PreprintPDF Available

Music Sentiment and Stock Returns

Authors:
Preprints and early-stage research may not have been peer reviewed yet.

Abstract

We develop a novel measure of investor sentiment based on the valence of songs that individuals listen to. We show that our measure of music sentiment captures seasonal mood swings. We further document that music sentiment is associated with a systematic pattern of mispricing correction. This relation is stronger for stocks with greater limits to arbitrage. Our findings add to a body of literature aiming to measure investor sentiment and study its impact on stock market returns.
1
Music Sentiment and Stock Returns
Adrian Fernandez-Perez
Alexandre Garel
Ivan Indriawan§
Auckland University of
Technology
Audencia Business School
Auckland University of
Technology
Abstract: We develop a novel measure of investor sentiment based on the valence of songs
that individuals listen to. We show that our measure of music sentiment captures seasonal mood
swings. We further document that music sentiment is associated with a systematic pattern of
mispricing correction. This relation is stronger for stocks with greater limits to arbitrage. Our
findings add to a body of literature aiming to measure investor sentiment and study its impact
on stock market returns.
JEL Classification: G12; G14
Keywords: Investor Sentiment; Music Sentiment; Mispricing
Auckland University of Technology, Private Bag 92006, 1142, Auckland, New Zealand. E-mail:
adrian.fernandez@aut.ac.nz
Corresponding author. Audencia Business School, 8 Route de la Jonelière, 44312 Nantes, France. Email:
agarel@audencia.com
§ Auckland University of Technology, Private Bag 92006, 1142, Auckland, New Zealand. E-mail:
ivan.indriawan@aut.ac.nz
2
1. Introduction
Baker and Wurgler (2007) define investor sentiment as the belief about future cash flows and
investment risks that is not justified by the facts at hand. De Long, Shleifer, Summers, and
Wadlmann (1990) show that investor sentiment leads to more noise trading, greater mispricing,
and excess volatility. Existing literature provides ample empirical support to investor sentiment
affecting stock prices (e.g., Baker and Wurgler, 2006; Tetlock, 2007; Da, Engelberg, and Gao,
2015; Huang, Jiang, Tu, and Zhou, 2015; Jiang, Lee, Martin, and Zhou, 2019).
An important issue in the literature is the measurement of investor sentiment (Baker and
Wurgler, 2007). Mood proxies have emerged as a viable alternative to survey- or market-
variable-based indicators.
1
Mood can be defined as a transient state of feeling (Kaustia and
Rantapuska, 2016). To capture mood swings, prior studies rely on events likely to affect the
mood of individuals such as specific weather conditions, the outcome of major sporting events,
or aviation disasters (e.g., Hirschleifer and Shumway, 2003; Kamstra, Kramer, and Levi 2003;
Edmans, Garcia, and Norli, 2007; Kaplanski and Levy, 2010).
In this paper, we propose an alternative way to capture the mood of individuals. Instead
of looking at events that cause individuals to feel good or bad, we intend to capture more
directly their mood at a given point in time. To do so, we rely on psychology literature which
shows that music choices reveal mood states.
2
Research in the field of musical psychology
documents strong relations between emotions and psychophysical dimensions (e.g., Behrens
and Green, 1993; Gabrielsson and Juslin, 1996). In addition, research in personality and social
1
Market-variable-based sentiment measures are subject to the criticism that it may contain economics forces other
than investor sentiment. Measuring investor sentiment with surveys is subject the potential bias associated with
survey responses and low data frequency (Baker and Wurgler, 2007). An interesting feature of mood proxies is
that they enable testing directional predictions. In that sense, Edmans et al. (2007) find that losses in major games
predict poor returns in the losing country the next day. Kaplanski and Levy (2010) predict that bad mood induced
by aviation disasters lead to negative rates of return in the stock market.
2
A strand of the psychology and marketing literature shows that music can induce emotions in customers (e.g.,
Bruner, 1990). For this reason, we concentrate on music that people choose to listen because it is more likely to
reveal their mood state.
3
psychology shows how individual differences in musical preferences are linked to personality
traits, values, and cognitive style. In particular, Rentfrow and Gosling (2003) show that
preferences for music dimensions (such as upbeat versus conventional) are related to a wide
array of personality dimensions. Follow-up research (e.g., Greenberg et al., 2015; Greenberg et
al., 2016) further supports that musical preferences are linked to personality traits (e.g.,
cheerfulness, depression, anxiety).
While prior literature establishes a link between music choices and people's emotions, a
key empirical challenge is to associate more precisely mood states to quantifiable song
characteristics. In this respect, we use the valence of a song, which quantifies the musical
positiveness it conveys. Songs with high valence sound more positive (happy, cheerful,
euphoric), while songs with low valence sound more negative (sad, depressed, angry).
Algorithms, trained on ratings of positivity by musical experts, can attribute a valence score to
songs, and music platform such as Spotify makes this data publicly available. In our empirical
analysis, we use the valence of the U.S. top-200 songs played on Spotify to derive a daily
indicator of the mood of U.S. individuals over the 2017-2019 period.
We start by testing whether our music-based mood proxy indeed captures mood
manifestations. We build on prior literature to identify seasonal factors likely to affect
individuals’ moods (e.g., Frieder and Subrahmanyam, 2004; Golder and Macy, 2011;
Bialkowski, Etebari, and Wisniewski, 2012; Bergsma and Jiang, 2016). Consistent with music
sentiment being responsive to mood swings of individuals, we document a strong positive
association with U.S. holidays and weekends and a strong negative association with post-
holidays and winter seasons.
Next, we investigate the relation between music sentiment and stock market returns.
Prior literature, both theoretically and empirically, shows that when investor sentiment is highly
positive, prices are temporarily high but revert to the fundamentals after sentiment retreats,
4
reflecting a pattern of price correction (Baker and Wurgler, 2006, 2007). We find that music
sentiment is associated with a strong reversal pattern from the second day up to one week later,
i.e., a one standard deviation increase in our mood proxy corresponds to a decrease of around
18 bps in the S&P 500 returns in the following week. Such return reversal is stronger for stocks
with greater limits to arbitrage, hence providing additional support for sentiment-induced
mispricing. For robustness, we show that, even after controlling for macroeconomic news
announcements, our results remain. Finally, using a placebo test, we show that quantifiable
audio features other than valence are not associated with future stock returns.
Our study contributes to the existing literature in several ways. Unlike prior literature
that uses low-frequency surveys and economic indicators to measure general investor
sentiment, we develop a high-frequency measure that relies on direct manifestations of the
mood of individuals. Our music-based mood proxy allows us to form directional predictions
that are motivated by psychological literature. As a result, our music-based sentiment measure
is well-suited to investigate the effect of investor sentiment on stock markets. We confirm the
pattern uncovered by prior literature that positive market sentiment is associated with future
stock return reversals, especially for stocks with greater limits to arbitrage.
The closest paper to ours is Da et al. (2015), who use daily internet searches to measure
sentiment.
3
Both our paper and theirs provide a direct high-frequency measure of market
sentiment. However, unlike Da et al. (2015), we do not rely on a specific selection of keywords
to capture individuals emotions. Instead, we use the average positivity of songs that individuals
choose to listen to infer their mood. We believe that both approaches complement each other
3
Our paper also relates to other studies investigating high-frequency proxies of sentiment. For instance, Tetlock
(2007) extracts sentiment from the tone of a popular Wall Street Journal column and Obaid and Pukthuanthong
(2019) measure sentiment through a sample of editorial news photos.
5
because music can give voice to emotions that cannot be expressed in words and would not be
captured by a word-based sentiment measure.
We organize the rest of the paper as follows. In Section 2, we discuss the data and the
music sentiment measure. We report the empirical findings in Section 3 and the robustness tests
in Section 4. Section 5 concludes.
2. Data & Sample
We obtain stock market data from Thomson Reuters Tick History (TRTH) maintained by
Refinitif. We collect daily closing prices for the S&P 500 index (ticker: SPX) for our
construction of stock market returns. To measure music sentiment, we rely on Spotify data.
4
Starting from January 1, 2017, Spotify releases daily statistics of the top 200 songs by the total
number of streams. These statistics include the artist’s name, the title of the song, and the total
number of streams. In addition, the firm has an algorithm that classifies a song’s valence, which
measures the musical positiveness conveyed by a song and ranges between 0.0 to 1.0. This
algorithm is trained on ratings of positivity by musical experts and can be linked to any
particular song using the Spotify application programming interface (API).
5
Songs with high
valence sound more positive (e.g., happy, cheerful, euphoric), while songs with low valence
sound more negative (e.g., sad, depressed, angry).
We gather statistics for the U.S. market for the period from January 1, 2017, to
December 31, 2019. We identify around 4,000 unique songs with over 80 billion streams in
total. On average, there are 79 million streams daily, with around 400,000 streams per song.
6
4
Source: https://spotifycharts.com/regional/us/daily/latest
5
While the algorithm itself is proprietary to Spotify, the valence can be obtained from
https://developer.spotify.com/documentation/web-api/reference/tracks/get-audio-features/. Spotify also reports
other audio features such as loudness, danceability, energy, speechiness, acousticness, instrumentalness, liveness
and tempo. We consider these other features in a robustness test discussed in Section 4.
6
The songs with the highest valence in our sample are: September by Earth, Wind & Fire (valence of 0.982), Here
Comes Santa Claus (Right Down Santa Claus Lane) by Gene Autry (0.976) and Little Saint Nick - 1991 Remix by
6
We then construct a stream-weighted average valence (henceforth SWAV) across the top-200
songs for each day. Figure 1 plots the SWAV over the sample period. Table 1 provides
descriptive statistics. The daily SWAV ranges from 0.369 to 0.552, with an average of 0.453.
We observe positive spikes in SWAV during the Christmas periods (December 25 of each year)
as well as during public holidays (Independence Day on July 4). Winter months (December to
February) tend to have a slightly lower valence with an average of 0.449 compared to 0.454 for
the rest of the year.
INSERT FIGURE 1 HERE
Table 1 reports descriptive statistics of the SWAV and the S&P500 Index. We also
compare our measure of music sentiment with other measures documented in the literature such
as the U.S. economic policy uncertainty (EPU) of Baker et al. (2016), the Aruoba, Diebold and
Scotty (2009) business condition index (ADP) and the S&P500 implied volatility index (VIX).
Panel A reports the statistics for the levels. All the sentiment measures are persistent with a
highly positive first autocorrelation coefficient (SWAV, for instance, has an AC of 0.941). The
ADF p-value further shows that, apart from EPU, all series have unit roots. Panel B reports
summary statistics for the first differences. The mean values of the first differences are close to
zero, although there is quite some variation. We observe that ΔSWAV is negative and serially
correlated, as shown by the AC. The ADF p-value further shows that all the first difference
series are stationary. Hence, in our empirical tests, we use the change in SWAV, as well as the
changes in the other sentiment measures.
INSERT TABLE 1 HERE
3. Empirical Analyses
The Beach Boys (0.971). The songs with the lowest valence are: RMP by Trippie Redd (valence of 0.0333), I'll
see you in 40 by Joji (0.0321) and Legion Inoculant by TOOL (0.0262).
7
3.1. Validation of music sentiment measure
Extant studies document that general mood is related to various seasonal factors. A line of
research studies the effect of holidays on stock market returns (e.g., Frieder and Subrahmanyam
2004; Bialkowski, Etebari, and Wisniewski 2012; Bergsma and Jiang, 2016). In particular,
Bergsma and Jiang (2016) document evidence suggesting that positive holiday moods generate
elevated stock prices. Conversely, the first working day after U.S. holidays is often associated
with negative mood due to post-vacation blues (e.g., Korstanje, 2016). The seasonal affective
disorder is most common during winter seasons (also known as winter blues) and linked to
negative mood (e.g., Mersch et al. 1999). Prior literature further shows that mood fluctuates
across days of the week due to lifestyle and socio-cultural factors (see Birru (2018) for a
review). Friday and the weekend, for instance, have a higher mood than Monday through
Thursday (e.g., Golder and Macy, 2011; Stone et al. 2012; Helliwell and Wang, 2014). In fact,
Mondays are associated with adverse health outcomes such as spike suicides, heart attacks, and
myocardial infractions (e.g., Jessen and Jensen, 1999; Collart et al. 2014).
To validate our music sentiment measure as a proxy for mood, we test how our measure
relates to the above seasonal mood patterns. More specifically, we estimate the following
regression:
       
  
where  is the change in stream-weighted average valence on day t.  are
variables to indicate: 1) U.S. holidays, 2) post-holidays (first working day after U.S. holidays),
3) calendar seasons, and 4) days of the week. We also control for the lags of  over the
previous week to remove the serial correlation (see Panel B of Table 1). are the residuals.
8
We estimate Equation (1) using Ordinary Least Squares (OLS). Following Da et al. (2015), we
calculate bootstrapped standard errors but report only the p-values.
7
Table 2 reports the regression estimates. We document a strong positive association with
U.S. holidays (model 1) and weekends (model 4). We also document a significant and negative
relation during post-holidays (model 2) and winter seasons (model 3). These findings remain
statistically significant even after we combine all the seasonal indicators together, as shown in
model 5, suggesting that our measure of music sentiment is responsive to mood swings of
individuals. Another interesting observation is that  is serially correlated across all
lags, except for the seventh. Therefore, we use the cumulative  in our next analysis.
INSERT TABLE 2 HERE
3.2. Music sentiment and stock market returns
We investigate the relation between music sentiment and stock market returns. Prior literature,
both theoretically and empirically, shows that when investor sentiment is highly positive, prices
are temporarily high but later become low, reflecting a pattern of mispricing correction (Baker
and Wurgler, 2006, 2007). We estimate the following model to examine return reversals:
          
where  is the cumulative S&P 500 return from day    to    with   ,
and  is the weekly music sentiment, i.e., the cumulative stream-weighted average
valence from day    to .
8
The variables we use as Controls are the cumulative S&P 500
7
We first estimate Equation (1). The estimated parameters and residuals are stored. The residuals are resampled
and then inserted back into the model which is then re-estimated. We repeat the process 10,000 times to obtain the
bootstrapped standard errors. Finally, we compare the actual standard errors of Equation (1) with its empirical
distribution and compute the p-values.
8
In Table 2, we observe a strong serial correlation across all models, i.e., all the music sentiment lag variables are
negative and highly significant. Hence, we use weekly cumulative changes in our music-based sentiment variable,
analogous to Tetlock (2007) and Obaid and Pukthuanthong (2019), for instance.
9
return from day    to (or weekly market return), weekly changes in EPU, ADS, and VIX
(see, e.g., Da et al., 2015), in addition to the month and day-of-the-week indicator variables.
 are the residuals. We estimate Equation (2) using OLS and report statistical tests using
bootstrapped standard errors.
Table 3 reports the estimation results. We find a strong return reversal pattern from the
second day (the second column) up to one week after (the last column) as shown by the negative
and significant coefficients for . Focusing on the last column, we find that a one-
unit increase in weekly music sentiment leads to a 0.134 decrease in the S&P 500 returns the
following week (alternatively, a one standard deviation increase in the weekly music sentiment
leads to an 18 bps decrease in the following week returns). Even after controlling for other
sentiment measures, the coefficient for music sentiment remains negative and significant,
suggesting that it adds explanatory power beyond what the other sentiment measures provide.
Overall, our findings suggest that our music-based mood proxy captures reasonably well mood
swings and is associated with a systematic pattern of mispricing correction, where a more
positive (negative) mood is associated with a negative (positive) expected stock market returns
in the following week.
INSERT TABLE 3 HERE
3.3. Music sentiment and limits to arbitrage
There are several factors that can exacerbate the effect of investor sentiment on asset prices.
One of the most salient ones is limits to arbitrage (Shleifer and Vishny, 1997). Arbitrage capital
moves slowly to take advantage of the irrational beliefs of sentiment investors. In line with
previous literature, we expect the effect of market sentiment to be stronger for stocks with
valuations that are highly subjective and difficult to arbitrage, i.e., small stocks, stocks with
high betas, or illiquid stocks as they are disproportionally held by local investors and are prone
10
to speculative trading of sentiment investors (see, e.g., Edmans et al. 2007; Baker, Bradley and
Wurgler 2011). Conversely, we expect the effect to be weaker for large, low-beta, and liquid
stocks.
To test our predictions, we examine how music sentiment relates to the average returns
of portfolios of stocks with greater limits to arbitrage versus portfolios of stocks with lower
limits to arbitrage. For the size-based portfolios, we use the first (small) and tenth (large)
percentile of the univariate sorted portfolios formed on the market capitalization of Fama and
French (1993) collected from Kenneth French’s website.
9
As proxies for the beta-based
portfolios, we use the S&P 500 High Beta Index (SP5HBI) and the S&P 500 Low Volatility
Index (SP5LVI) collected from TRTH. Finally, to represent the liquidity-based portfolios, we
use the Vanguard U.S. Liquidity Factor ETF (VFLQ), and iShares Edge MSCI USA Quality
Factor ETF (QUAL) collected from Yahoo Finance.
Table 4 reports the results. The average music sentiment coefficients and adjusted-R2
for the portfolios of stocks with greater limits to arbitrage (Panel A) is around -0.27 and 4.00%,
respectively, compared with that for the portfolios of stocks with lower limits to arbitrage (Panel
B) of merely -0.09 and 2.00%. Relative to a model with all the controls but SWAV
(unreported), adding music sentiment increases the adjusted-R2 by 2.77% for the portfolios of
stocks with greater limits to arbitrage, but only by 0.69% for the portfolios of stocks with lower
limits to arbitrage. Our results, therefore, indicate that the effect of music sentiment on stock
market return reversal is much stronger for stocks with greater limits to arbitrage, which
provides additional support for sentiment-induced mispricing.
INSERT TABLE 4 HERE
4. Robustness Tests
9
Source: https://mba.tuck.dartmouth.edu/pages/faculty/ken.french/
11
We conduct several robustness tests to further validate our results. First, we control for days
with macroeconomic news announcements because they can affect stock returns and reversal
patterns (e.g., Da et al., 2015). We collect dates of 24 U.S. Macroeconomic news
announcements from Thomson Reuters and add an indicator variable which equals one if there
is a news release that day and zero otherwise. Appendix A lists all the macroeconomic news we
consider in this study. As reported in Table 5, Column 1, we find that, even after controlling
for the macroeconomic news releases, music sentiment remains negative and statistically
significant associated with stock market return reversals.
Second, we use other Spotify audio features as a placebo test. These features are
loudness, danceability, energy, speechiness, acousticness, instrumentalness, liveness, and
tempo. We extract the first principal component of the changes in the stream-weighted average
of all the above features, i.e., ΔPC1. For consistency, we use the cumulative ΔPC1 over the past
week as our main regressor in Equation (2). The regression results are reported in the second
column of Table 5. The coefficient on cumulative ΔPC1 is not statistically significant, which
suggests that, apart from valence, other audio features do not seem to affect stock market
returns.
5. Conclusion
In this paper, we develop a novel market sentiment measure based on the valence of
songs that individuals listen to. We find that our measure of music sentiment captures
reasonably well mood swings and is associated with a systematic pattern of mispricing
correction. Our findings add to a body of studies investigating the effect of investor sentiment
on financial markets.
12
References
Aruoba, S. B., Diebold, F. X., & Scotti, C. (2009). Real-time measurement of business
conditions. Journal of Business & Economic Statistics, 27, 417-427.
Baker, M., & Wurgler, J. (2006). Investor sentiment and the cross‐section of stock returns.
Journal of Finance, 61, 1645-1680.
Baker, M., & Wurgler, J. (2007). Investor sentiment in the stock market. Journal of Economic
Perspectives, 21, 129-152.
Baker, M., Bradley, B., & Wurgler, J. (2011). Benchmarks as limits to arbitrage: Understanding
the low-volatility anomaly. Financial Analysts Journal, 67, 40-54.
Baker, S. R., Bloom, N., & Davis, S. J. (2016). Measuring economic policy uncertainty.
Quarterly Journal of Economics, 131, 1593-1636.
Behrens, G. A., & Green, S. B. (1993). The ability to identify emotional content of solo
improvisations performed vocally and on three different instruments. Psychology of
Music, 21, 20-33.
Bergsma, K., & Jiang, D. (2016). Cultural New Year holidays and stock returns around the
world. Financial Management, 45, 3-35.
Białkowski, J., Etebari, A., & Wisniewski, T. P. (2012). Fast profits: Investor sentiment and
stock returns during Ramadan. Journal of Banking & Finance, 36, 835-845.
Birru, J. (2018). Day of the week and the cross-section of returns. Journal of Financial
Economics, 130, 182-214.
Bruner, G. C. (1990). Music, mood, and marketing. Journal of Marketing, 54, 94-104.
Collart, P., Coppieters, Y., Godin, I., & Levêque, A. (2014). Day-of-the-week variations in
myocardial infarction onset over a 27-year period: the importance of age and other risk
factors. American Journal of Emergency Medicine, 32, 558-562.
Da, Z., Engelberg, J., & Gao, P. (2015). The sum of all FEARS investor sentiment and asset
prices. Review of Financial Studies, 28, 1-32.
De Long, J. B., Shleifer, A., Summers, L. H., & Waldmann, R. J. (1990). Noise trader risk in
financial markets. Journal of Political Economy, 98, 703-738.
Edmans, A., Garcia, D., & Norli, Ø. (2007). Sports sentiment and stock returns. Journal of
Finance, 62, 1967-1998.
Fama, E., & French, K. (1993). Common risk factors in the returns on stocks and bonds. Journal
of Financial Economics, 33, 356.
Frieder, L., & Subrahmanyam, A. (2004). Nonsecular regularities in returns and volume.
Financial Analysts Journal, 60, 29-34.
Gabrielsson, A., & Juslin, P. N. (1996). Emotional expression in music performance: Between
the performer's intention and the listener's experience. Psychology of Music, 24, 68-91.
Golder, S. A., & Macy, M. W. (2011). Diurnal and seasonal mood vary with work, sleep, and
daylength across diverse cultures. Science, 333, 1878-1881.
Greenberg, D. M., Baron-Cohen, S., Stillwell, D. J., Kosinski, M., & Rentfrow, P. J. (2015).
Musical preferences are linked to cognitive styles. PloS one, 10.
Greenberg, D. M., Kosinski, M., Stillwell, D. J., Monteiro, B. L., Levitin, D. J., & Rentfrow, P.
J. (2016). The song is you: Preferences for musical attribute dimensions reflect
personality. Social Psychological and Personality Science, 7, 597-605.
Helliwell, J. F., & Wang, S. (2014). Weekends and subjective well-being. Social Indicators
Research, 116, 389-407.
Hirshleifer, D., & Shumway, T. (2003). Good day sunshine: Stock returns and the weather.
Journal of Finance, 58, 1009-1032.
Huang, D., Jiang, F., Tu, J., & Zhou, G. (2015). Investor sentiment aligned: A powerful
predictor of stock returns. Review of Financial Studies, 28, 791-837.
13
Jessen, G., & Jensen, B. F. (1999). Postponed suicide death? Suicides around birthdays and
major public holidays. Suicide and Life‐Threatening Behavior, 29, 272-283.
Jiang, F., Lee, J., Martin, X., & Zhou, G. (2019). Manager sentiment and stock returns. Journal
of Financial Economics, 132, 126-149.
Kamstra, M. J., Kramer, L. A., & Levi, M. D. (2003). Winter blues: A SAD stock market cycle.
American Economic Review, 93, 324-343.
Kaplanski, G., & Levy, H. (2010). Sentiment and stock prices: The case of aviation disasters.
Journal of Financial Economics, 95, 174-201.
Kaustia, M., & Rantapuska, E. (2016). Does mood affect trading behavior? Journal of Financial
Markets, 29, 1-26.
Korstanje, M. E. (2016). Divorcing after holidays: From sacredness to post-vacation blues
syndrome. In Global Dynamics in Travel, Tourism, and Hospitality, 70-84. IGI Global.
Mersch, P. P. A., Middendorp, H. M., Bouhuys, A. L., Beersma, D. G., & van den Hoofdakker,
R. H. (1999). Seasonal affective disorder and latitude: a review of the literature. Journal
of Affective Disorders, 53, 35-48.
Obaid, K., & Pukthuanthong, K. (2019). A Picture is Worth a Thousand Words: Measuring
Investor Sentiment by Combining Machine Learning and Photos from News. Available
at SSRN 3297930.
Rentfrow, P. J., & Gosling, S. D. (2003). The do re mi's of everyday life: the structure and
personality correlates of music preferences. Journal of Personality and Social
Psychology, 84, 1236.
Shleifer, A., & Vishny, R. W. (1997). The limits of arbitrage. Journal of Finance, 52, 35-55.
Stone, A. A., Schneider, S., & Harter, J. K. (2012). Day-of-week mood patterns in the United
States: On the existence of ‘Blue Monday’,‘Thank God it's Friday’and weekend effects.
Journal of Positive Psychology, 7, 306-314.
Tetlock, P. C. (2007). Giving content to investor sentiment: The role of media in the stock
market. Journal of Finance, 62, 1139-1168.
14
Appendix A. List of US macroeconomic news releases
This table list all the US macroeconomic news releases considered in our study. The sample
period is from January 1, 2017, to December 31, 2019. * indicates that Personal Income and
Personal Consumption Expenditure have the same release dates. ** indicates that Industrial
Production and Capacity Utilization have the same release dates.
No
Macroeconomic News
Releases
Frequency
1
GDP Advance
12
Quarterly
2
GDP Preliminary
12
Quarterly
3
GDP Final
12
Quarterly
4
Personal Income, Personal Consumption Expenditures*
36
Monthly
5
Trade Balance
36
Monthly
6
Nonfarm Payroll Employment
36
Monthly
7
PPI
36
Monthly
8
CPI
36
Monthly
9
Retail Sales
36
Monthly
10
New Home Sales
36
Monthly
11
Durable goods orders
36
Monthly
12
Factory Orders
36
Monthly
13
Business Inventories
36
Monthly
14
Construction Spending
36
Monthly
15
Housing Starts
37
Monthly
16
Consumer Confidence Index
36
Monthly
17
Chicago PMI
35
Monthly
18
Leading Indicator Index
36
Monthly
19
Industrial Production, Capacity Utilization**
36
Monthly
20
Consumer Credit
36
Monthly
21
Government Budget
36
Monthly
22
Target Federal Funds Rate
24
6-Week
BEA = Bureau of Economic Analysis
BLS = Bureau of Labour Statistics
BC = Bureau of the Census
CB = Conference Board
FRB = Federal Reserve Bank
FMS = Financial Management Service
15
Figure 1. Music sentiment
This figure plots the daily stream-weighted average valence, SWAV, over our sample period
from January 1, 2017, to December 31, 2019. Valence measures the musical positiveness
conveyed by a song and ranges between 0.0 to 1.0. Songs with high valence sound more positive
(happy, cheerful, euphoric), while songs with low valence sound more negative (sad, depressed,
angry).
16
Table 1. Descriptive statistics
This table reports summary statistics of the daily series used in this paper. The sample period
is from January 1, 2017, to December 31, 2019. Statistics for SWAV are based on calendar days,
whereas statistics for the rest of the variables are based on trading days. Panel A reports
statistics for the levels, and Panel B reports statistics for the first differences (log returns in the
case of SPX). AC is the autocorrelation function, and ADF is the p-value for the Augmented
Dickey-Fuller test. *, ** and *** denote significance at the 10%, 5% and 1% level, respectively.
Panel A: Summary statistics for levels
SWAV
EPU
ADS
VIX
SPX
Mean
0.453
93.605
-0.067
14.374
2703.163
Median
0.449
85.740
-0.089
13.155
2723.530
Maximum
0.552
305.980
0.715
37.320
3240.020
Minimum
0.369
10.920
-0.629
9.140
2257.830
Std. Dev.
0.025
43.425
0.268
4.139
227.625
Skewness
0.134
1.363
0.461
1.555
-0.038
Kurtosis
2.864
5.829
3.394
6.506
2.230
AC
0.941***
0.326***
0.997***
0.932***
0.989***
ADF
0.619
0.001***
0.499
0.156
0.969
#Obs.
1095
754
754
754
754
Panel B: Summary statistics for the first difference
ΔSWAV
ΔEPU
ΔADS
ΔVIX
SPX Ret
Mean
-0.00004
0.02112
-0.00071
0.00124
0.00048
Median
0.000
-0.850
-0.001
-0.070
0.001
Maximum
0.046
232.620
0.103
20.010
0.048
Minimum
-0.102
-218.970
-0.074
-7.340
-0.042
Std. Dev.
0.009
50.456
0.017
1.527
0.008
Skewness
-3.388
0.208
0.439
3.331
-0.704
Kurtosis
37.403
5.619
8.181
44.643
8.422
AC
-0.084**
-0.456***
0.663***
-0.114***
-0.036
ADF
0.001***
0.001***
0.001***
0.001***
0.001***
#Obs.
1094
753
753
753
753
17
Table 2. Music sentiment as a mood proxy
This table reports the regression estimates from Equation (1) over the sample period from January 1, 2017, to December 31, 2019 (including
weekends). The dependent variable is the changes in stream-weighted average valence, ΔSWAV. Holidays is a dummy equal to one for U.S. national
holidays and zero otherwise. Post_holidays is a dummy that is equal to one on the first weekday after US national holidays, zero otherwise. Seasonal
and day-of-the-week indicators are included. Bootstrap p-values are reported in parenthesis. *, **, and *** denote significance at the 10%, 5% and
1% level, respectively.
(1)
(2)
(3)
(4)
(5)
Constant
0.000
(0.308)
0.000
(0.157)
0.001
(0.234)
-0.000
(0.812)
0.000
(0.773)
Holidays(t)
0.009***
(0.000)
0.010***
(0.000)
Post_holidays(t)
-0.012***
(0.000)
-0.012***
(0.000)
Winter(t)
-0.001*
(0.064)
-0.001*
(0.062)
Spring(t)
0.000
(0.812)
-0.000
(0.850)
Summer(t)
-0.001
(0.220)
-0.001
(0.234)
Monday(t)
-0.000
(0.748)
-0.001
(0.124)
Tuesday(t)
0.000
(0.726)
0.002*
(0.058)
Wednesday(t)
-0.000
(0.747)
-0.000
(0.924)
Thursday(t)
0.000
(0.939)
0.000
(0.924)
Friday(t)
-0.001
(0.258)
-0.001
(0.423)
Saturday(t)
0.003***
(0.007)
0.0028***
(0.002)
ΔSWAV(t-1)
-0.156***
(0.000)
-0.105***
(0.007)
-0.146***
(0.001)
-0.135***
(0.003)
-0.113***
(0.005)
ΔSWAV(t-2)
-0.220***
(0.000)
-0.192***
(0.000)
-0.215***
(0.000)
-0.210***
(0.000)
-0.197***
(0.000)
ΔSWAV(t-3)
-0.142***
(0.001)
-0.122***
(0.002)
-0.146***
(0.002)
-0.141***
(0.002)
-0.132***
(0.002)
ΔSWAV(t-4)
-0.089**
(0.012)
-0.084**
(0.016)
-0.092**
(0.016)
-0.086**
(0.020)
-0.085**
(0.013)
ΔSWAV(t-5)
-0.115***
(0.006)
-0.124***
(0.003)
-0.129***
(0.002)
-0.123***
(0.002)
-0.116***
(0.004)
ΔSWAV(t-6)
-0.093**
(0.014)
-0.116***
(0.004)
-0.108***
(0.009)
-0.097**
(0.013)
-0.096**
(0.010)
ΔSWAV(t-7)
0.016
(0.564)
-0.001
(0.944)
0.013
(0.642)
-0.002
(0.906)
-0.025
(0.393)
Adj-R²(%)
10.01
11.58
6.74
7.73
16.61
#Obs.
1087
1087
1087
1087
1087
18
Table 3. Music sentiment and market reaction
This table reports the regression estimates from Equation (2) over the sample period from January 1, 2017, to December 31, 2019. The dependent
variable is the log returns for the S&P 500 index at various horizons. The independent variables are the weekly changes in the stream-weighted
average valence, ΔSWAV(t-4:t), the weekly log returns of the dependent variable, Ret(t-4:t), and the weekly changes in EPU, ADS and VIX (ΔEPU(t-
4:t), ΔADS(t-4:t), ΔVIX(t-4:t), respectively). We also include the week and month dummy variables. Bootstrap p-values are reported in parenthesis. *,
**, and *** denote significance at the 10%, 5% and 1% level, respectively.
Ret(t+1)
Ret(t+1:t+2)
Ret(t+1:t+3)
Ret(t+1:t+4)
Ret(t+1:t+5)
Constant
0.000
0.003**
0.001*
0.005***
0.001*
0.006**
0.002*
0.007**
0.002*
0.007
(0.100)
(0.013)
(0.075)
(0.009)
(0.066)
(0.030)
(0.060)
(0.039)
(0.046)
(0.114)
ΔSWAV(t-4:t)
-0.012
-0.019
-0.049
-0.055
-0.081*
-0.083*
-0.107*
-0.113**
-0.134**
-0.144**
(0.546)
(0.368)
(0.164)
(0.125)
(0.083)
(0.091)
(0.066)
(0.048)
(0.047)
(0.034)
Ret(t-4:t)
-0.074**
-0.148***
-0.163**
-0.171*
-0.134
(0.019)
(0.006)
(0.022)
(0.056)
(0.188)
ΔEPU(t-4:t)
0.000
0.000
0.000
0.000
0.000
(0.168)
(0.479)
(0.940)
(0.854)
(0.903)
ΔADS(t-4:t)
-0.007
-0.014
-0.020*
-0.026*
-0.033*
(0.114)
(0.103)
(0.091)
(0.085)
(0.064)
ΔVIX(t-4:t)
0.000
-0.001*
-0.001*
-0.001
0.000
(0.206)
(0.078)
(0.083)
(0.280)
(0.786)
Time dummies
NO
YES
NO
YES
NO
YES
NO
YES
NO
YES
Adj-R² (%)
-0.09
0.10
0.24
1.56
0.57
1.91
0.78
2.15
0.99
2.26
#Obs.
748
748
747
747
746
746
745
745
744
744
19
Table 4. Music sentiment and limits to arbitrage
This table reports the regression estimates from Equation (2) over the sample period from
January 1, 2017, to December 31, 2019 (except for the low liquidity portfolio, which is from
February 16, 2018, to December 31, 2019). The dependent variable is the log returns from day
t+1 to day t+5, i.e. Ret(t+1:t+5). Panel A represents stock portfolios with greater limits to
arbitrage, e.g., small-cap, high beta, and low liquidity. Panel B represents stock portfolios with
lower limits to arbitrage, e.g., large-cap, low volatility, and quality. The independent variables
are the weekly changes in the stream-weighted average valence, ΔSWAV(t-4:t), the weekly log
returns of the dependent variable, Ret(t-4:t), and the weekly changes in EPU, ADS and VIX
(ΔEPU(t-4:t), ΔADS(t-4:t), ΔVIX(t-4:t), respectively). We also include the week and month indicators.
Bootstrap p-values are reported in parenthesis. *, **, and *** denote significance at the 10%,
5% and 1% level, respectively.
Panel A: Greater limits to arbitrage
Small cap Ret(t+1:t+5)
High beta Ret(t+1:t+5)
Low liquidity Ret(t+1:t+5)
Constant
0.002
0.002
0.002
0.007
0.002
0.018**
(0.246)
(0.725)
(0.361)
(0.296)
(0.257)
(0.027)
ΔSWAV(t-4:t)
-0.269***
-0.290***
-0.258**
-0.262***
-0.283***
-0.282***
(0.002)
(0.000)
(0.008)
(0.009)
(0.003)
(0.003)
Ret(t-4:t)
0.111
-0.011
0.036
(0.179)
(0.897)
(0.760)
ΔEPU(t-4:t)
0.000
0.000
0.000
(0.492)
(0.388)
(0.807)
ΔADS(t-4:t)
-0.028
-0.049*
-0.065*
(0.247)
(0.066)
(0.058)
ΔVIX(t-4:t)
0.001
0.000
0.001
(0.147)
(0.642)
(0.246)
Time dummies
NO
YES
NO
YES
NO
YES
Adj-R² (%)
2.40
4.54
1.73
2.92
3.91
9.79
#Obs.
744
744
744
744
462
462
Panel B: Lower limits to arbitrage
Large cap Ret(t+1:t+5)
Low volatility Ret(t+1:t+5)
Quality Ret(t+1:t+5)
Constant
0.003**
0.008*
0.002**
0.006**
0.003**
0.006
(0.015)
(0.064)
(0.015)
(0.031)
(0.022)
(0.204)
ΔSWAV(t-4:t)
-0.136**
-0.148**
0.000
-0.023
-0.122*
-0.135**
(0.038)
(0.035)
(0.978)
(0.616)
(0.078)
(0.049)
Ret(t-4:t)
-0.136
-0.204**
0.019
(0.182)
(0.010)
(0.836)
ΔEPU(t-4:t)
0.000
0.000
0.000
(0.883)
(0.343)
(0.820)
ΔADS(t-4:t)
-0.032*
-0.017
-0.035*
(0.080)
(0.190)
(0.059)
ΔVIX(t-4:t)
0.000
0.000
0.001
(0.823)
(0.415)
(0.231)
Time dummies
NO
YES
NO
YES
NO
YES
Adj-R² (%)
0.99
2.36
-0.13
6.17
0.77
2.29
#Obs.
744
744
744
744
744
744
20
Table 5. Robustness table
This table reports the regression estimates from Equation (2) over the sample period from
January 1, 2017, to December 31, 2019. The dependent variable is the log returns from day t+1
to day t+5, i.e. Ret(t+1:t+5) for the S&P 500 index. The independent variables are the weekly
changes in the stream-weighted average valence, ΔSWAV(t-4:t). ΔPC1 is the first principal
component of the changes in the stream-weighted average of other audio features (such as
loudness, danceability, energy, speechiness, acousticness, instrumentalness, liveness, and
tempo). Ret(t-4:t) is the weekly log returns of the dependent variable, ΔEPU(t-4:t), ΔADS(t-4:t), and
ΔVIX(t-4:t) is the weekly changes in EPU, ADS, and VIX, respectively. We also include the week
and month indicators, as well as indicator variables for US macroeconomic news
announcements. Bootstrap p-values are reported in parenthesis. *, **, and *** denote
significance at the 10%, 5% and 1% level, respectively.
S&P 500 Ret(t+1:t+5)
Constant
0.007*
0.006
(0.098)
(0.133)
ΔSWAV(t-4:t)
-0.146**
(0.032)
ΔPC1(t-4:t)
0.000
(0.839)
Ret(t-4:t)
-0.132
-0.130
(0.205)
(0.205)
ΔEPU(t-4:t)
0.000
0.000
(0.833)
(0.841)
ΔADS(t-4:t)
-0.033*
-0.033*
(0.065)
(0.071)
ΔVIX(t-4:t)
0.000
0.000
(0.809)
(0.675)
Time dummies
YES
YES
Macroeconomic news dummies
YES
NO
Adj-R² (%)
2.44
0.99
#Obs.
744
744
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
Research suggests that musical preferences are linked to personality, but this research has been hindered by genre-based theories and methods. We address this limitation using a novel method based on the actual attributes that people perceive from music. In Study 1, using 102 musical pieces representing 26 genres and subgenres, we show that 38 perceived attributes in music can be organized into three basic dimensions: arousal, valence, and depth. In Study 2 (N = 9,454), we show that people’s preferences for these musical attributes reflected their self-ratings of personality traits. Importantly, personality was found to predict musical preferences above and beyond demographic variables. These findings advance previous theory and research and have direct applications for the music industry, recommendation algorithms, and health-care professionals.
Article
Full-text available
Why do we like the music we do? Research has shown that musical preferences and personality are linked, yet little is known about other influences on preferences such as cognitive styles. To address this gap, we investigated how individual differences in musical preferences are explained by the empathizing-systemizing (E-S) theory. Study 1 examined the links between empathy and musical preferences across four samples. By reporting their preferential reactions to musical stimuli, samples 1 and 2 (Ns = 2,178 and 891) indicated their preferences for music from 26 different genres, and samples 3 and 4 (Ns = 747 and 320) indicated their preferences for music from only a single genre (rock or jazz). Results across samples showed that empathy levels are linked to preferences even within genres and account for significant proportions of variance in preferences over and above personality traits for various music-preference dimensions. Study 2 (N = 353) replicated and extended these findings by investigating how musical preferences are differentiated by E-S cognitive styles (i.e., 'brain types'). Those who are type E (bias towards empathizing) preferred music on the Mellow dimension (R&B/soul, adult contemporary, soft rock genres) compared to type S (bias towards systemizing) who preferred music on the Intense dimension (punk, heavy metal, and hard rock). Analyses of fine-grained psychological and sonic attributes in the music revealed that type E individuals preferred music that featured low arousal (gentle, warm, and sensual attributes), negative valence (depressing and sad), and emotional depth (poetic, relaxing, and thoughtful), while type S preferred music that featured high arousal (strong, tense, and thrilling), and aspects of positive valence (animated) and cerebral depth (complexity). The application of these findings for clinicians, interventions, and those on the autism spectrum (largely type S or extreme type S) are discussed.
Article
By applying machine learning to the accurate and cost-effective classification of photos based on sentiment, we introduce a daily market-level investor sentiment index (Photo Pessimism) obtained from a large sample of news photos. Consistent with behavioral models, Photo Pessimism predicts market return reversals and trading volume. The relation is strongest among stocks with high limits to arbitrage and during periods of elevated fear. We examine whether Photo Pessimism and pessimism embedded in news text act as complements or substitutes for each other in predicting stock returns and find evidence that the two are substitutes.
Article
This paper constructs a manager sentiment index based on the aggregated textual tone of corporate financial disclosures. We find that manager sentiment is a strong negative predictor of future aggregate stock market returns, with monthly in-sample and out-of-sample R²s of 9.75% and 8.38%, respectively, which is far greater than the predictive power of other previously studied macroeconomic variables. Its predictive power is economically comparable and is informationally complementary to existing measures of investor sentiment. Higher manager sentiment precedes lower aggregate earnings surprises and greater aggregate investment growth. Moreover, manager sentiment negatively predicts cross-sectional stock returns, particularly for firms that are difficult to value and costly to arbitrage.
Article
Long-short anomaly returns are strongly related to the day of the week. Anomalies for which the speculative leg is the short (long) leg experience the highest (lowest) returns on Monday. The opposite pattern is observed on Friday. The effects are large. Monday (Friday) alone accounts for over 100% of returns for all anomalies examined for which the short (long) leg is the speculative leg. Consistent with a mispricing explanation, the pattern is driven by the speculative leg. The observed patterns are consistent with the abundance of evidence in the psychology literature that mood increases on Friday and decreases on Monday.
Chapter
Divorcing is a common-place practices that characterized our modern daily life. What is more thaninteresting to explore is why one of third takes after holidays. Are holidays accelerating the social decompositionthe founding parents of sociology denounced in earlier century? or simply, as Maccannellputs it, tourism serves as a valve (mechanism) of cohesion similarly to the role played by religion/totems in primitive societies? To answer these questions, we present statistical information given by theDirección General de Estadística y Censos (Ministerio de Hacienda GCBA). Although this informationhad its own limitations, which suggests more investigation is needed, we strongly beliefs the outcomesshed light to embrace the theory of vital cycles, as it has been formulated by Turner.
Article
We develop a new index of economic policy uncertainty (EPU) based on newspaper coverage frequency. Several types of evidence – including human readings of 12,000 newspaper articles – indicate that our index proxies for movements in policy-related economic uncertainty. Our US index spikes near tight presidential elections, Gulf Wars I and II, the 9/11 attacks, the failure of Lehman Brothers, the 2011 debt-ceiling dispute and other major battles over fiscal policy. Using firm-level data, we find that policy uncertainty is associated with greater stock price volatility and reduced investment and employment in policy-sensitive sectors like defense, healthcare, finance and infrastructure construction. At the macro level, innovations in policy uncertainty foreshadow declines in investment, output, and employment in the United States and, in a panel VAR setting, for 12 major economies. Extending our US index back to 1900, EPU rose dramatically in the 1930s (from late 1931) and has drifted upwards since the 1960s.
Article
This paper investigates the role of seasonal affective disorder (SAD) in the seasonal time-variation of stock market returns. SAD is an extensively documented medical condition whereby the shortness of the days in fall and winter leads to depression for many people. Experimental research in psychology and economics indicates that depression, in turn, causes heightened risk aversion. Building on these links between the length of day, depression, and risk aversion, we provide international evidence that stock market returns vary seasonally with the length of the day, a result we call the SAD effect. Using data from numerous stock exchanges and controlling for well-known market seasonals as well as other environmental factors, stock returns are shown to be significantly related to the amount of daylight through the fall and winter. Patterns at different latitudes and in both hemispheres provide compelling evidence of a link between seasonal depression and seasonal variation in stock returns: Higher latitude markets show more pronounced SAD effects and results in the Southern Hemisphere are six months out of phase, as are the seasons. Overall, the economic magnitude of the SAD effect is large.
Article
That music affects human beings in various ways has probably been presumed as long as people have played music. Many marketing practitioners already accept this notion, given that music is increasingly used as a stimulus in the retail environment as well as in radio and television advertising. Yet, fewer than 20 published empirical studies in marketing have music as their focus. The author reviews the small body of marketing literature, surveys relevant literature outside marketing, and provides research propositions to guide future studies.