There is no individual step or any universally accepted sentiment step. Many steps have been proposed in the literature to quantify sentiment, or at least to come close it as its subjective nature makes it unobservable. Two chief classs of steps can be distinguished: direct measurings and indirect measurings. Direct steps are derived from sentiment polls that straight ask persons how they feel about current or future economic and stock market conditions. The indirect steps assume that certain economic and fiscal variables, such as closed terminal fund price reduction, common fund flows, aggregative trading volume, informations from IPOs or derived functions, convey outlooks non justified by cardinal and hence a step of investor sentiment. Table 1 below briefly lists the advantages and disadvantages of both types of steps.
The inquiry of the impact of sentiment on stock monetary values is connected with the nature of finance, finance is a societal scientific discipline. Once we lift the premise of substantial reason and see that the fiscal agent is human, finance becomes a societal scientific discipline: security monetary values are determined by human behaviour, the latter being capable to psychological prejudices. Lazer et Al. ( 2009 ) indicate that surveies of human interactions are frequently based on informations collected outright andA assertively, this raises the inquiry of the quality of steps as studies implicitly assume that respondents are honest. Lazer et Al. ( 2009 ) and King ( 2011 ) put frontward that technological developments, development of the Internet and societal webs for case, produce highly rich informations ( really big samples, longitudinal analysis ) leting research workers to do important advancement in the geographic expedition and apprehension of major social jobs. This expanded entree to the observation of human behaviour has the possible to help research workers in placing new behaviours and developing new theories.
Table 1: Advantages and disadvantages of assorted sentiment steps
Steady and big clip series
Sentiment steps that do non depend on theories
Variables observed in existent clip
Correctly reflect market public presentation and therefore the strength of the optimism or the pessimism of investors
Offset publication indices
Questionable representativity of the population
Make non take into history the comparative weight of each investor in the responses.
Unable to right grok investor ‘s degrees of optimism and pessimism
Possible divergency between responses and reel behaviours of investors
Name upon theories that are frequently controversial
Measures really endogenous to the market and economic activity, doing it hard to divide sentiment from the cardinal economic
Surveies utilizing really rich databases are besides available in finance, some of which have identified psychological prejudices. A good illustration is the article of Barber and Odean ( 2001, 2008 ) who explain plus monetary values with psychological prejudices: they show, for illustration, that single investors are net purchasers of attention-grabbing stocks, for example, stocks in the intelligence, stocks sing high unnatural trading volume, and stocks with utmost one-day returns. Lazer et Al ( 2009 ) and King ( 2011 ) put frontward the thought that technological developments – development of the Internet and societal webs for case – green goods highly rich informations set ( e.g. big samples or long clip series ) leting research workers to do important advancement in the geographic expedition and apprehension of major social jobs. These databases, nevertheless, are non widespread and their entree is sometimes restricted. Furthermore, these databases are inactive, informations are collected over a specific period and the update – when it occurs – is long and dearly-won.
For this article, we use a big, unrestricted, on a regular basis updated ( hebdomadal ) and free to entree database. Specifically, the purpose of this paper is to build a new step of investor sentiment based on the corporate intelligence of 100s of 1000s of persons. This attack overcomes the jobs associated with the traditional steps of sentiment presented in Table 1. The database we selected, Google Trends, is put frontward by Google. Google Trends informations provides a quantitative step of the text questions initiated by indviduals utilizing Google hunt engine. In a recent survey, Dzielinski ( 2011 ) shows that Google hunt engine histories for 70 per centum of the entire traffic affecting research. In France, over the period 2004-2010, Google has continued to increase its market portion. AT Internet reported that in January 2004, Google accounted for 68.2 % of the entire French cyberspace traffic, 82.6 % in January 2008 and 91.5 % in December 2010, taking us to reason that in the instance of France the represenativeness of the database is more than satisfactory.
Such a tool has attracted the attending of research workers both in experimental scientific discipline and in societal scientific disciplines. In natural scientific disciplines for case, entree to internet users ‘ questions has been instrumental in developing a public wellness tool. Ginsberg et Al. ( 2009 ) monitor the development of an epidemic of grippe viruses utilizing cyberspace users ‘ questions performed in the United States through Google hunt engine. The advantage of this attack is that it provides faster consequences than the traditional tools of influenza surveillance: the steps are available within a twenty-four hours against one to two hebdomads for the old 1. This shorter clip frame has the possible to let public governments to move faster than earlier.
This type of method is besides widely used in societal scientific discipline. In political scientific discipline, Ripberger ( 2011 ) inquiries the value of utilizing Google Trends to place public attending. He compares the traditional step of political scientific discipline of development of media coverage of a topic with the figure of questions related to the same topic on the Internet. His consequences show that both steps are strongly related. Scheitle ( 2011 ) reaches a similar decision.
In the field of economic sciences, Goel et Al ( 2010 ) note that the internet hunt volume aid foretell consumer behaviour by foretelling the gross revenues of the films box office, gross revenues of video games or the ranking of the vocals on the charts. They conclude that the volume of questions on the cyberspace is a usher of the close hereafter. Goel et Al ( 2010 ) besides indicate that the utility of this index does non lie in its high quality over other indexs, but is related to its about immediate handiness and its easy handiness. McLaren ( 2011 ) efforts to place how the internet hunt volume can be used as indexs of economic activity ( labour market and lodging market ) . He concludes by saying that “ the Bank [ of England ] will go on to supervise these informations as portion of the scope of different indexs it considers in organizing its position about the mentality for the economic system of the United Kingdom. As farther developments are made in this country [ … ] these informations are likely to go an progressively utile beginning of information about economic behaviour. “
In the fiscal field, Da et Al. ( 2011a ) use the the hunt volume on Google to measure the grade of attending paid by investors for certain securities. Specifically, they test and validate the hypothesis of Barber and Odean ( 2008 ) that when investors are net purchasers of some securities, they invite attending on these securities and exercise an upward force per unit area on their monetary values. In add-on, they stress the involvement of internet hunt volume: the volume of petitions is a revealed step of attending. Investors seek information about a security, because they are interested in that security. One can therefore reason that if an investor seeks a term on the cyberspace, it is because he is paying attending to that security. If the internet hunt volume carries a negative intension, investors are pessimistic. Finally, Da et Al. ( 2011a ) show that the internet hunt volume captures the position of single investors, the alleged unworldly investors. These surveies lead to the thought that the building of a step of investor sentiment utilizing the internet hunt volume is far from meaningless.
Tetlock ( 2007 ) investigates the nexus between media content and stock returns. To this terminal, he focuses on the influence of a popular Wall Street Journal column. By carry oning a textual analysis between 1984 and 1998, utilizing a chief constituent analysis, he derives a step of sentiment of the media. The texts are coded utilizing the psychosocial General Inquirer Harvard IV-4 lexicon and the first constituent is correlated with the different classs of the lexicon. This analysis shows that the first constituent reflects the pessimism of the media. Empirical trials show that this step of media pessimism is consistent with the formal attacks of DeLong et Al. ( 1990 ) in that it predicts impermanent lessenings in profitableness followed by average reversion. Tetlock ( 2007 ) ‘s analysis is consistent with established determination in the field of psychological science back uping the thought that “ bad is stronger than good ” ; people react more strongly to a negative phenomenon than to a positive 1. Simonton and Baumeister ( 2005 ) indicate that this type of human reaction demands to be qualified as one of the cardinal, basic, cosmopolitan and general truths in psychological science. Finally, Garcia ( 2012 ) states that investors use different determination regulations in roar than in flop, while being particularly sensitive to intelligence during downswings.
Ripberger ( 2011 ) shows that a step of public sentiment from the media and a step utilizing hunt volume on Google converge toward one another. Therefore, its easiness of entree, the simpleness of its execution, the frequence of its updates and its cheap nature suggest that, based on the internet hunt volume for words with negative intensions pessimism. This hypothesis is strengthened by the reasoning comments of Da et Al. ( 2011a ) : “ aˆ¦ hunt volume is an nonsubjective manner to uncover and quantify the involvements of investors and hence should hold many other possible applications in finance “ , Google Trends can be a valid step of investor.
Finally, Baker and Wurgler ( 2006, 2007 ) argue that the houses most prone to investor sentiment, little market capitalisation, are besides those complicated to measure and those for which arbitrations prove hard. Furthermore, these writers show that when sentiment is high – investors are optimistic – these houses profitableness are by and large smaller in the beginning of the period than towards the terminal. These elements lead to the undermentioned hypothesis:
H1: A step of investor pessimism based on the volume of Internet hunts of words with negative intensions should be positively correlated with little capitalisation hereafter returns.
Methodology and empirical findings
The attack developed in this paper is similar to that presented by Da et Al. ( 2011b ) , although it besides has a figure of important differences. These writers propose a new investor sentiment utilizing the hunt volume on Google. Our step emerges from Google Trends ( GTNS: Google Trends Negative Sentiment ) .
To build a step of pessimism from the internet hunt volume, we start by placing the footings searched by persons utilizing the classs defined by the General Inquirer. Since our end is to proxy investor ‘s pessimism, we consider the lists of words in the classs “ economic system ” and “ negative ” . The intersection of these two lists leads to a “ starting list ” of 63 economic and negative words.
Several grounds justify the pick of the class “ Economy ” . On the one manus, direct steps of sentiment ask persons about their perceptual experience of economic conditions. On the other manus, Dzielinski ( 2011 ) constructs a step of uncertainness about the province of the economic system utilizing the hunt volume for the word “ economic system ” . The writer paperss that A«A economyA A» captures the results of fiscal uncertainness and has a important relationship with the stock returns. As for the pick of the class “ negative ” , it merely follows from the fact that we construct a step of investor ‘s pessimism.
The proposed step implicitly relies on the fact that people collect information on the Internet utilizing hunt engines. In a laboratory experiment, Holscher and Strube ( 2000 ) effort to understand how persons proceed to research economic constructs utilizing the Internet. Their consequences indicate that if persons are non at the same time experts in economic sciences and the Internet, they overpoweringly use a hunt engine. Jansen and Spink ( 2006 ) carry on a meta-analysis of articles from a really big database of questions. Their consequences show that one tierce of questions on the cyberspace usage merely one term and that the operators such as and/or are seldom used. These surveies suggest that a step of investor sentiment from a question of a specific term with a hunt engine is consistent with the ascertained behaviour of Internet users.
The “ starting list ” of 63 was reduced to a list of eight footings. All footings have been introduced in Google Trends by restricting the geographic range to France in order to mensurate the sentiment of Gallic families. First, we eliminated the words non widely hunt that do non hold a history of more than 96 back-to-back hebdomads. We so excluded the words that do non hold for cyberspace users a certified economic. We operate as described because our end is to develop an single investor sentiment step. Consequently, merely words with a familiar general economic significance instead than specific meaning are selected. To make this, we searched every word with Google Insights For Search. This hunt engine allows to sort internet hunts by class and to cognize the most frequent hunts associated with the term analyzed. When the most frequent hunts attest to the fact that Gallic cyberspace users do non give to the word analyzed a regular economic definition, we remove the word from our list. Ultimately, this work leads to the list of words presented in Table 2 below.
Table 2: List of words representative of the feeling of Gallic families
Economic words with a negative significance
Two elements distinguish our attack from that of Da et Al. ( 2011b ) . First, we do non spread out the initial list of words by the add-on of qualifiers, i.e. we do non seek for rising prices and rising prices rate or nucleus rising prices. The add-on of qualifiers, i.e. words that restricts the significance of another word such as nucleus rising prices vs rising prices, could bring forth extremely correlative historical informations which automatically increase the explanatory power of the footings in the Principal Component Analysis we performed to pull out the step of sentiment. Second, we use the volumes of research on Google Trends instead than the step of Google Insights For Search used by Da et Al. ( 2011b ) . This may look fiddling, but is really rather of import as the steps differ depending on the hunt engine. Whatever the step, the hunt volumes are normalized in the same manner. However, the grading differs depending on the hunt engine: Google Insights For Search normalized values by the highest point, while Google Trends uses the mean traffic associated with the term sought ( fixed graduated table: the default is January 2004, i.e. the get downing day of the month of informations handiness ; comparative graduated table: the norm is calculated over the full period for which information is requested. ) . Google trends attack is therefore superior as it avoids the hindsight prejudice.
The development of the GTNS index
Since each word is likely to include a sentiment constituent every bit good as an independent idiosyncratic constituent, we apply a chief constituent analysis on the eight words retained to insulate the component sentiment. The chief aim is to sum up, every bit expeditiously as possible, into a individual index the common information contained in the set of words. Our step of sentiment represents the first chief constituent based on the correlativity matrix of the eight selected footings. Using hebdomadal informations on the hunt volumes between January 2004 and December 2010, we obtain the following GTNS index
The chief constituent analysis reveals three dimensions that explain about 62 % of the entire discrepancy of the common factor. The first constituent summarizes the most important fluctuation, entirely it accounts for approximately 36 % of the discrepancy. The footings “ bankruptcy ” , “ shortage ” , “ recession ” and “ crisis ” acquire a comparatively big weight in the equation. These footings are therefore closely related to investor sentiment. Furthermore, the coefficients of the different words all display the expected positive marks. Indeed, the internet hunt volume associated with a word is reciprocally related to persons ‘ province of head.
Figure 1 below illustrates the development of the GTNS for the period January 04, 2004 to December 26, 2010. The index increases important during the 2008 crisis, a good mark that our ( beasrish ) index right reflects investors ‘ pessimistic temper during the crisis. Our GTNS index reached its highest degree during the hebdomad of September 28, 2008. This coincides with the stock market clang frequently tied to the subprime crisis.
Nous pouvons remarquer un garrison alignement entre notre indicatuer de sentiment GTNS et la volatilite implicite qui mesure lupus erythematosus emphasis et l’inquietude diethylstilbestrols intervenants sur le marche . Le coefficient de correlation s’eleve a 62 % ( significatif a 1 % ) .
L’indice de confiance est une mesure directe du sentiment largement utilisee dans la litterature. L’indice de confiance diethylstilbestrols menages est etabli a travers une enquete mensuelle qui rend compte diethylstilbestrols perceptual experiences et diethylstilbestrols expectancies diethylstilbestrols menages concernant tant leur state of affairs economique et financiere que leur propension a depenser et leur sentiment Sur la state of affairs economique globale.
Nous Avons donc calcule une moyenne mensuelle de notre indicateur GTNS. Nous pouvons egalement remarquer que notre indicateur GTNS presente une garrison alignement avec IC/ Le coefficient de correlation s’eleve a 56 % environ ( significatif a 1 % ) .
Nous Avons egalement regresse l’indicateur de sentiment diethylstilbestrols menages ICM Sur notre indicateur de sentiment GTNS decale d’un mois. Nous trouvons que notre indicateur GTNS prevoit avec fiabilite l’indicateur de sentiment ICM ( t-stat =- 6.84 ) . Une augmentation de GTNS predit une baisse significative de ICM le mois prochain. Les trials de causalite de Granger confirment Ce invariable, le sens de causalite va de GTNS a l’indice de confiance ICM. Ceci confirme que les donnees Google sont en avance par resonance aux donnees d’enquete ( au mois d’un mois ) .
Au concluding, l’indicateur GNTS semble s’aligner fortement avec les indicateurs traditionnels de sentiment utilises dans la litterature. De plus, notre indicateur semble reproduire de maniere assez fidele les clefts boursiers de notre periode d’etude. Tous Ces resultats sont interessants pour la suite de notre parturiency auto ils laissent presager que notre indicateur composite capte lupus erythematosuss fluctuations appropriees du facteur sentiment Sur notre periode d’etude.
Methodology and empirical consequences
Harmonizing to the theory of investor sentiment, sentiment is an first-class market clocking index. Advocates of behavior finance believe that noisetraders tend to underestimate stocks when they are pessimistic and overestimate them when they are optimistic. Since monetary values finally return to their cardinal value, we expect a negative correlativity between sentiment and future returns. In other words, the best clip to purchase stocks is when investors are pessimist, while it is more advantageous to sell them when they are optimistic
To analyze the interaction between investor sentiment and stock returns, the usage of a theoretical account vector autoregression ( VAR ) is appropriate ( Brown and Cliff ( 2004 ) ) . VAR patterning assumes that each variable is modeled as an endogenous variable depending on its ain slowdown and those of all other endogenous variables included in the system. Therefore, capturing the dynamic relationships that drive the system between the endogenous variables. The VAR theoretical account non merely to foreground the dynamic interactions between the variables, but besides measure the velocity, range and continuance of the impact of sentiment on returns via the impulse response maps. In a formal manner, the VAR theoretical account is as follows
Where Yt is the column vector incorporating the three endogenous variables: GTNS, Rsmall and Rlarge. GTNS is the sentiment index constructed with informations from Google Trends. Rsmall and Rlarge are the returns of little and big market capitalisations severally. is the vector of changeless footings. the matrix of coefficients to be estimated. provides information on causalities between variables in the theoretical account. is the vector of remainders, besides called daze vestor or pulsations vector. is the optimum figure of slowdowns determined by minimising the Akaike information standard ( = 2 in our instance ) . Brown and Cliff 04 province: “ Model choice prosodies such as AIC and BIC suggest P = 2 in the monthly informations and P = 4 in the hebdomadal information. Likelihood ratio trials for the theoretical account order indicate several extra slowdowns are needed. In the involvement of parsimoniousness, we stick to the smaller theoretical account orders. ” Do we hold a good justification for choosing p=2
Table 2 is ne’er introduced nor discussed. I suggest to add the followingA : Table 2 shows descriptive statistics for the variables GTNS, VIX, Rsmall and Rlarge. In peculiar, the tabular array shows the mean, average and standard divergence ( I do non believe we need column 8 may be the last column: figure of points? ? ? ? Number of observations? ? ? ) The tabular array besides show the consequences of ou rating of the void hypothesis that each variable contains a unit root against the alternate hypothesis that the series are stationary. The nothing is rejected at 5 per centum by the ADF and PP statistics. Results therefore suggest that we deal with stationary time-series. ( Did we test for common unit root as good? ( Levin, Lin, and Chu, 2002 ) )
Table 2A : Descriptive Statisticss
This tabular array presents the descriptive statistics for our hebdomadal observations over the period January 2004-December 2010. GTNS is the sentiment index constructed with informations from Google tendencies. VIX is the implied volatility on the CAC 40 index. Rsmall and Rlarge represent the returns of little and big market capitalisations. ADF and PP are unit root trials of Augmented Dickey-Fuller and Phillips-Perron estimated with a changeless and two slowdowns.
Nombre de points
1 % a†’ -3,448
5 % a†’ -2,869
10 % a†’ -2,570
The consequences of the VAR theoretical account are presented in Table 3. We find that the sentiment index is hardly impacted by stock returns. Indeed, merely the relationship between the big capitalisation one month lagged returns and the sentiment index is important at 10 % . An addition of big capitalisation returns is followed by an addition of the GTNS in subsequent periods. It therefore seems that Gallic investors are “ negative feedback bargainers, ” that is to state, investors who purchase securities after monetary values diminution and sell them after a monetary values rise. This consequence disagrees with the profile of noisemakers presented in DeLong et Al ( 1990 ) .
Table 3: Appraisal consequences of the VAR theoretical account
This tabular array presents the appraisal consequences of the VAR theoretical account over the period January 2004-December 2010. The figure of slowdowns is determined by minimising the Akaike information standards ( p = 2 ) . The GTNS sentiment index is constructed through the informations from Google tendencies. Rpetites and Rlarges represent the returns of little and big market capitalisations. The statistics tabulated in parentheses are t-statistics. *** , ** , * Denote the grades of significance at the 1 % , 5 % and 10 %
GTNS ( -1 )
( 17.926 )
( -5.799 ) ***
( -3.915 ) ***
GTNS ( -2 )
( -1.256 )
( 4.624 ) ***
( 3.422 ) ***
Rpetites ( -1 )
( 0.887 )
( -2.231 )
( 1.307 )
Rpetites ( -2 )
( 0.019 )
( 1.717 )
( 1.340 )
Rlarges ( -1 )
( 1.779 ) *
( -0.184 )
( 2.651 )
Rlarges ( -2 )
( -1.191 )
( -0.118 )
( 2.226 )
( -0.663 )
( 0.755 )
( 3.472 )
As expected, the public presentation of little market capitalisation is significantly affected by investor sentiment. Results show a important negative relationship between sentiment index lagged one hebdomad and the returns of little capitalisation. The phenomenon is reversed when the index is delayed by two hebdomads. We record a important positive relationship between the sentiment index delayed by two hebdomads and samll market capitalisations. The same phenomenon, but of lesser magnitude, is observed for the larger market capitalisations. It seems that little market capitalisations are more vulnerable to alterations in investor sentiment than larger 1. This consequence validates the cardinal hypothesis of our survey
A CONTRASTER AVEC CLIFF ET BROWN ( 2004 )
COMPARAISON RESULTATS AVEC DiezlinskiA ; TwitterA ; Facebook avec reviews des 2 derniers ( mesures – revelees ) et Cliff et Brown par exemple
CORRELATIONSA ? ? ? ? RESULTATS AVEC modele VAR Sur fluctuation indicateur de sentiment ET aussi AVEC VARIABLE DE DEPARTA : -GTS auto mesure sentiment negatif
Suggestion to be discussed ____ Consequences are different from those reported by Brown and Cliff ( 2004 ) . Brown and Cliff ( 2004 ) papers that market returns trigger future alterations in sentiment but that sentiment does non look to do subsequent market returns. They besides show that institutional sentiment appears positively related to subsequent big stock returns. Brown and Cliff ( 2004 ) findings are slightly antagonistic intuitive. Indeed, as mentioned by the writers, it is more likely that single investors would be most easy influenced by sentiment whose impact should be observe in little stocks.
Of Granger causality trials were conducted to clear up the way of causality between investor sentiment and stock returns. Consequences in Table 4 show that investor sentiment influences significantly, at the 1 % , both the public presentation of little and big market capitalisations. Furthermore, consequences indicate market returns predict sentiment. All critical chances are below 0.01, which suggests the being of a feedback consequence between stock returns and investor sentiment. However, this analysis does non state us anything about the “ quantification ” of these impacts. This will be studied by analyzing the impulse response maps. I prefer___ To quantify these relationships, we examine the impulse response maps. ___ Or even better for the flow of the document start instantly with the undermentioned paragraphA : The resultsaˆ¦
Table 4: Trial consequences for Granger causality between returns and investor sentiment
This tabular array presents the trial consequences of Granger causality between returns and investor sentiment. The statistics tabulated are F-statistics followed by their P values. The figure of slowdowns in the VAR theoretical account is determined by minimising the Akaike information standards ( p = 2 ) . Test 1: H0: Sentiment does non Granger predict returns Test 2: H0: Tax returns do non Granger predict sentiment
Petites capitalizations boursieres
Larges capitalizations boursieres
Les resultats de l’estimation du modele VAR ne fournissent pas d’indication quant aux proprietes dynamiques du systeme. Comme le fait remarquer Sims ( 1980 ) , les systemes autoregressifs ne peuvent pas etre interpretes d’une maniere succincte en se focalisant Sur les coefficients estimes du modele. Comme les coefficients individuels estimes dans lupus erythematosuss modeles VAR sont souvent difficiles a interpreter, les praticiens de cette technique recommandent l’etude de l’analyse diethylstilbestrols chocs aleatoires sur lupus erythematosuss variables via les fonctions de reponses impulsionnelles. A cet egard, nous completons notre etude par l’examen de Ces fonctions. Ces dernieres permettent de prevoir le profil temporel diethylstilbestrols reactions diethylstilbestrols variables endogenes a chacune diethylstilbestrols inventions ( forty-nine s’agit d’analyser les reponses a la fluctuation de l’ecart type d’une invention ) . Dans notre Ca, les fonctions de reponses impulsionnelles caracterisent deux types de reactionA : La reaction du sentiment de l’investisseur face a des chocs imprevus apportes aux rentabilites et la reaction diethylstilbestrols rentabilites a diethylstilbestrols chocs aleatoires sur l’indicateur de sentiment. Pour chaque type de reaction, Ces fonctions apparaissent dans lupus erythematosuss graphiques de la figure 2.
Consequences of the VAR theoretical account do non supply counsel as to the dynamic belongingss of the system. As noted by Sims ( 1980 ) , autoregressive systems can non be interpreted briefly by concentrating on the estimated coefficients of the theoretical account. As single coefficients estimated in the VAR theoretical accounts are frequently hard to construe, users ( whoA ? ? Beginning is needed ) of this technique urge to analyse the impact of random dazes on the variables via impulse response maps. These maps allow to foretell the temporal profile of responses of endogenous variables to inventions ( this is to analyse the variables responses to the fluctuation of the standard divergence of an invention, ) . In our instance, the impulse response maps characterize two types of reaction: the reaction of investor sentiment to unexpected dazes in returns and the reaction of returns to random dazes of the sentiment index. These impulse response fonctions are depicted in Figure 2
Figure 2: Impulse Response Function
The graphs represent the impulse response maps. The dotted lines represent the upper and lower sets of the assurance interval associated with the impulse response maps obtained utilizing Monte Carlo simulation. The responses are statistically important at the 5 % degree when the upper set and lower set have the same mark. The impulse response maps characterize two types of reaction: the reaction of investor sentiment to unexpected dazes in returns and the reaction of returns to random dazes of the sentiment index ( GTNS )
Reponse de la rentabilite des petites capitalizations boursieres suite a un choc Sur le sentiment
Reponse de la rentabilite diethylstilbestrols grandes capitalizations boursieres suite a un choc Sur le sentiment
Reponse du sentiment de l’investisseur suite a un choc Sur La rentabilite des petites capitalizations boursieres
Reponse du sentiment de l’investisseur suite a un choc Sur La rentabilite diethylstilbestrols grandes capitalizations boursieres
Chemical reaction of investor sentiment to unexpected dazes in returns of little market capitalization/large ____ Reaction of little market capitalization/large to random dazes of the sentiment index
Overall, we find that dazes are died rapidly. The returns revert to their long-term equilibrium within 5 to 7 hebdomads. We besides find that the responses of returns to floor of sentiment are larger for little market capitalisation. Examination of the impulse response maps of these stocks indicates that a positive daze on our bearish GTNS sentiment index lead to a diminution in returns in the first period. ( Returns ) The consequence remains negative in the 2nd period but becomes significantly positive in the 3rd. These consequences reflect the rectification happening for little market capitalisations. Indeed, remember that a high degree of our bearish GTNS sentiment index predicts that monetary values will return to their cardinal degrees. Furthermore, the returns of companies with big market capitalisations show a weaker dependance on structural dazes related to the sentiment index ( explicate more ) __ bearish GTNS sentiment index = excessively long.
In add-on, a daze on stock retuns has no immediate consequence on the sentiment index. The consequence is non noticeable until the 2nd period and rapidly slices over clip. We note that merely a daze on stock returns of big capitalisations generates important impact the sentiment index. Examination of the impulse response maps indicates that a positive daze on returns of big capitalisations is followed by a important addition in the GTNS. The impact of these dazes lasts about 4 hebdomads
A ETAYER AVEC COMPARAISON CLIFF BROWN & A ; MESURES TWITTER FACEBOOK
Ariely D. ( 2008 ) , Predictably Irrational, Harper Collins
Baker M. , Wurgler J. ( 2006 ) , A«A Investor Sentiment and the Cross-Section of Stock ReturnsA A» , Journal of Finance, vol. 61 ( 4 ) , pp.1645-1680
Baker M. , Wurgler J. ( 2007 ) , A«A Investor Sentiment in the Stock MarketA A» , Journal of Economic Perspectives, vol. 21 ( 2 ) , pp. 129-152
Odean T. , Barber B. ( 2001 ) , A«A Boys will be Boys: Gender, Overconfidence, and Common Stock InvestmentA A» , Quarterly Journal of Economics, February, vol. 116 ( 1 ) , pp. 261-292
Odean T. , Barber B. ( 2008 ) , A«A All that Glitters: The Effect of Attention and News on the Buying Behavior of Individual and Institutional InvestorsA A» , Review of Financial Studies, vol. 21 ( 2 ) , pp. 785-818
Barberis N. , Shleifer A. , Vishny R. ( 1998 ) , A«A A theoretical account of investor sentimentA A» , Journal of Financial Economics, vol. 49, pp. 307-343
Black F. ( 1986 ) , A«A NoiseA A» , Journal of Finance, vol. 51, pp. 529 – 543
Brown G. W. , Cliff, M. T. ( 2004 ) , A«A Investor Sentiment and the Near-Term Stock MarketA A» , Journal of Empirical Finance, vol. 11 ( 1 ) , pp. 1-27
Campbell J. , Kyle A. ( 1993 ) , A«A Smart money, noise trading and stock monetary value behaviorA A» , Review of Economic Studies, vol. 60, pp. 1 – 34
Cao M. , Wei J. ( 2005 ) , A«A Stock market returns: A note on temperature anomalyA A» , Journal of Banking & A ; Finance, vol. 29 ( 6 ) , pp. 1559-1573
Da Z. , Engelberg J. , Gao P. ( 2001a ) , A«A In Search of Attention ” , Journal of Finance, vol. 66 ( 5 ) , pp. 1466-1491
Da Z. , Engelberg J. , Gao P. ( 2001b ) , A«A The Sum of All Fears: Investor Sentiment and Asset Prices ” , Working Paper, Revise & A ; Resubmit Review Of Financial Studies
Daniel K. , Hirshleifer D. , Subrahmanyam A. ( 1998 ) , A«A Investor psychological science and investor security market under-and overreactionsA A» , Journal of Finance, vol. 53, pp. 1839-1886
Das ( 2011 ) A , News Analytics: Model, Techniques and Metrics, Chapter 2 in The Handbook of News Analytics in Finance, May, John Wiley & A ; Sons, UK
De Long J. B. , Shleifer A. , Summers L. H. , Waldmann R. J. ( 1990 ) , A«A Noise bargainer hazard in fiscal marketsA A» , Journal of Political Economy, vol. 98 ( 4 ) , pp. 703-739
Dzielinski, M. ( 2011 ) , A«A Measuring economic uncertainness and its impact on the stock marketA A» , Finance Research Letters
Edmans A. , Garcia D. , Norli O. ( 2007 ) , A«A Sports Sentiment and Stock ReturnsA A» , Journal of Finance, vol. 62 ( 4 ) , pp. 1967-1998
Garcia D. ( 2012 ) , A«A Sentiment during recessionsA A» , Kenan-Flagler Business School Working Paper
Ginsberg J. , Mohebbi M.H. , Patel R. S. , Brammer, Lynnette, Smolinski, Mark S. , Brilliant, L. ( 2009 ) , A« Detecting grippe epidemics utilizing hunt engine question informations A» , Nature, vol. 457, 19/02/2009
Goel S. , Hofman J.M. , Lahaie S. , Pennock D.M. , Watts D.J. ( 2010 ) , A«A Predicting consumer behaviour with web searchA A» , Proceedings of the National Academy of Sciences of the United States of America, vol. 107, pp.17486-17490
Hirshleifer D. , Shumway T. ( 2003 ) , A«A Good Day Sunshine: Stock Returns and the WeatherA A» , Journal of Finance, vol. 63 ( 3 ) , pp. 1009 – 1032
Holscher, C. , Strube, G. ( 2000 ) , A«A Web hunt behaviour of Internet experts and newbiesA A» , Computer Networks, vol. 33, pp. 337-346
Jansen B. J. , Spink A. ( 2006 ) , A«A How are we seeking the World Wide Web? A comparing of nine hunt engine dealing logsA A» , Information Processing & A ; Management, vol. 42 ( 1 ) , pp. 248-263
Tversky A. , Kahneman D. ( 1974 ) , A«A Judgment under Uncertainty: Heuristics and aˆ?BiasesA A» , Science, vol. 185, pp. 1124-1131
Tversky A. , Kahneman D. ( 1979 ) , A«A Prospect Theory: An Analysis of Decision under RiskA A» , Econometrica, vol. 47, pp. 313-327
Kamstra, Mark J. , Kramer, Lisa A. , Levi, Maurice D. , 2003a. Winter blues: A Sad stock market rhythm. American Economic Review 93 ( 1 ) , 324-343.
King Gary ( 2011 ) , A«A Guaranting the Data-Rich Future of the Social SciencesA A» , Science, vol. 331, nA°6018, pp. 719-21
Lamont O.A. , Thaler R.H. ( 2003 ) , A«A Can the Market Add and Subtract? Mispricing in Tech Stock Carve-outsA A» , Journal of Political Economy, vol. 111 ( 2 ) , pp. 227-268, April.
Lazer D. , Pentland A. , Adamic L. , Aral S. , Barabasi A.-L. , Brewer D. , Christakis N. , Contractor N. , Fowler J. , Gutmann M. , Jebara T. , King G. , Macy M. , Roy D. , Van Alstyne M. ( 2009 ) , A«A Computational Social ScienceA A» , Science, vol. 323, nA°5915, 06/02/2009 ) , pp. 721-723
Ripberger J. T. ( 2011 ) , A«A Capturing Curiosity: Using Internet Search Trends to Measure Public AttentivenessA A» , Policy Studies Journal, pp. 239-259
Scheitle C. P. ( 2011 ) , A«A Google ‘s Penetrations for Search: A Note Measuring the Use of Search Engine Data in Social ResearchA A» , Social Science Quarterly, vol. 92 ( 1 ) , pp. 285-295
Simonton D. K. , Baumeister R. F. ( 2005 ) , A«A Positive psychological science at the summitA A» , Review of General Psychology, vol. 9, pp. 99-102
Tetlock P.C. ( 2007 ) , A«A Giving Content to Investor Sentiment: The Role of Media in the Stock MarketA A» , Journal of Finance, vol. 62, pp. 1139-1168
Figure 1: Temporal development of the sentiment index GTNS