GOOGLE TRENDS AND STOCK RETURNS A STUDY OF INVESTOR SENTIMENTS USING BIG DATA. School of Business, Amrita Vishwa Vidyapeetham, Coimbatore.

Transcription

1 Volume 118 No , ISSN: (on-line version) url: ijpam.eu GOOGLE TRENDS AND STOCK RETURNS A STUDY OF INVESTOR SENTIMENTS USING BIG DATA 1 Hari Krishnan.A.V, 2 Gopakumar.V, 3 Varsha Sureshkumar School of Business, Amrita Vishwa Vidyapeetham, Coimbatore. 1 avhari47@gmail.com, 2 gopakumarv@cb.amrita.edu, 3 s_varsha@cb.amrita.edu Abstract: This paper tries to check for the presence or absence of a relationship between the individual returns of the stocks in NIFTY50 and Google trends data for the word buy <company name>. We have employed Granger causality test to investigate the association between Google Trends data and individual returns. For the study, the adjusted closing prices of Nifty 50 shares for a period of 5 years is used to compute the weekly stock yield and the Google trend data from India for the keyword buy<company name> is taken. It was observed that the returns of some individual stocks showed a bi-directional causality while some showed a unidirectional causality. Among NIFTY 50 stocks, many did not show any causal relationship between the weekly returns and the relevant Google Trend keyword: buy<company name>. Keywords: Google trend data, Unit root test, ranger causality test, Buy, Nifty 50 returns. 1. Introduction Investor sentiment acts as a good indicator for other investors for trading in stock markets traditionally. With the advent of the Internet and easy availability of information, market sentiment has recently been observed online too, where investor sentiments are studied based on the search keywords about companies. The keyword buy is usually used along with a company name in websites like google.com when a person searches for information, indicating an interest to buy stocks of that company. Prospective investments decisions are usually taken with the help of technology. Google search is one such major tool for collecting such data. The data from Google search is aggregated and available as Google Trends, which is an opensource and searchable database. By using the search keyword buy along with the company name, data is extracted from Google Trends for a period. We then test for any possible relationships between the word buy and Nifty50 stocks. NIFTY is a broad-based index representing 50 large and liquid stocks on the National Stock Exchange, India. Nifty50 stocks were chosen for the study because it represents more than 60% of free float market capitalization in the exchange. The adjusted daily closing price for all the 50 stocks of the Nifty50 index was extracted for a period of 5 years starting December 2012 and ending December 2017 from the NIFTY website. This was then averaged into the weekly closing price for analysis. 2. Literature Review Reference [1] examined the influence of information contained in news on the stock market, by sourcing news articles from Lexis Nexis, a searchable database of digitized print media. Results showed that the content in the recording revealed a substantial connection with the stock yields. Reference [2] applied the technique of collecting data from social media platforms like Facebook, Twitter and Google. The study performed a sequence analysis and transmission mechanism of how social media is a good reflector towards the capital markets. It also put forth the relevance of reach of data from social media as a tool in monitoring the society s behaviour. Reference [3] tested for the presence of any relationship between online search intensity and stock trading behaviour. The test was carried out in Japanese stock markets for 189 companies. All of them were Japanese stocks, which were searched from 2008 to 2011 in google.com. The conclusion drawn was that the relationship between search intensity and stock price is positive but weak. But the relationship between search intensity and trading volume was positive and strong. Reference [4] examined if stock returns could be forecasted by using data from Google Trends. Previous studies stated that the returns were high for the first 1 to 2 weeks with Google search volumes, followed by a subsequent price reversal. The dataset used for this study pertains to 2008 to The key conclusion of the paper was that higher Google search volumes for a particular stock yielded negative returns. The paper also attempted a trading strategy by selling the stocks with higher search volumes and buying those with less frequent Google search volumes. Reference [5] tested the relationship between Twitter feeds and financial markets by collecting data on the volume of data churned out and performing a 941

2 sentiment analysis for 30 companies from the Dow Jones Industrial Average (DJIA) index for 15 months. The study found that Twitter sentiment had a positive relationship with the stock returns when the volume is at peak. It also identified that Granger Causality and Pearson Correlation were lower in the time-series for the given sample time frame. Reference [6] presented an overview of the mechanisms operating between social media tools and the capital markets explaining the rationale using theories of behavioural finance. It also contemplated on less rational factors (investors sentiment/public mood) as a powerful indicator of asset pricing and capital market volatility. Reference [7] identified the emergence of cyclical patterns, as a result of linearity. The NIFTY data was used from 1994 to 2013.The study found a cyclical pattern where Indian stock markets were observed to have shifted between periods of efficiency and inefficiency, and settling to becoming efficient from Reference [8] analyzed changes in the volume of search terms in Google. The findings were patterns that can be interpreted as early warning signs of stock market moves. Data from January 2004 and February 2011 was used. It was found that adding huge behavioural data sets (financial trading and search query volumes) can bring up new understandings into the different phases of large-scale collective decision making Reference [9] used Granger Causality to comprehend the association between stock returns and selected macroeconomic variables like oil prices and exchange rates from India and US for the years 2000 to It was found that the GDP growth rate of India and USA are significant predictors of NIFTY returns. Similarly, reference [10] used Granger Causality to study the relationship between exchange rate returns and stock indices returns. The paper established that the indices (commodity, IT, MNC and energy) significantly caused exchange rate returns. 3. Methodology The study has taken weekly closing data of Nifty 50 stocks for the period starting December 2012 to December 2017.Weekly index returns of Nifty50 individual stocks were then calculated by the following log normal formula =ln ( / ) Where, = Nifty 50 index closing price of present week = Nifty 50 index closing price of previous week This study has taken the Google Trend data using the keywords buy<company name> weekly from December 2012 to December The Granger Causality test, a statistical technique used to predict the relationship between variables using time series data. One prerequisite for using Granger Causality test is that the data has to be stationary. So using unit root test, weekly returns of Nifty50 and Google Trend data are tested for statistical stationarity. The unit root test is directed to find whether the dataset is stationary or not. Statistical stationarity implies that the properties of statistics (mean, variance and autocorrelation) are perpetual for the given particular period of time. Equation (1) and (2) explains the causal relationship between Buy and NReturns =!"#!$ + ' & ' ' +( (1)!"#!$ = ' & '!"#!$ ' +( (2) + Where, BUY= buy<company name> search term used in Google trends NRETURNS = Nifty 50 index weekly returns ),& ',+,, ' = coefficients of the model (i.e., the contributions of each laggedobservation) (,( = residuals (prediction errors) for each time series. 4. Hypothesis The subsequent hypothesis was formulated which is tested using Granger Causality test. H01a: GT does not Granger cause RET H01b: RET does not Granger cause GT Where, GT = buy<company name> search term used in Google trends RET = Nifty50 individual stocks weekly returns Figure Bosch HCL Technologies 942

3 Figure Reliance Industries Figure1 shows the Google Trends chart for Bosch and HCL Technologies for the period of 2012 to Figure 2 shows the Google Trends chart for Reliance Industries and Hindustan Unilever for the period of 2012 to The dataset was downloaded from the Google Trends in CSV format and then it was used for the analysis. Google Trends is a representation of search interest relative to the highest value for the specified time and period. A value of 100 indicate the maximum popularity of the term. A value of 50 indicates that the term is semi-popular. Similarly, a score of zero implies the term was searched by less than 1% of the population. 5. Results and Discussion The Granger causality results for lag 2 is shown in the Tables below Date: 01/09/18 Time: 20:39 Sample: Table 1. Table 2. Hindustan Unilever WGT does not Granger Cause WRET WRET does not Granger Cause WGT Date: 01/11/18 Time: 22:32 Sample: LGT does not Granger Cause LRET LRET does not Granger Cause LGT Date: 01/27/18 Time: 22:09 Table 3. IOCRET does not Granger Cause IOCGT IOCGT does not Granger Cause IOCRET Date: 01/27/18 Time: 22:11 Table 4. MSRET does not Granger Cause MSGT MSGT does not Granger Cause MSRET Date: 01/11/18 Time: 22:34 Sample: Table 5. ITGT does not Granger Cause ITRET ITRET does not Granger Cause ITGT Date: 01/27/18 Time: 22:13 Table 6. BOSCHRET does not Granger Cause BOSCHGT BOSCHGT does not Granger Cause BOSCHRET

4 Date: 01/27/18 Time: 22:15 Table 7. HCLRET does not Granger Cause HCLTECHGT HCLTECHGT does not Granger Cause HCLRET Date: 01/27/18 Time: 22:21 Table 8. RILRET does not Granger Cause RILGT RILGT does not Granger Cause RILRET Date: 01/27/18 Time: 22:23 Table 9. HULRET does not Granger Cause HULGT HULGT does not Granger Cause HULRET WRET:Wipro Returns WGT:Wipro Google Trends ITRET:ITC Returns ITGT:ITC Google Trends LRET: Lupin Returns LGT: Lupin Google Trends IOCRET: Indian Oil Corporation (IOC) Returns IOCGT: Indian Oil Corporation Google Trends MSRET: Maruti Suzuki Returns MSGT: Maruti Suzuki Google Trends BOSCHRET: BOSCH Returns BOSCHGT: BOSCH Google Trends HCLRET: HCL Returns HCLGT: HCL Google Trends RILRET: Reliance Industries Returns RILGT: Reliance Industries Google Trends HULRET: Hindustan Unilever (HUL) Returns HULGT: Hindustan Unilever Google Trends The same testing protocol was followed for the remaining 41 companies. After examining the association between Google Trends data and the individual stock returns, it was found that the results were mixed in nature. r companies like Wipro, ITC, Lupin, IOC and Maruti Suzuki the stock returns Granger Caused Google search trends (Table 1-5). Bosch and HCL displayed the opposite causality where Google Trends were causing stock returns (Table 6 and 7). Reliance Industries and HUL were observed to have a bidirectional causal relationship for the study (Table 8 and 9). There were 41 companies that did not show any kind of causality between Google trends and their respective stock returns. Table 10 shows the directional relationship using Granger causality for the selected 50 companies. Table 10. Causality Direction No of Companies RETURN Granger Cause GT 5 GT Granger Causes RETURN 2 Bi-DIRECTIONAL 2 No Causality Conclusion The study examined the association between Google Trends data and the Nifty50 individual stock returns. Results showed that for certain stocks such as Wipro, ITC, Lupin, IOCL, Maruti Suzuki, the stock returns had a relationship with Google Trends data,. However, Bosch and HCL Google Trends Granger caused return and for HUL, RIL both Google Trends data and stock returns had a bidirectional relationship for a period from December 2012 to December The Google Trends data pertaining to search word buy<company name> was in use for this study. There is scope for studies using other keywords like sell<company name> and their relationship with individual stock returns. References [1] Clive B. Walker, The direction of media influence: Real-estate news and the stock market, Journal of Behavioral and Experimental Finance 10 (2016) [2] Laurens Bijl, Glenn Kringhaug, Peter Molnár and EirikSandvik, Google searches and stock returns, International Review of Financial Analysis 45 (2016) [3] FumikoTakeda, Takumi Wakao, Google search intensity and its relationship with returns and trading volume of Japanese stocks, Pacific-Basin Finance Journal 27 (2014) 1 18 [4] Hyungyoungchoi and Hal Varian, Predicting the Present with Google Trends, The Economic Record, vol. 88,, June, 2012,

5 [5] Gabriele Ranco, Darko Aleksovski, Guido Caldarelli, Miha Grˇcar and Igor Mozeti, The Effects of Twitter Sentiment on Stock Price Returns [6] Jaroslav Bukovina, Social media big data and capital markets An overview, Journal of Behavioral and Experimental Finance 11 (2016) [7] Gourishankar S Hiremath and JyotiKumari, Stock Returns Predictability And The Adaptive Market Hypothesis In Emerging Markets: Evidence From India, SpringerPlus (2014) 3:428 [8] Tobias Preis; Helen Susannah Moat; H. Eugene Stanley (2013). "Quantifying Trading Behavior in Financial Markets Using Google Trends". Scientific Reports. 3: [9] B. Nikita,P. Balasubramanian, Lakshmi Yermal, Impact of key macroeconomic variables of India and USA on movement of the Indian stock return in case of S&P CNX nifty [10] Prashant Krishnamurthy, P. Balasubramanian, Deepti Mohan, Study on relationship between exchange rate return and various stock indices returns 945

6 946