I.INTRODUCTION

Social network sites, known as SNS, are a powerful tool for reaching customers and building a strong brand digital presence. Social media have modified the way companies promote their products. Posts advertising new features or sales periods appear in, most of the time, potential customers‟ news feeds. Company news and promotions published through different social channels are generating growing amounts of data. Business analysts search for valuable signals drawn from such data (Bello-Orgaz, Jung, & Camacho, 2016).

Platforms such as Twitter and Facebook have gained considerable popularity among consumers, and corporations managers are realizing the huge potential that social media has to engage with customers and reach new audiences (Gallaugher & Ransbotham, 2010).

Managers have been focused on the central question whether their firm‟s presence in social media add value to the firms bottom line (Kaske et al., 2012), or it is just the new buzzword. Since strategic decisions in a firm are mainly based on profit maximization, managers need to verify whether social medias long-term benefits outweigh the costs of the investment (Lenskold, 2003).

Quantifying the online attention received by companies and their correlations with financial indicators across time might help to understand better the impact that interactions in these new channels have on enterprises‟ profitability (Vaca Ruiz et al., 2014). It is needed to design strong signals to approximate real companies‟ performance from existing data.

In this study, we aim to find out relationships between financial performance of a firm and online attention metrics measured with Facebook and Twitter interactions. Finding a pattern that relates social media activity with financial indicators can help firms to track the customer perception across time with a low-cost approach. We have collected Facebook and Twitter data because those platforms are the most influential SNS in Ecuador. However, the metrics proposed can be calculated on datasets collected from any other platform such as Instagram or LinkedIn. Correlating online signals with offline performance we make three main contributions:

∙Design quantitative metrics for industries to measure the activity and attention they attract in SNS.

∙Visualize the existing correlations between social attention and financial performance reached by companies in different industries in Ecuador.

∙Propose a model to predict the revenue of a firm based on the normalized attention it receives of Facebook.

The paper proceeds as follows. Section II gives a brief overview of previous researches and related work. Section III describes the data, the time and its features. The theory and the statistical methods are described in Section IV. Section V and VI provide the results and the details of our findings. Finally, in Section VII we make some conclusions and suggest future research directions.

II.RELATED WORK

There is a considerable amount of literature on sentiment analysis and its relationship with the financial performance of a firm like stock returns. Sentiment analysis is the computational study of opinions, feelings and emotions expressed in text. The use of data mining techniques to predict the financial markets has been extensively studied in numerous publications (Schumaker & Chen, 2009). The effect of social media and conventional media, their relative importance, and their interrelatedness have been studied on short term firm stock market performances (Yu et al., 2013). The relationship between social media-based brand engagement and purchase intention has been found to be associated with a potential return on investment (Coursaris et al., 2016).

A framework specifically can foster the understanding of customer centered social media and provides executives with a tool to decide on whether to implement social media or not. Selected empirical examples illustrate that social media lead to positive ROI across the analyzed social media platforms (Kaske et al., 2012). Web visibility was found as a significant determinant of firm web traffic, and it improves both firm short- and long-term performances. It has also revealed that web visibility creates higher values for business-to-consumer (B2C) than business-to-business (B2B) firms, for firms with a value appropriation focus than those with a value creation focus, and for firms of search goods than those of experience goods (Wang & Xu, 2017).

A recent review of the literature on this topic found that a learn model is able to predict the social network sentiment and the stock performance; and its recent news and social data are also closely correlated (Attigeri et al., 2015). In statistical tests, the correlation of Facebook brand page fan counts and their respective brand company stock prices were found to be statistically significant (O‟Connor, 2013). And the accuracy of DJIA (Dow Jones Industrial Average) predictions can be significantly improved by the inclusion of specific public mood dimensions, like Twitter public sentiment, but not others (Bollen et al., 2010).

III.DATASET

Using Facebook and Twitter APIs, we collect social media data for 32 Ecuadorian firms. Such firms were selected based on the ranking built by Llorente & Cuenca in 2015 (Rivas & Llanos, 2016). In such ranking, socially engaged companies are ordered according to an influence measure proposed by Llorente & Cuenca. We started searching information for the top 50 companies in the ranking. However, we realized that just 32 companies in that list could be analyzed since the others were brands of bigger enterprises and the financial statements were not disaggregated.

For the financial indicators dataset, we collected the public data available at “Superintencia de Compañías del Ecuador” website. There were three companies with more than one brand, and instead of using their official social media accounts, we chose to use their most popular brands accounts. For instance, in the case of Cervecería Nacional we selected the social media accounts of „Pilsener‟ and „Club Premium‟. For Nestle we used „Galletas Amor‟ and „Nescafe‟. Finally, for DK Management we used „Quicentro Shopping‟ and „San Marino Shopping‟. We took such decisions since we can assume that the social media activity of these companies is well represented by their most popular brands. In Figure 1 we show the process to get the results of each model used in this research.

The Facebook dataset was collected using the Graph API that let us to obtain posts, likes, shares and their corresponding meta data for the 32 firms since they maintain public accounts. For the Facebook platform, we obtained the total number of likes, posts and shares by each of the 32 firms. The resulting dataset contains 94,542 posts with a total of 1,829,080 likes and 158,276 shares received across time.

Fig. 1
Architecture used.
Datasets extracted from Social Media and financial statements are fed to the model to predict financial performance of a company.

The Twitter dataset was obtained using the Twitter REST API to collect, for each of the firms in the study, tweets, retweets, favorites and the corresponding dates. The total number of tweets collected were 111,473 with a total of 47,378 favorites and 20,828 retweets. Due to the Twitter API rate limits and the different level of activity, for some firms we collected data starting two or three years ago, and for some firms, data corresponding only to this year. Figure 2 shows the number of posts and tweets published for each firm using their national accounts in 2017, since January to July. The 32 firms were categorized in 9 industries as shown on Table 2, in which we provide the companies that belong to each industry.

To summarize we collected six social media metrics from Facebook and Twitter that are the average likes, average shares, average retweets, average favorites and the total number of followers in each SNS. And we collected six financial indicators that are the total amount of assets, equity, cash, revenue, profit and advertising expenditure. Table 1 provides the principal variables and its description. In Section 5 we discuss and analyze the online activity and attention using social media metrics of Twitter and Facebook in each industry.

Tab. 1
Variable Description.

Fig. 2
Posts and Tweets per firms of 2017.

Tab. 2
Industry Description.

IV. METHODOLOGY

We quantify social media engagement of a company using activity and attention indicators. Activity indicators correspond to the number of posts, reposts, and comments done, while attention is determined by the number of reposts or comments received from others (Vaca Ruiz et al., 2014). To measure activity, we consider the number of posts and tweets published by the company on Facebook and Twitter respectively. On the other hand, to measure attention we count the number of likes and retweets that the company received.

In order to describe and find which industries generate and attract more activity and attention in Twitter and Facebook we use the tweets and posts of the first seven months of 2017. To analyze the activity, we sum the posts and tweets in each month of every company, and then joining that by industry. In order to analyze the online attention, we defined the term: “normalized attention” in Equation 1.

[Ec. 1]

Where:

(1)

∙metrict: Total of a determined social media metric in the post “t”.

∙posts: Total posts of the industry in a month.

∙normalized_attentioni: Normalized attention of the industry “i” in a month.

To find a relationship between normalized metrics and financial indicators we use the Spearman‟s rank correlation (Spearman, 1904) that is a nonparametric method for the Pearson product-moment correlation. This method was chosen because it is one of the most practical ways to find a relationship in small samples, also it is not necessary to assume that we have normal data. For Spearman‟s correlation we have the following formula:

[Ec. 2]

We can test if the correlation is significant with 95% of confidence. Using correlations as a first step, we propose next a linear regression model to predict financial performance from normalized attention.

We chose the average likes of a firm‟s Facebook, to predict the annual revenue of a company. In Figure VII Advertising Expenditure and Revenue have the highest correlation with the average likes of a post in Facebook. Given the inherent power of popularity to foster social conformity, enhanced by the opinion- influencing capacity of social media (O‟Connor, 2013), average likes per post of the most popular firms are associated with the Revenue of the consumer brand companies that own these brands. This relationship is represented by the following formula:

[Ec. 3]

Where:

∙metrici: normalized attention metric in a firm

∙Revenuei: Revenue of a company

∙i: Error term

∙β: Return of an additional like you receive in a post

∙α: Normalized attention of the industry “i” in a month

The results of the regression analysis and correlation analysis are specified in Section VI.

V.ACTIVITY AND ATTENTION

The online activity of a firm can be measured by the total of posts and tweets published in a given period of time. The online attention can be measured by the total of shares and likes in Facebook, and the total of retweets and favorites in Twitter. An additional metric to quantify online attention is the total of followers in the respective SNS. We aim to analyze temporarily the social activity and attention of the industries, using posts and tweets for activity, and likes and retweets for the normalized attention defined in Equation 1.

Twitter

To analyze the behavior of each industry in Twitter, we use the tweets of the 32 firms in 2017. Figure 3 shows the activity on Twitter of each industry. Telecommunication‟s industry is the most active on Twitter, with two peaks in February and June. In contrast, the Clothing industry is the least active in this platform. Interestingly, the first months of the year are the most active in most of the industries. To know the effectiveness of each tweet in the audience of the company, we chose to quantify the industry‟s normalized attention by apply in Equation 1 using retweets for the variable metric. In Equation 1 we use the total retweets of a given tweet in the metric variable.

In Figure 4 we reveal the normalized attention and we notice that there is not any industry with a sustained level of average attention during the first seven months of 2017. However, we observed that Beverage and Food industry in January and February have a high level of normalized attention, and Retail industry represented by “Almacenes Tia” have a high peak of normalized attention in July.

Previous studies have shown that a higher level of activity do not necessarily correspond to a higher level of online attention (Vaca Ruiz et al., 2014). We observe that such pattern is also reflected in Ecuadorian Companies.

Facebook

We also analyze the posts published in the public Facebook pages the 32 firms during 2017. In Figure 5 we present the activity on Facebook of each industry. The Automobile industry has the highest level of activity, being March the month with more posts in Facebook. There are other industries like Retail, Banking and Entertainment with more posts on March than the other months. In contrast, the Food and Beverage and Telecommunication industries has the lowest activity of the nine sectors.

We chose the Likes metric to calculate the normalized attention in an industry. In Equation 1 we use the total likes of a given post in the metric variable. Figure 6 shows the normalized attention of each industry in 2017. The Automobile industry and the Telecommunication industry have the highest level of normalized attention. Instead of this, the Food and Beverage industry and Restaurant industry have the lowest normalized attention.

We can see several differences between two figures of each SNS, and conclude that there are not patterns between the social activity of monthly posts and the normalized attention measured by the average likes and retweets in an industry. Based on the related work (Section II) we test the hypothesis that the six social media metrics: average likes, average shares, average retweets, average favorites and total followers; individually are positively correlated with the six financial metrics: assets, equity, cash, revenue, profit and advertising.

Fig. 3
Activity in Twitter

Fig. 4
Normalized Attention in Twitter

Fig. 5
Activity in Facebook

Fig. 6
Normalized Attention in Facebook

VI. RESULTS

Correlation Analysis

H1. Average Likes. We correlated the average likes per firm with each financial metric and found that in the six cases there is a positive correlation. We obtained r = 0.61 for Assets and Equity, r = 0.54 for Cash, r = 0.69 for Income, r = 0.63 for Profit and r = 0.70 for Ads.

H2. Average Shares. We correlated the average shares per firm with each financial metric and found that in the six cases there is a positive correlation between them. We obtained r = 0.45 for Assets and Equity, r = 0.49 for Cash, r = 0.57 for Income, r = 0.50 for Profit and r = 0.61 for Ads.

H3. Followers Facebook. We correlated the followers of Facebook of each firm with each financial metric. There was no correlation in any case except with Advertising expenditure, obtaining r = 0.48. It is a well-known phenomenon that the number of followers do not correlate well with influence on financial performance (Cha et al., 2010).

H4. Average Retweets. We correlated the average retweets per firm with each financial metric. There was no correlation for any of the firms.

H5. Average Favorites. We correlated the average favorite per firm with each financial metric. There was no correlation in any case.

H6. Followers Twitter. We correlated the followers of Twitter of each firm with each financial metric. There was no correlation in any case.

Fig. 7
Spearman Correlation heatmap

We learned through the analysis (H4, H5) that for Ecuadorian audience the attention obtained in Facebook is notoriously higher but also it reflects better the financial performance of the company than the activity in Twitter.

All the attention metrics and financial indicators are log-transformed, to counter for skewness before any calculation of correlations. The results obtained are statistically significant, at least with p-value < 0.05.In Figure 7 there is a Heat Map of the correlations between attention metrics and financial metrics. Through the analysis we observed that the correlation between financial performance and the banks was really poor. Therefore, we removed the Banking industry from the analysis, its social attention is not very strong in the Ecuadorian market. We can conclude that attracting social attention is not a priority for Ecuadorian Banking.

Analyzing Facebook‟s social attention, the Advertising Expenditure has the highest correlation with attention metrics. Specially for likes and shares the correlation is 0.7 and 0.61 respectively. For Revenue there is also high correlations with likes and shares, specifically 0.69 and 0.57. This supports previous findings in the literature about the relationship between social attention and the financial performance of a firm.

Regression Analysis

For a more precise measure of the relationship among the variables, a regression analysis is expressed in Equation 3. Particularly, Model 1 defines the revenue of a company as the dependent variable and the average likes for the independent variable, as we show in Equations 4. Model 2 is defined in Equation 5 using the revenue of a company for the dependent variable and the average shares for the independent variable.

[Ec. 4]

[Ec. 5]

In Table 3 we showed the results for the regression analysis. The variables are log-transformed to correct skewness and the banking and retail industries were separated from the sample. The results revealed are statistically significant, at least with p-value < 0.01. We selected av_likes and av_shares as independent variables for own model predictive since those are the variables with a higher correlation results.

In the case of the Model 1 (Equation 4) where the normalized attention metric is the average likes per post, if the average likes changes by one percent, we would expect the revenue to change by 0.74%. In Model 2 (Equation 5) if the average shares changes by one percent, we would expect the revenue to change by 0.64%. Finally, the Adjusted R2 is much better in Model 1, with 0.40, than Model 2, with 0.28.

Tab. 3
Linear Regression Models

VII. CONCLUSIONS

The evidence from this study points towards the idea that there is a relationship between social attention metrics and the financial performance of a firm. Our findings are in line with previous results even when there is not available stock data for Ecuador to make statistical analysis. We are aware that our research may have two limitations. The first is that it is not always easy to find the financial metrics of a specific brand in companies that agglo has a lot of them. The second is that stock market data could be a more reliable metric to measure the financial performance.

Unlike other research carried out in this area, we did not find a significant correlation between Twitter metrics and financial indicators. This apparent lack of correlation can be attributed to the fact that Ecuadorian consumers are not highly connected to this social network. Apparently, Facebook could be a most powerful and effective tool to reach and engage new customers.

We have confirmed that Facebook metrics have relationship with financial metrics, specially revenue and advertising expenditure. We also propose a specific model to study the relationship between the revenue and Facebook metrics, given certain conditions. This cutting-edge model for firm‟s valuation is still developing new practical and theoretical implications, so to further our research we plan to make a sentiment analysis to the comments and tweets of this SNS and financial performance of firms.

Referencias

[1]Attigeri, G. V, Manohara Pai M M, Pai, R. M., & Nayak, A. (2015). Stock market prediction: A big data approach. In TENCON 2015 - 2015 IEEE Region 10 Conference (pp. 1–5). IEEE. https://doi.org/10.1109/TENCON.2015.7373006

[2]Bello-Orgaz, G., Jung, J. J., & Camacho, D. (2016). Social big data: Recent achievements and new challenges. Information Fusion, 28, 45–59.

[3]Bollen, J., Mao, H., & Zeng, X.-J. (2010). Twitter Mood Predicts the Stock Market. Computers in Human Behavior, 15(4), 506–515.

[4]Cha, M., Haddadi, H., Benevenuto, F., & Gummadi, K. P. (2010). Measuring User Influence in Twitter: The Million Follower Fallacy. AAAI Conference on Weblogs and Social Media, 14.

[5]Coursaris, C. K., v. Osch, W., & Balogh, B. A. (2016). Do Facebook Likes Lead to Shares or Sales? Exploring the Empirical Links between Social Media Content, Brand Equity, Purchase Intention, and Engagement. In 2016 49th Hawaii International Conference on System Sciences (HICSS) (pp. 3546– 3555).

[6]Gallaugher, J., & Ransbotham, S. (2010). SOCIAL MEDIA AND CUSTOMER DIALOG MANAGEMENT AT STARBUCKS. Mis Quarterly Executive, 9(4), 197–212.

[7]Kaske, F., Kugler, M., & Smolnik, S. (2012). Return on Investment in Social Media--Does the Hype Pay Off? Towards an Assessment of the Profitability of Social Media in Organizations. In 2012 45th Hawaii International Conference on System Sciences (pp. 3898–3907). IEEE. https://doi.org/10.1109/HICSS.2012.504

[8]Lenskold, J. (2003). Marketing ROI: the path to campaign, customer, and corporate profitability. McGraw-Hill.

[9]O‟Connor, A. J. (2013). The Power of Popularity. Social Science Computer Review, 31(2), 229–235.

[10]Rivas, A., & Llanos, J. C. (2016). Mapa de poder en la red Ecuador 2015.

[11]Schumaker, R. P., & Chen, H. (2009). Textual Analysis of Stock Market Prediction Using Breaking Financial News: The AZFin Text System. ACM Trans. Inf. Syst., 27(2), 12:1--12:19.

[12]Spearman, C. (1904). The Proof and Measurement of Association between Two Things. The American Journal of Psychology, 15(1), 72–101. Retrieved from http://www.jstor.org/stable/1412159

[13]Vaca Ruiz, C., Aiello, L. M., & Jaimes, A. (2014). Modeling dynamics of attention in social media with user efficiency. EPJ Data Science, 3(1), 5.

[14]Wang, F., & Xu, B. (2017). Who Needs to Be More Visible Online? The Value Implications of Web Visibility and Firm Heterogeneity. Inf. Manage., 54(4), 506–515.

[16]Yu, Y., Duan, W., & Cao, Q. (2013). The impact of social and conventional media on firm equity value: A sentiment analysis approach. Decision Support Systems, 55(4), 919–926.

Notas de autor

paestrad@espol.edu.ec