MSFE-adjusted t-statistics indicate that the OOS weekly predictive effects of both Release and ΔMovie are statistically significant at the 5 percent level or higher. Additionally, we report the results of OOS forecasts for the comparable economic, sentiment, and weather variables, namely PE, DE, TMS, DFY, CAY, CAPE, SENT^AAII, SENT^advisor, and ΔDCC. The OOS R² of these variables are substantially lower than those of movie variables. Consistent with Welch and Goyal (2008), none of the individual economic variables exhibit significant OOS predictive power. While ΔDCC presents a significant OOS R² at a 10 percent significance level within a rolling in-sample training window, its OOS predictive capacity is not statistically significant within an expanding training window. None of the remaining comparable sentiment predictors exhibit significant OOS predictive effects. Collectively, these OOS forecast results align with the findings from our baseline regressions. To the best of our knowledge, this is the first study to demonstrate a mood metric with significant OOS forecasting ability.

Table 7.

OOS forecasts.

This table presents the OOS test results for forecasting the stock market return over the next week. OOS R square and the MSFE-adjusted statistic of Clark and West (2007) are reported for univariate predictive regressions. Release is the number of blockbuster movie releases in a given week. ΔMovie is the recursively deseasonalized percentage change in weekly box office sales. PE is the price–earnings ratio of S&P 500 index; DE is the dividend–payout ratio; TMS is the term spread; DFY is the default yield spread; CAY is the consumption–wealth ratio, as defined in Lettau and Ludvigson (2001); CAPE is Shiller’s cyclically adjusted price-to-earnings ratio. SENT^AAII is the sentiment index based on the survey of the AAII. SENT^advisor is the sentiment of financial advisors reported by Investor Intelligence. ΔDCC is the weekly change in cloudiness. The initial training period consists of 520 weeks for the extending-window estimates. For the rolling training windows, the in-sample training duration is maintained at 520 weeks.

Training window	Extending		Rolling
	OOS R2 (%)	MSFE t-stat	OOS R2(%)	MSFE t-stat
Release	0.94	(2.858)	0.61	(2.434)
ΔMovie	0.88	(2.138)	0.93	(2.184)
PE	0.56	(1.389)	0.11	(0.866)
DE	−0.27	(−0.982)	−0.24	(−1.082)
TMS	0.00	(0.150)	−0.20	(−0.008)
DFY	−0.02	(−0.603)	0.05	(0.687)
CAY	−0.31	(1.397)	−0.62	(0.563)
CAPE	0.27	(1.041)	−0.10	(0.516)
SENTAAII	0.41	(1.313)	0.17	(0.981)
SENTadvisor	0.08	(0.687)	0.32	(1.476)
ΔDCC	0.29	(1.479)	0.37	(1.847)

4.5 Movie attention

In this subsection, we explore the correlation between Internet search trends for movies and subsequent stock market returns. Table 8 presents the findings where Internet search for movies (ISM) is the variable of interests. Analysis from both univariate (Column (1)) and multivariate predictive regressions (Columns (2)–(10)) indicates that ISM positively correlates with market returns in the following week at a 5 percent significance level. On an economical scale, the coefficient estimation from Column (10) suggests that a one-standard-deviation increase in ISM is associated with an increase of 17.13 basis points in the subsequent weekly stock market returns, corresponding to an annualized return of 8.91 percent. The predictive effect ISM remains significant when other three comparable predictors (i.e., SENT^AAII, SENT^advisor, and ΔDCC) are included as controls. Overall, these results further affirm that movies significantly influence public mood, and consequently impacts equity market returns.

Table 8.

Internet search for movies.

This table illustrates the estimation results for tests with Internet search for movies. The dependent variable is the stock market return over the subsequent week proxied by S&P 500 index. ISM is the index of Internet search for movies. ISM has been available since 2005, due to the data availability of Google Trends and its construction process. SENT^AAII is the sentiment index based on the survey of the AAII. SENT^advisor is the sentiment of financial advisors reported by Investor Intelligence. ΔDCC is the weekly change in cloudiness. Control variables include lagged returns up to five lags, changes in a news-based measure of EPU, the CBOE VIX, changes in the ADS business conditions index and a holiday dummy, which equals one if there is a federal holiday in a given week. Month dummies are included to control for potential seasonality in stock returns. Newey–West standard t-statistics are reported. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
ISM	0.179**	0.216**	0.183**	0.217**	0.182**	0.218**	0.182**	0.218**	0.185**	0.221**
	(2.21)	(2.43)	(2.23)	(2.44)	(2.23)	(2.45)	(2.25)	(2.46)	(2.27)	(2.48)
SENTAAII			−0.004	−0.006					−0.003	−0.004
			(−0.59)	(−0.83)					(−0.39)	(−0.61)
SENTadvisor					−0.004	−0.008			−0.003	−0.007
					(−0.50)	(−0.90)			(−0.34)	(−0.77)
ΔDCC							1.548**	1.576**	1.545**	1.582**
							(2.31)	(2.39)	(2.29)	(2.39)
Month dummies	No	Yes	No	Yes	No	Yes	No	Yes	No	Yes
Controls	No	Yes	No	Yes	No	Yes	No	Yes	No	Yes
Adj. R2	0.0048	0.0299	0.0043	0.0296	0.0042	0.0298	0.0091	0.0344	0.0075	0.0334
N	782	782	782	782	782	782	782	782	782	782

4.6 Expected volatility and risk aversion

Psychological studies have shown that entertainment such as blockbuster movies can foster increased optimism in individuals’ risk evaluations. Apart from stock market returns, our investigation further delves into the potential influence of blockbuster movie releases on investors’ expected volatility and risk aversion in the subsequent week. Specifically, we examine the correlation between major movies releases and two metrics: the VIX and VRP.

Consistent with the approach of Bollerslev, Tauchen, and Zhou (2009), we utilize the five-minute S&P 500 index time series sourced from TICKDATA to compute the daily realized variance. For the computation of the conditional expectation of one-month variance, we employ the daily logarithmic HAR-RV-VIX model as proposed by Bekaert and Hoerova (2014) and Hu, Jacobs, and Seo (2022). The VRP is then calculated as the difference between the risk-neutral expectation and the physical (real-world) expectation of future variance.

The results in Table 9 reveal that blockbuster movie releases (Release) negatively predict the VIX (in Columns (1)–(5)) and VRP (in Columns (6)–(10)) in the subsequent week. All coefficients of Release are significant at the 5 percent level or higher, both with and without control variables and other predictors. These findings suggest that, in the week following the release of blockbuster movies, investors anticipate lower market uncertainty and exhibit diminished risk aversion. These results further affirm the economic distinction between movies and high-frequency investor sentiments, as the latter exert only a transitory impact without predictive effects on expected stock market volatility (Da, Engelberg, and Gao 2015). Overall, these outcomes align with insights from psychological literature.

Table 9.

Expected volatility and risk aversion.

This table presents the estimation results for the relationship between blockbuster movie releases and expected volatility in the subsequent week, proxied by the VIX in Columns (1)–(5), and risk aversion in the subsequent week, proxied by the VRP in Columns (6)–(10). Release is the number of blockbuster movie releases in a given week. SENT^AAII is the sentiment index based on the survey of the AAII. SENT^advisor is the sentiment of financial advisors reported by Investor Intelligence. ΔDCC is the weekly change in cloudiness. Control variables include lagged dependent variables up to five lags, lagged market returns up to five lags, changes in a news-based measure of EPU, changes in the ADS business conditions index and a holiday dummy, which equals one if there is a federal holiday in a given week. Month dummies are also included to control for potential seasonality. Newey–West standard t-statistics are reported. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
	VIX(t + 1)	VIX(t + 1)	VIX(t + 1)	VIX(t + 1)	VIX(t + 1)	VRP(t + 1)	VRP(t + 1)	VRP(t + 1)	VRP(t + 1)	VRP(t + 1)
Release	−0.467**	−0.465**	−0.472**	−0.501**	−0.514**	−0.086***	−0.101***	−0.074**	−0.090***	−0.088**
	(−2.16)	(−2.03)	(−2.21)	(−2.30)	(−2.27)	(−2.58)	(−2.63)	(−2.29)	(−2.66)	(−2.50)
SENTAAII		0.000			−0.001		−0.002			−0.001
		(0.04)			(−0.17)		(−1.32)			(−1.02)
SENTadvisor			0.004		0.005			−0.005		−0.004
			(0.31)		(0.38)			(−1.45)		(−1.37)
ΔDCC				−1.756**	−1.759**				−0.178	−0.185
				(−2.20)	(−2.19)				(−1.16)	(−1.19)
Month dummies	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Controls	Yes	Yees	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Adj. R2	0.8781	0.8779	0.8780	0.8786	0.8784	0.8155	0.8158	0.8164	0.8155	0.8165
N	1,037	1,037	1,037	1,037	1,037	1,037	1,037	1,037	1,037	1,037

4.7 Movie sentiment and ratings

A number of studies in the movie literature argue that, apart from entertainment effects, movies of different genres and content can impact mood in various ways (e.g., Gross and Levenson 1995; Andrade and Cohen 2007; Sjöberg and Engelberg 2010; Bartsch 2012). Although most of these studies are based on laboratory experiments in which participants passively watch films without actively selecting them—thereby significantly reducing the entertainment value compared to real-life settings—this may still represent another channel through which movies influence mood. In other words, positive-genre movies can improve mood, while negative-genre films can induce sadness. Thus, the overall positive predictive effects on the stock market may be driven by the fact that the average sentiment of movies, particularly blockbusters, is positive. To better understand the economic mechanism behind return predictability and distinguish between the entertainment and sentiment effects of movies, we consider proxies for movie positivity and examine their impact on stock market returns.

We consider three measures for movie sentiment. The first measure is based on each movie’s introduction. Specifically, we collect movie introduction from the Box Office Mojo and apply natural language processing (NLP) analysis to evaluate their content sentiment and positivity. Using the widely recognized National Research Council (NRC) Emotion Lexicon, we retrieved sentiment categories for each movie. The NRC lexicon categorizes words into ten distinct emotional and sentiment categories, encompassing negative, anger, fear, disgust, sadness, surprise, anticipation, trust, joy, and positive. Each movie can have multiple sentiment categories based on this lexicon. To quantify movie sentiment, scores from zero (most negative) to nine (most positive) are assigned to these sentiment categories. A movie’s positivity is then estimated as the weighted sum of these scores, with weights determined by the frequency of each category in the movie’s introduction. Next, we calculate the overall weekly movie positivity as the box-office-weighted average of each movie’s positivity within a given week. Then, we apply the changes in weekly movie positivity (ΔPositivity^intro) and examine how it correlates with stock market return in subsequent week.

In addition to using positivity based on movie introductions, we consider a more straightforward measure derived from movie genres. Specifically, we assign a positivity score of one to action, comedy, adventure, fantasy, romance, and sci-fi films, which are known to enhance mood according to movie literature, and a score of zero to thrillers, horror, crime, documentaries, dramas, mysteries, and war films. Since each movie can have multiple genres, we define a movie’s genre positivity as the average score of its top three genres. We then calculate the overall weekly genre positivity as the box office-weighted average of individual movie scores in a given week, testing whether changes in weekly genre positivity (ΔPositivity^genre) predict stock market returns.

As a robustness test, we also consider an alternative positivity scale suggested by ChatGPT. This scale assigns scores between zero and ten based on the emotional impact of different genres, with higher scores indicating greater positivity. Specifically, ChatGPT assigns a positivity score of ten to comedies; nine to romance movies; eight to adventure and action movies; seven to fantasy, animation, family, and musicals; six to drama and mystery films; five to thrillers; four to horror, crime, and documentaries; three to war movies; two to tragedies, and zero to sad dramas. Consistently, we define each movie’s genre positivity as the average of its top three genres, calculate the weekly genre positivity weighted by each movie’s box office sales for a given week, and test whether changes in ChatGPT-based weekly genre positivity (ΔPositivity^GPT) present predictive power.

The results in Columns (1)–(9) of Table 10 show that none of those positivity measures have significant predictive effects on subsequent stock market returns. This suggests that the effect of movies sentiment does not directly influence investors’ mood or investment decisions. When both blockbuster movie releases (Release) and a movie positivity measure are included in the predictive regressions, the predictive effect of Release on aggregate stock market returns remains robust. These findings suggest that the entertainment effect dominates the content effect in real life and is the primary channel through which blockbuster movies influence investors’ mood and future stock market returns.

Table 10.

Move positivity and ratings.

This table reports the estimation results for predictive regressions where movie positivity and ratings are included. The dependent variable is the stock market return over the subsequent week proxied by S&P 500 index. Release is the number of blockbuster movie releases in a given week. ΔPositivity^intro measures change in weekly weighted average movie positivity, using NLP analysis based on movie introduction. ΔPositivity^genre measures the weekly change in weighted average movie positivity based on movie genre. ΔPositivity^GPT is an alternative measure of movie genre positivity using the scale suggested by ChatGPT. ΔRating measures the weekly change in weighted average movie IMDb ratings. Control variables include lagged returns (up to five lags), changes in a news-based measure of EPU, the CBOE VIX, changes in the ADS business conditions index and a holiday dummy, which equals one if there is a federal holiday in a given week. Month dummies are included to control for potential seasonality in stock returns. Newey–West standard t-statistics are reported. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)
Release		0.416***	0.503***		0.472***	0.554***		0.425***	0.506***		0.422***	0.511***
		(2.75)	(2.99)		(3.03)	(3.19)		(2.81)	(3.02)		(2.70)	(2.97)
ΔPositivityintro	0.317	0.318	0.256
	(1.21)	(1.23)	(0.96)
ΔPositivitygenre				−0.356	−0.804	−0.832
				(−0.53)	(−1.18)	(−1.16)
ΔPositivityGPT							−0.088	−0.123	−0.134
							(−0.43)	(−0.60)	(−0.60)
ΔRating										0.028	−0.041	−0.103
										(0.09)	(−0.13)	(−0.35)
Month dummies	No	No	Yes	No	No	Yes	No	No	Yes	No	No	Yes
Controls	No	No	Yes	No	No	Yes	No	No	YES	NO	NO	YES
Adj. R2	0.0001	0.0033	0.0147	−0.0007	0.0033	0.0151	−0.0008	0.0026	0.0143	−0.0009	0.0023	0.0141
N	1043	1,043	1,037	1,043	1,043	1,037	1,043	1,043	1,037	1,042	1,042	1,037

Table 11.

International markets.

This table illustrates the estimation results for international markets. The dependent variable is the one-week ahead local stock market returns in each country proxied by country-specific MSCI indices. Panels A and B report the results for country-by-country and joint tests, respectively. ΔMovie is the percentage change in weekly box office sales in each country. Control variables include the MSCI world market return, lagged local stock market returns, the VIX, changes in a news-based measure of EPU, and changes in the ADS business conditions index. In regressions for the pooled sample (Panel B), country and month fixed effects are controlled. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

Panel A: Country-by-country tests
Country	Without controls		With controls
	ΔMovie	t-stat	ΔMovie	t-stat
Australia	0.670***	(2.74)	0.677***	(2.75)
China	0.153**	(2.34)	0.140**	(2.11)
Italy	0.065	(0.36)	0.048	(0.26)
Mexico	0.089	(0.87)	0.072	(0.73)
Russia	0.094	(0.24)	0.016	(0.04)
South Korea	0.095	(0.65)	0.057	(0.44)
UK	0.228*	(1.80)	0.216*	(1.68)
USA	0.523***	(3.40)	0.597***	(3.77)

Panel B: Pooled sample tests
Excluded country	Without controls		With controls
	ΔMovie	t-stat	ΔMovie	t-stat
None	0.161***	(3.15)	0.150***	(2.96)
Australia	0.141***	(2.85)	0.132***	(2.68)
China	0.166***	(2.90)	0.155***	(2.72)
Italy	0.166***	(3.16)	0.154***	(2.95)
Mexico	0.220***	(3.22)	0.210***	(3.08)
Russia	0.160***	(3.11)	0.150***	(2.93)
South Korea	0.166***	(3.13)	0.156***	(2.96)
UK	0.156***	(2.88)	0.146***	(2.70)
USA	0.134**	(2.44)	0.122**	(2.22)

Beyond movie genres and sentiment, Niemiec and Wedding (2013) suggest that the effects of movies on viewers’ optimism vary depending on the film. Specifically, movies that allow viewers to better identify with the main character are more likely to induce optimism. To account for these effects and test the robustness of our findings, we consider an additional factor—the “quality” of movies. It is reasonable to assume that high-quality films are more likely to help viewers identify with the main character’s situation and be more captivating and entertaining. To investigate this, we merge the data from Box Office Mojo with ratings from IMDb, a well-known online database of information related to films. We compute a weekly average IMDb rating, weighted by each movie’s box office sales for that particular week, and designate this metric as Rating. We then test whether changes in this weekly average rating (ΔRating) mirror the effects observed from blockbuster movie debuts. These results are reported in Columns (10)–(12) of Table 10.

As noted, none of the ΔRating coefficients are of statistical significance. These findings might need to be interpreted with caution. One explanation could be group bias, as only a subset of movie watchers provide ratings. Additionally, IMDb shows only the most recent ratings, not a historical series, which may skew estimations, especially since current ratings may not reflect mood during the movie’s debut week, when peak box office revenues occur. If we dismiss the impact of these biases, the results suggest that the quality of entertainment, as a proxy for audience identification with the main character, may be less significant than the act of being entertained and the volume of entertainment consumed in real-life settings. This raises further questions, encouraging exploration in future studies.

4.8 International evidence

In the final phase of our analysis, we broaden our scope to encompass international box office sales. Our focus narrows to the world’s largest movie markets, selected based on their total box office sales and widespread popularity of films. Taking into account both the size of these markets and data availability, we include seven countries in our study: Australia, China, Italy, Mexico, Russia, South Korea, and the UK, alongside the USA.

Panel A of Table 11 presents the results of the relationship between weekly box office sales changes and subsequent week’s stock market returns, country by country. Notably, countries like Australia, China, and the UK display significant coefficients for the changes in local box office sales (ΔMovie), underscoring a compelling link between box office performance and stock market returns. This association still holds after controlling for a range of global and local economic factors in these countries. In contrast, markets including Italy, Russia, Mexico, and South Korea do not manifest a statistically significant link between the two metrics, though the estimated coefficients remain positive for all.

Panel B of Table 11 presents the results based on pooled samples. It is noteworthy that the correlation between box office fluctuations and subsequent stock market returns is significantly positive across all specifications. This relationship appears even more pronounced when incorporating US data and remains robust when accounting for country-specific and monthly fixed effects.

Overall, the results from international markets are consistent with those observed in the USA. These results collectively suggest that movies, as a form of entertainment, can indeed influence investors’ mood and, therefore, their investment decisions in financial markets.

4.9 Discussion on economic magnitude

Our findings suggest that blockbuster movies influence mood, raising an intriguing question: to what extent can we expect the stock market to respond to a $10 movie ticket purchase? The coefficient estimates for changes in U.S. box office sales (Appendix IA.9) indicate that when weekly box office revenues increase by one standard deviation from the sample mean to approximately $257.3 million, the stock market gains around $40 billion in value in the subsequent week. This implies a multiplier effect of approximately 155.

To interpret this multiplier, it might be essential to explore further: if an investor feels entertained and uplifted after watching a $10 blockbuster movie, how much additional investment might they allocate to the stock market? Would their mood enhancement translate into $10, $100, or $1,000 in stock purchases? Furthermore, what proportion of the movie audience would exhibit this behavior to account for such substantial market gains? A related consideration is whether high-impact investors, such as fund managers or affluent individuals, disproportionately drive this relationship. These questions encourage further research into the linkage between movie ticket sales and dollar-amount stock market gains at individual and institutional levels.

Moreover, a multiplier of 155 seems to suggest a substantial return on investment from producing additional blockbuster movies, given their significant impact on box office revenues. However, the observed relationship warrants careful interpretation. While the release of a blockbuster film, such as an Avengers installment, generates widespread excitement and positivity, frequent releases of blockbuster movies may not sustain this effect over time. This notion aligns with the hedonic adaptation theory, which posits that individuals gradually adapt to repeated positive stimuli, thereby diminishing the intensity of subsequent mood enhancements. In this context, another interesting direction for future research could be to explore the interaction between the effects of mood metrics and the frequency of such stimuli, particularly in financial markets.

5. Conclusion

Motivated by psychological evidence that entertainment strongly affects mood, this article explores the relationship between blockbuster movie releases and stock market returns. Our results reveal that releases of blockbuster films are positively correlated with stock market returns in the subsequent week. The absence of a significant correlation between contemporaneous market returns and blockbuster movie releases allows us to reject the notion that movies simply serve as an endogenous indicator of investor sentiment. Instead, we interpret these findings as evidence that blockbuster movies exert a predictive influence through their impact on investor mood.

These patterns are further affirmed by changes in movie box office revenue, Internet search volumes for movie-related terms and are also evident in international markets, suggesting that movie-related variables have a positive predictive capacity for stock returns in countries with well-established movie industries.

Our study contributes to the literature by introducing a novel methodology that demonstrates how mood shifts can be anticipated using exogenously determined blockbuster movie release schedules. Contrary to traditional market efficiency theories, which suggest that such predictable effects should be fully traded away, our findings indicate otherwise, supported by a psychological rationale. Overall, our study challenges the misconception that mood variables are either insignificant in driving returns, inherently unpredictable, or incapable of distinguishing cause from effect.

Our findings align with the literature emphasizing the role of public mood in financial markets. Understanding the impact of the entertainment sector on investor mood opens new avenues for risk evaluation and investment strategy. This encourages a broader examination of the intersections between entertainment, mood, and economic activity.

Footnotes

1

Source: Gitnux Marketdata Report.

2

It is worth mentioning that our study focuses on the “incremental” entertainment effects provided by blockbuster movies. Specifically, when people actively choose to go to cinemas for blockbuster movies over other types of entertainment, they anticipate greater entertainment value, which could further enhance their mood compared to their usual activities.

3

The movie market came to a complete standstill during the lockdowns of 2020, gradually rebounding as restrictions were lifted. Supplementary Appendix IA.3 provides an overview of annual box office sales surrounding the COVID-19 outbreak in 2020. Nevertheless, as a robustness test, we extend our sample to December 2021 and our results remain qualitatively unchanged. The results are available upon request.

4

The rationale for selecting the 4,000-theater threshold is to establish a conservative definition of a blockbuster movie. Lower thresholds of 3,500 and 3,000 theaters would result in inflated counts of 444 and 1,047 blockbuster movies, respectively, significantly diluting the concept of a “blockbuster” by suggesting the release of 22–52 such movies annually. To ensure the robustness of our definition, we conduct sensitivity analyses with alternative thresholds both below and above 4,000 theaters. These analyses corroborated our initial criterion; higher thresholds consistently enhanced the predictive power, statistical and economic significance of blockbuster movie releases, reinforcing their perceived impact on entertaining investors.

5

Some might argue that a measure like box office sale changes merely reflects underlying economic conditions or consumer confidence. To address these concerns, we consider alternative measures for box office change. Specifically, we use ΔMovie_PCE calculated by dividing the deseasonalized dollar value change in box office sales by the total personal consumption expenditure from the previous quarter. Additionally, we introduce another measure ΔMovie^⊥ for robustness. This measure represents the portion of ΔMovie that is independent of consumer confidence trends. We calculate this by orthogonalizing ΔMovie against the consumer confidence index, using the University of Michigan’s monthly Survey of Consumers as a proxy. We collect the data on total personal consumption expenditure and University of Michigan’s monthly Survey of Consumers from https://fred.stlouisfed.org/. Results for these alternative measures are consistent and reported in Supplementary Appendix IA.9.

6

Supplementary Appendix IA.6 presents the list of all 46 terms we considered for Internet search.

7

We calculate the weekly SVI from daily SVI data instead of relying on Google Trends’ weekly SVI, as Google Trends defines a week as Sunday through Saturday, a definition that does not align with our research design.

8

As robustness tests, we also consider alternative rankings such as the top 5, 20, 30, and 46 ranked search terms. We find consistent results with those alternative rankings and the results are available upon request.

9

See https://www.the-numbers.com/international-charts-overview

10

Data are available from September 2008 to December 2019, sourced from hedonometer.org. Further details and foundational studies on the index can be found at hedonometer.org/about.

11

See https://www.ncei.noaa.gov/access/search/data-search/local-climatological-data.

12

Due to the unavailability of high-frequency macroeconomic variables for non-US countries and considering the relatively greater influence of US macroeconomic policy on non-US stock markets compared to local macroeconomic policy (Brusa, Savor, and Wilson 2019), we follow Edmans et al. (2022) and use weekly changes in US EPU and ADS as controls for international markets.

13

For weeks where Monday is a federal holiday and the stock market is closed, the dummy variable for that week is set to zero, while the dummy for the previous week is set to one. This approach allows us to better control for potential holiday effects on the stock market in the subsequent week.

14

It is worth noting that these R² values we reported are also in the range of those found in related research. Specifically, Hirshleifer and Shumway (2003) report an R² of 0.02 percent, when examining the contemporaneous correlation between daily stock market returns and cloudiness over 16 years.

15

While music might intuitively be thought of as another form of entertainment, our findings on movies differ significantly from those of Edmans et al. (2022). Edmans et al. report that changes in music positivity act as endogenous sentiment indicators, positively correlating with contemporaneous equity returns but negatively predicting stock performance the following week, which the authors attribute to noise trading theory (De Long et al. 1990; Da, Engelberg, and Gao 2015). In contrast, our analysis shows that movie variables, such as blockbuster releases and box office sales, do not correlate with contemporaneous market returns. Since movies are mostly consumed on weekends, they act more as mood influencers, similar to sports outcomes, rather than endogenous sentiment indicators. Furthermore, unlike music, movie variables positively predict future stock returns, aligning with psychological literature on mood enhancement through entertainment. This highlights the different ways media can influence investor behavior and market dynamics.

16

In Table 4, the coefficient of CAY is negative, which contrasts with previous studies that report positive predictive effects of CAY (e.g., Lettau and Ludvigson 2001; Bollerslev, Tauchen, and Zhou 2009). Our empirical analysis indicates that this is primarily driven by the sample period. Specifically, using data from 1990 to 2007, as in Bollerslev, Tauchen, and Zhou (2009), we find positive predictive effects of CAY on stock market returns across weekly, monthly, and quarterly frequencies. However, during the 2000–2019 period, we document negative coefficients for CAY on the same frequencies. This shift might be influenced by broader structural changes in the economy and financial markets, which falls outside the scope of our study and invites further investigation.

17

For all additional tests, we also use deseasonalized changes in box office sales (ΔMovie) as an alternative measure and find consistent results. For brevity, we do not report these results, but they are available upon request.

18

In a supplementary test not presented in the tables, we examine the relationship between contemporaneous market returns and changes in cloudiness (ΔDCC). This regression yields a coefficient of −0.67 for ΔDCC, with a t-statistic of −1.13. These results align with the findings of Hirshleifer and Shumway (2003), who observed a negative, albeit statistically insignificant, correlation between cloudiness and concurrent stock market returns in the USA (see columns (3) and (4) of Table III in their study).

19

Accordingly, the MSFE-adjusted statistic tests the null hypothesis that the historical average MSFE does not surpass the predictive regression forecast MSFE against the one-sided alternative hypothesis that the historical average MSFE exceeds the predictive regression forecast MSFE.

Acknowledgments

We thank David Solomon (the editor), an anonymous associate editor, and an anonymous reviewer for their insightful comments, which significantly improved the quality of our article. We are grateful to Jedrzej Bialkowski, Kee Chung, Ivan Diaz-Rainey, David Hirshleifer, Ivan Indriawan, Marta Khomyn, Yuting Liu, Jane Luo, Daniel Marcus Orlovsky, Limin Xu, Alfred Yawson for their helpful feedback. Research support from the Research Compute Cluster (RCC) at the University of Canterbury is gratefully acknowledged.

Supplementary material

Supplementary material is available at Review of Finance online.

Funding

None declared.

Data availability

The data underlying this article are collected from Box Office Mojo by IMDbPro, available at https://www.boxofficemojo.com.

The data for analysis of international market are collected from theNumbers, available at https://www.the-numbers.com/international-charts-overview.

Appendix A: Variable definition

Variable	Definition
Ret	Weekly US stock market returns (in percentage) proxied by S&P 500 index
Release	Number of blockbuster movie releases within a given week
ΔMovie	Deseasonalized percentage change in weekly total box office sales for all movies
ΔMovie_PCE	Deseasonalized dollar value change in box office sales scaled by the total personal consumption expenditure from the previous quarter
ΔMovie⊥	The residual of ΔMovie orthogonalized to the consumer confidence index
ISM	The average of deseasonalized and standardized Google Trends search volume indices for top ranked movie-related terms in the U.S.
SENTAAII	Sentiment index based on the survey of the American Association of Individual Investors
SENTadvisor	Sentiment of financial advisors reported by Investor Intelligence
ΔDCC	The weekly change in cloudiness as reported by all the weather stations across the U.S.
VIX	The CBOE volatility index
VRP	Variance risk premium calculated as the difference between the risk-neutral expectation and the physical expectation of future variance
EPU	Change in the news-based measure of economic policy uncertainty
ADS	Change in the Aruoba-Diebold-Scotti business conditions index
Holiday	A dummy variable set to one if any federal holidays—such as Christmas Day, Columbus Day, Independence Day, Labor Day, Martin Luther King Jr. Day, Memorial Day, New Year’s Day, Thanksgiving Day, Veterans Day, or Washington’s Birthday—occur within a given week, and zero otherwise.
PE	Price–earnings ratio, defined as the difference between the log of prices and the log of earnings on the S&P 500 index
DE	Dividend–payout ratio, defined as the difference between the log of dividends and the log of earnings on the S&P 500 index
TMS	Term spread, calculated as the long-term yield minus the T-bill rate
DFY	Default yield spread, defined as the difference between BAA- and AAA-rated bond yields
CAY	Consumption–wealth ratio, as defined in Lettau and Ludvigson (2001)
CAPE	Cyclically adjusted price-to-earnings ratio, developed by Robert Shiller
ΔPositivityintro	Movie positivity measure defined as the change in weekly weighted average movie positivity based on movie introduction
ΔPositivitygenre	Movie positivity measure defined as the change in weekly weighted average movie positivity based on movie genre
ΔPositivityGPT	An alternative measure of weekly movie genre positivity using the scale suggested by ChatGPT
ΔRating	Movie quality measure defined as the weekly change in weighted average movie IMDb ratings
Happiness	Change in deseasonalized weekly trading-day happiness proxied by Hedonometer Happiness Index based on Twitter
Photo_pes	Photo pessimism derived from news photos in the Wall Street Journal, as in Obaid and Pukthuanthong (2022)
Text_pes	Text pessimism derived from news texts in the Wall Street Journal, as in Obaid and Pukthuanthong (2022)