Music sentiment and the stock market in Vietnam

Abstract

Purpose

By extending Edmans et al.’s (2021) music sentiment measures to the Vietnam market, the authors aim to investigate the impacts of music sentiment on stock market returns and volatility.

Design/methodology/approach

The authors adopted Edmans et al.’s (2021) music-based sentiment to proxy for investor mood. The current study uses linear regression analysis.

Findings

The authors find that music sentiment is significantly and positively related to both stock returns and stock market volatility. The authors also show that music sentiment has a contagious effect: Global music sentiment and those in the United States, France and Hong Kong are significant drivers of the Vietnamese stock market. The authors also examine the effect on different industry returns and find that returns on stocks of firms in the communication services, consumer discretionary, consumer staples, energy, financials, healthcare, real-estate, information technology and utility sectors are significantly related to music sentiment. In addition to valence, the authors find that other Spotify audio features can be used to quantify music sentiment.

Originality/value

This study contributes to the behavioral finance literature that focuses on investor sentiment. The authors address this topic in Vietnam using high-frequency data.

Citation

Can, T.L., Le, M.D. and Yu, K.-C. (2024), "Music sentiment and the stock market in Vietnam", Journal of Asian Business and Economic Studies, Vol. 31 No. 1, pp. 74-83. https://doi.org/10.1108/JABES-07-2022-0170

1. Introduction

Classical finance theories hinge on the assumption that stock markets are efficient, which means that investors have all available information and act rationally on that information. Nevertheless, this school of thought has failed to explain several observed calendar anomalies, such as the January (Gultekin and Gultekin, 1983; Rozeff and Kinney, 1976; Wachtel, 1942), weekend and reverse-weekend (Bampinas et al., 2016; Brusa et al., 2000, 2003, 2005, 2011; Olson et al., 2015) effects. In the 1990s, finance researchers started to apply theories underpinned by psychology to adjust robust investor-rationality assumptions (Lo, 2004; Shiller, 2003); therefore, behavioral finance can superiorly explain stock price fluctuations that classical finance theory cannot (Baker and Wurgler, 2007; Brown and Cliff, 2004; Kim et al., 2014; Ryu et al., 2017).

Against this backdrop, we contribute to the literature in several respects. First, the study is among the first to investigate the impacts of music-based sentiment measures in Vietnam. Compared with those in neighboring countries, Vietnam's stock exchanges are young, are underdeveloped and contain several noise traders. In this setting, most stocks are traded based on investor sentiment and daily mood rather than on technical analysis (Truong et al., 2021). However, Vietnamese citizens more often choose to listen to music than other relaxing or reactional activities [1]. In addition, in Vietnam, stocks are more frequently traded than other assets, such as bonds, derivatives or currency. These specific features offer an ideal environmental setting to test the effects of music-based sentiment on the stock market. Focusing on studies in Vietnam alone, Truong et al. (2021) and Phan et al. (2021) found that investor sentiment significantly affected stock market returns. In our study, we find that music sentiment drives stock market returns and volatility. Second, we extend Edman et al.’s (2021) research by tracing the baseline findings in several respects, and we explore whether music sentiment affects stock market returns differently in different industry sectors and when Spotify's audio features are used as proxies for music sentiment. Third, consistent with the current literature, we confirm both that music sentiment has contagious effects on the Vietnamese stock market (Aissia, 2016) and that limits to arbitrage drive the effects of music sentiment on stock returns (Fernandez-Perez et al., 2020).

The remainder of this paper is structured as follows: First, we present our research designs and measurements. We then present the research results and conclude this paper.

2. Data and methodology

2.1 Main variable

2.1.1 Music sentiment

In the literature, authors have proposed two core approaches to quantifying unobserved investor sentiment: the exogenous and endogenous approaches (see Edmans et al., 2021). We intend to use a music-based sentiment measure developed by Fernandez-Perez et al. (2020) and Edmans et al. (2021) because it solves the problems of language divergence, increases global comparability and is highly frequent and available in most countries.

According to Spotify's application programming interface (API), the word “valence” is a metric to measure whether a song makes the listener feel happy (high valence) or sad (low valence). The scale of valence ranges from 0.0 to 1.0. We calculated the stream-weighted average valence (SWAV) (at time t), SWAV_t, based on songs' streams, Stream_it, which refers to the frequency with which the i-th song (i=1,200‾) is played for at least 30 s, as follows:

SWAVt=∑i=1200(Streamsit∑i=1200StreamsitValenceit)

Our music-based mood proxy at time t, Music Sentiment_t, is thus defined as changes in the SWAV of the top 200 songs to which the Vietnamese listen at time t:

Music Sentimentt=∆SWAVt=SWAVt−SWAVt−1

2.1.2 Market reactions

We proxy Vietnamese stock market returns using the VN-Index (VNI) (Truong et al., 2021). The VNI is a market capitalization-weighted index of 533 companies listed on the Ho Chi Minh Stock Exchange (HOSE) and is the most widely quoted Vietnam stock market index in the international press.

The so-called variable stock market returns (Ret) are constructed as the daily returns on the VNI. The volatility of stock market returns is the standard deviation of daily stock market returns from t−1 to t (7-day calendar window) [2].

(1)Rett=VNIt−VNIt−1VNIt

(2)σt=∑i=t−7t(Reti−Ret(t−7):t‾)24

2.2 Econometric models

We estimate the following time-series model to examine the relationship between music sentiment and stock market returns:

(3)Rett=α+∑i=07βi∆SWAVt−i+∑Γ.Controls+Time dummies+εt

In this model, Ret_t denotes market returns at time t. ΔSWAV_t−i denotes music sentiment at time t−i.

It is believed that investor sentiment and noise trading can impact the volatility and level of asset prices (Black, 1986; Edmans et al., 2021; De Long et al., 1990) because sentiment should cause prices to first deviate from the fundamentals and then correct. Thus, a large sentiment may cause more trading, thus affecting stock price volatility. To test this hypothesis, we estimate the following equation:

(4)σt=α+βi|∆SWAVt|+∑Γ.Controls+Time dummies+εt

In the aforementioned function, σ_t denotes the volatility of stock market returns over the period t−7 to t (one-week window). |∆SWAVt| denotes the average absolute music sentiment for the period t to t−7 (one-week calendar window).

In both models, Controls denotes control variables, including economic growth as proxied by gross domestic product (GDP) growth rates. Following the literature (Edmans et al., 2021; Fernandez-Perez et al., 2020), we add weekday and month dummies to control for time-fixed effects in terms of the seasonal mood and swings not captured by our music-based sentiment measure. We are interested in β_i, which represents market reactions to music sentiment.

2.3 Data and sample

The current study uses a time-series dataset framed using 1,474 observations from March 15, 2018, to March 25, 2022, based on the availability of valence data. We refer to and replicate the code from the GitHub source [3] to collect all Spotify audio features from the top 200 songs listened to daily in Vietnam; we then construct the weighted average for each audio feature, including valence, speechiness, danceability, energy, key, loudness, mode, acousticness, instrumentalness, liveness, tempo, duration and time signature. The stream-weighted valence of the top 200 songs is the main variable as a music-based mood proxy. In addition, we repeat these steps to construct the SWAV across the top 200 songs for the global, United States (US), UK, France, Hong Kong, Japan, Korea, Singapore, Thailand and Taiwan scenarios. Investors from these countries have inflows of foreign direct investment and foreign indirect investment in Vietnam, according to the General Statistics Office [4] of Vietnam and other sources on the internet (see Subsection 3.3.3).

We then collect all the daily stock market returns for the VNI and HNX (the Hanoi Stock Exchange) with the symbols “VNI” and “HNXI,” respectively, obtained from a financial platform and news website https://www.investing.com. Industry-specific return data are obtained from Capital IQ, and we classify common stock tickers from the two stock market exchanges, HOSE and HNX, into 11 industry sectors based on the Global Industry Classification Standard [5]. We then construct the daily average return for each sector (see Subsection 3.3.1). To test the limit arbitrage, we also use Capital IQ data to construct the high-beta and low-beta portfolios (see Section 3.3.4).

For the controlled variables, we collect quarterly data on the GDP growth rate from the World Bank database.

3. Empirical results

3.1 Descriptive statistics

We identify 3,475 unique songs with over 1.65 billion streams in total. In addition, an average of 1.12 million streams occur daily, with approximately 6,170 streams per song [6]. We then construct a daily SWAV across the top 200 songs. The daily SWAV ranges from 0.3909 to 0.5451, with an average of 0.4556. Table 1 (please see it on the Online Appendix) describes the summary statistics for the SWAV, music sentiment and GDP growth rate. Our music sentiment measure ranges from −0.043% to 0.050%.

Table 2 (please see it on the Online Appendix) shows the correlation coefficients of the independent variables. The correlation coefficients of all pairs fall in the interval [−0.7; 0.7]; therefore, colinearity is not a concern.

Figure 1 plots a time series of the daily SWAV and VNI over our sample period from March 15, 2018, to March 25, 2022, for Vietnam. Overall, these indexes share fairly common trends.

3.2 Baseline results: market reactions

In Table 3 (please see it on the Online Appendix), we examine the relationship between music sentiment and market reaction. We find a strong positive relationship between the VNI daily return (Ret_VNI) and the music sentiment one day before, as shown in the second column of Table 3. Even after controlling for GDP growth rates [7] and time-fixed effects (weekdays and months), the coefficient of music sentiment remains positive and significant. Focusing on the third column, we observe that a one-unit increase in ΔSWAV_t−1 leads to a 0.1767 increase in the next day's VNI daily return. We also observe that a one-unit increase in weekly music sentiment (ΔSWAV_(t;t−7)) leads to a 0.0459 increase in the volatility of the VNI (Column 6).

3.3 Further analysis

3.3.1 Industry-specific returns

First, the behavioral-finance literature has shown that the relationship between investor sentiment and stock returns varies across industries. For instance, Sayim et al. (2013) found that although investor sentiment positively influenced stock returns in the US auto, finance, food, oil and utility industries, it negatively impacted stock volatility only in the US auto and finance industries. Most recent studies focus on the aggregate stock market returns; therefore, filling the gaps regarding industry-specific stock returns would extend the behavioral-finance literature. In this study, we conducted empirical tests across different industry sectors in two major stock exchanges: the HOSE (VNI) and HNX [8]. The construction of industry-specific stock returns was as follows. First, we collected data on the daily stock prices for all Vietnamese enterprises in the Capital IQ database and categorized them based on their industries (including communication services, consumer discretionary, consumer staples, energy, financials, healthcare, real-estate and utility firms) and trading platforms (the HOSE and HNX). We then computed the daily returns using Equation (1). The final industry-specific stock returns in an industry were computed as the average stock returns for all the firms in that industry.

Table 4 (please see it on the Online Appendix) shows the regression results for stocks listed on the HOSE. Overall, we observe that music sentiment is significantly related to stock returns for communication services, consumer discretionary, consumer staples, energy, financials, healthcare, real-estate and utility firms. We also find an inverse relationship between music sentiment and stock returns in the communication services, consumer discretionary, healthcare and utility sectors.

Table 5 (please see it on the Online Appendix) shows the regression results for stocks listed on the HNX. Overall, we observe that music sentiment is significantly related to stock returns for communication services, consumer discretionary, energy, financials, healthcare, information technology and utility companies. Specifically, the financials and healthcare sectors exhibit an inverse relationship between stock returns and music sentiment.

3.3.2 Spotify audio features

In addition to valence, Spotify provides other audio features [9]. Therefore, we replace “valence” with other audio features defined by Spotify's API in calculating music sentiment and repeat the baseline regressions to determine whether other features can replace valence. This also provides a robustness test for our baseline findings. Table 6 (please see it on the Online Appendix) exhibits the estimation results for Spotify's audio features. Overall, in addition to valence, music sentiment measures that utilize other features (e.g. danceability, energy, loudness, mode, acousticness, duration and time signature) show a significant relationship with the VNI daily return, with some showing (e.g. danceability and mode) strong impacts at the 1% significance level. Among them, only danceability has a contemporaneous impact. The time-, duration-, acousticness- and mode-based music sentiment measures show an inverse relationship with stock returns.

3.3.3 Global music sentiment and the Vietnam stock market

A few studies highlight that investor sentiment can have a contagious effect, that is, foreign and global investor sentiment may impact local stock markets (Aissia, 2016). Therefore, we replace the Vietnamese music sentiment with other markets' music sentiment and repeat the baseline regression analysis. Table 7 [10, 11] (please see it on the Online Appendix) reports the regression results for music sentiment in selected markets, which show that, as well as the home music sentiment, foreign music sentiment is a significant predictor of stock market returns. Overall, we find that global music sentiment and those in the US, France, Hong Kong, Japan and Thailand are strongly related to Vietnam's stock market returns. Among the selected markets, US and Hong Kong music sentiments show a negative relationship with Vietnam's stock market returns, suggesting an inverse effect. Global music sentiment and those in the US, France and Hong Kong have contemporaneous impacts on Vietnam's stock market returns.

3.3.4 Limits to arbitrage

Limits to arbitrage can exacerbate the effect of investor sentiment on asset prices (Fernandez-Perez et al., 2020; Shleifer and Vishny, 1997). In this sense, we should expect different effects of music sentiment for stocks with greater limits to arbitrage (e.g. small stocks, stocks with high-beta and illiquid stocks) and those with lower limits to arbitrage because retail investors are more likely to hold them disproportionally and speculatively trade them based on sentiment due to information scarcity (Fernandez-Perez et al., 2020; Truong et al., 2021). To test this prediction, we repeat the baseline regression by replacing the stock market index with a market capitalization portfolio, a high-beta portfolio and a liquidity-based portfolio [12]. Table 8 (please see it on the Online Appendix) shows that stocks with lower limits to arbitrage are more strongly impacted by music sentiment.

4. Conclusions

First, we again confirm that music sentiment significantly affects stock market returns and volatility and show that it is a proxy for mood (Edmans et al., 2021; Fernandez-Perez et al., 2020). Second, the additional analyses again confirm the role of limits to arbitrage in driving the effects of music sentiment on stock market returns (Fernandez-Perez et al., 2020). Moreover, we find that the effects of music sentiment on stock market returns vary across different industry sectors and are contagious. Finally, we show that besides valence, other Spotify audio features (e.g. danceability, energy, loudness, acousticness, mode, duration and time signature) can also be used to quantify investor sentiment.

Our study has limitations that future researchers must consider. First, recent literature highlights the nexus between changes in trading volume and stock market returns (Chordia and Swaminathan, 2000; Chen, 2012; Lee and Rui, 2002). Our study does not include analyses regarding this important aspect due to a scarcity of sources of the necessary data, and thus, we leave this for future studies. Second, using Spotify data may introduce substantial biases. For instance, despite acknowledging that Spotify audiences are mostly from Generation Z, who may not be major retail investors, we could not conduct a robustness test on this issue because of a lack of access to data on Spotify's audience demography (e.g. age and gender). In addition, Spotify's market share is not high in Vietnam compared to those of local platforms, which challenges the representativeness of the data used in the formal analysis. Finally, using daily stock returns in our baseline model may introduce bias due to non-synchronicity between the opening and closing times of the HOSE and the time of day that Spotify reports its daily statistics. Future studies should consider using weekly stock returns, as stated in Edmans et al. (2021).

Figures

Figure 1

Daily stream-weighted average valence and stock returns from March 15, 2018, to March 25, 2022, in Vietnam

Notes

1. Retrieved from: https://www.slideshare.net/asiaplus_inc/survey-about-vietnamese-music-listening-behavior

2. The denominator in Equation (2) is 4 (=5−1) instead of 6 (=7−1) because we eliminate Saturday and Sunday in the construction. These days are non-trading in Vietnam.

3. Retrieved from: https://github.com/simon-th/spotify-data project/blob/master/Extracting%20Spotify%20Audio%20Features.ipynb

4. General Statistics Office of Vietnam website: https://www.gso.gov.vn/en/homepage/

5. The Global Industry Classification Standard (GICS) is an industry taxonomy developed by Morgan Stanley Capital International (MSCI) and Standard & Poor's in 1999 for use by the global financial community.

6. The songs with the highest valence in our sample are Running Over (feat. Lil Dicky) by Justin Bieber (valence of 0.977), Ăn Sáng Nha – Original Version by ERIK (0.974), and There's Nothing Holdin’ Me Back by Shawn Mendes (0.969). The songs with the lowest valence are The Plan - From the Motion Picture “TENET” by Travis Scott (valence of 0.0363), Wednesday Afternoon by FAIR GAME (0.0359) and Lo Vas A Olvidar (with ROSALÍA) by Billie Eilish (0.0320)

7. The coefficient of GDP shows that a higher GDP growth rate would lead to lower stock market returns in Vietnam, indicating that Vietnamese stockholders do not necessarily benefit from economic growth. This surprising result is consistent with Ritter (2005).

8. The difference between these two stock market platforms is presented in the supplemental materials. We limit the information due to space considerations.

9. Retrieved from: https://developer.spotify.com/documentation/web-api/reference/#/operations/get-audio-features

10. Retrieved from: https://www.gso.gov.vn/en/homepage/

11. Retrieved from: https://fia.mpi.gov.vn/en/

12. For the size-based portfolios, we use the VN-Small Cap Index and VN30-Index for small and large stock portfolios, respectively, which are collected from https://www.investing.com. As proxies for the beta-based portfolios, we first calculate each stock's beta (compiled from the Capital IQ database) and then form high-beta (greater than 1) and low-beta (less than 1) stock portfolios. Finally, to represent the liquidity-based portfolios, we use VNAllShare Index and UPCOM Index, obtained from https://www.investing.com

Appendix

The supplementary materials for this article can be found online.