classic red imitation Hermes dynamic Take a look at it from zroessgs viesoess's blog
Quantifying Wikipedia Usage Patterns Before Stock Market Moves
AbstractFinancial crises result from a catastrophic combination of actions. Vast stock market datasets offer us a window into some of the actions that have led to these crises. Here, we investigate whether data generated through Internet usage contain traces of attempts to gather information before trading decisions were taken. We present evidence in line with the intriguing suggestion that data on changes in how often financially related Wikipedia pages were viewed may have contained early signs of stock market moves. Our results suggest that online data may allow us to gain new insight into early information gathering stages of decision making.IntroductionThe complex behaviour of our society emerges from decisions made by many individuals. In certain combinations, these numerous decisions can lead to sudden catastrophe, as demonstrated during crowd disasters and financial crises. Stock market data provide extremely detailed records of decisions that traders have made, in an area in which disasters have a widespread impact. As a result, these stock market records have generated considerable scientific attention1,2,3,4,5,6,7,8,9,10,11,12,13,14.Human decision making does not, however, consist solely of the final execution of a chosen action, such as a trade recorded at a stock exchange. Instead, within the constraints of available resources, we often begin by gathering information to help us identify what the consequences of possible actions might be15.With Internet provision becoming so widespread, online resources have become the first port of call in many quests for new information. As a rule, providers of such online resources collect extensive data on their usage, adding to a range of new large scale measurements of collective human behaviour16,17,18,19,20,21. In this way, the ubiquity of the Internet in everyday life has not only changed the way in which people collect information to make decisions, but has opened up new avenues for scientists to investigate the early information gathering stages of decision making processes.Previous studies have demonstrated that analysis of search data can provide insight into current or even subsequent behaviour in the real world. For example, changes in the frequency with which users look for certain terms on search engines such as Google and Yahoo! have been correlated with changes in the numbers of reports of flu infections across the USA22, the popularity of films, games and music on their release23, unemployment rates24,25, tourist numbers25, and trading volumes in the US stock markets26,27. A recent study showed that Internet users from countries with a higher per capita gross domestic product (GDP) search for proportionally more information about the future than information about the past, in comparison with Internet users from countries with a lower per capita GDP28.In work most closely related to the study presented here, Preis, Moat and Stanley outline an analysis of historic data which suggests that changes in search volume for financially relevant search terms can be linked to stock market moves29. A further study analysed data from Twitter and considered the emotions of traders, rather than their information gathering processes, suggesting that changes in the calmness of Twitter messages could be linked to changes in stock market prices30.In this study, we investigate whether data on the usage of the popular online encyclopaedia Wikipedia31,32,33,34 can be linked to subsequent decisions made in the stock markets. Specifically, can we find any evidence that changes in the numbers of views or edits to articles relating to companies and other financial topics on Wikipedia may provide insight into the information gathering process of investors?ResultsTo investigate the relationship between changes in large scale information gathering behaviour on Wikipedia and market participants' trading decisions, we consider data on how often pages on the English language Wikipedia have been viewed, and how often pages on the English language Wikipedia have been edited. Wikipedia entries can be both viewed and edited by any Internet user.We calculate two measures of Wikipedia user activity: the average number of page views and the average number of page edits that have taken place for a given Wikipedia page in week t, where we define weeks as ending on a Sunday. All names of Wikipedia pages used and further details on data pre processing are provided in the Supplementary Information. To quantify changes in information gathering behaviour, we choose one measure of Wikipedia user activity n(t), either page view or page edit volume, and calculate the difference between the page view or page edit volume for week t, to the average copy cartier earrings page view or page edit volume for the previous t weeks: n(t, t) = n(t) N(t 1, t) with N(t 1, t) = (n(t 1) + n(t 2) + + n(t t))/t, where t is measured in units of weeks.We begin our comparison of changes in Wikipedia usage to subsequent stock market movements in this historic data by implementing a hypothetical investment strategy that uses data on either Wikipedia page views or Wikipedia page edits to trade on the Dow Jones Industrial Average (DJIA), following the approach introduced by Preis, Moat, and Stanley29. In this hypothetical strategy, we sell the DJIA at the closing price p(t + 1) on the first trading day of week t + 1 if the volume of views or edits has increased in week t such that n(t, t) > 0. We then close the position by buying the DJIA at price p(t + 2) at the end of the first trading day of the following week t + 2. Note that mechanisms exist which make it possible to sell stocks on a financial market without first owning them. If instead the volume of views or edits has decreased or remained the same in week t such that n(t, t) 0, then we buy the DJIA at the closing price p(t + 1) on the first trading day of week t + 1, and sell the DJIA at price p(t + 2) at the end of the first trading day of the coming week t + 2 to close the position.We calculate the cumulative return R of a strategy by taking the natural log of the ratio of the final portfolio value to the initial portfolio Hermes necklace imitation value. If we take a short position selling at the closing price p(t + 1) and buying back at price p(t + 2) then the change in the cumulative return R for a strategy is log(p(t + 1)) log(p(t + 2)). If we take a long position buying at the closing price p(t + 1) and selling at price p(t + 2) then the change in the cumulative return R is log(p(t + 2)) log(p(t + 1)). In this way, buy and sell actions have symmetric impacts on the cumulative return R of a strategy. In addition, we neglect transaction fees, since the maximum number of transactions per year when using this strategy is only 104, allowing one closing and one opening transaction per week. We note that inclusion of transaction fees would of course diminish any profit if this hypothetical strategy were to be used in the real world. However, this assumption does not have consequences for conclusions about the relationship between user activity on Wikipedia and movements in the DJIA.We compare the returns from the Wikipedia data based strategies to the returns from a random strategy. In the random strategy, a decision is made each week to buy or sell the DJIA. The probability that the DJIA will be bought rather than sold is always 50%, and the decision is unaffected by decisions in previous weeks. This random strategy leads to no significant profit or loss. For the statistical comparisons reported in the following sections, we use 10,000 independent realisations of this random strategy for the period between 10th December 2007 and 30th April 2012. We find no evidence that the overall return from these 10,000 realisations is significantly positive or significantly negative (mean return = 0.0002, V = 25012353, p = 0.97, = 0.05, two tailed one sample Wilcoxon signed rank test of symmetry of distribution of returns around 0). We use a non parametric test to check this point, as the distribution of returns deviates significantly from the normal distribution (D = 0.1716, p = 0.05, Kolmogorov Smirnov test). Similarly, the remainder of the analyses of return distributions reported here also use non parametric tests. Throughout the rest of the results, the cumulative returns R of all non random strategies are stated in terms of standard deviations above or below the mean cumulative return of the random strategy.Views and edits of Wikipedia articles about companies listed in the DJIAFigure 1 shows the distributions of returns from two portfolios of 30 hypothetical strategies, trading weekly on the DJIA. These trading strategies are based on changes in how often the 30 Wikipedia pages describing the companies in the DJIA were viewed (blue) and edited (red) during the period December 2007 April 2012, with t = 3 weeks. The distribution of returns from 10,000 independent realisations of a random strategy is also shown (gray). The distribution of returns from 10,000 independent realizations of a random strategy is also shown (gray). Data is displayed using a kernel density estimate and the ggplot2 library36, with a Gaussian kernel and bandwidth calculated using Silverman's rule of thumb37. Whereas we show in the text that random strategies lead to no significant profit or loss, we find that the returns of Wikipedia article view based strategies for this period are significantly higher than the returns of the random strategies (mean R = 0.50; W = 199690, p = 0.005, = 0.05, two tailed two sample Wilcoxon rank sum test, Bonferroni correction applied). There is however no statistically significant difference between the returns from the Wikipedia edit based strategies and the random strategies (mean R = 0.09; W = 140781, p > 0.9, = 0.05, two tailed two sample Wilcoxon rank sum test, Bonferroni correction applied).We find that there are significant differences between these three distributions (2 = 10.21, df = 2, p = 0.006, = 0.05; Kruskal Wallis rank sum test). Our analysis shows that the returns of Wikipedia page view based strategies for this period are significantly higher than the returns of the random strategies (mean R = 0.50; W = 199690, p = 0.005, = 0.05, two tailed two sample Wilcoxon rank sum test, Bonferroni correction applied). There is however no statistically significant difference between the returns from the Wikipedia edit based strategies and the random strategies (mean R = 0.09; W = 140781, p > 0.9, = 0.05, two tailed two sample Wilcoxon rank sum test, Bonferroni correction applied).Views and edits of Wikipedia articles about financial topicsWe investigate whether these results extend to Wikipedia articles on more general financial topics. To address this question, we make use of the fact that Wikipedia contains lists of pages relating to specific topics. Here, we examine view and edit data for 285 pages relating to general economic concepts, as listed in the subsection "General economic concepts" on the English language Wikipedia page "Outline of economics".Figure 2 shows the results of an analysis of the distribution of returns from two portfolios of 285 hypothetical strategies, trading weekly on the DJIA. These strategies are based on changes in how often these 285 financially related Wikipedia pages were viewed (blue) and edited knock off Hermes necklace (red) during the same period, again with t = 3 weeks. As before, we find that there is a significant difference between the returns generated by the random strategies, the Wikipedia view based strategies and the Wikipedia edit based strategies (2 = 307.88, df = 2, p = 0.05; Kruskal Wallis rank sum test). As before, the returns of Wikipedia page view based strategies are significantly higher than the returns of random strategies for this period (mean R = 1.10; W = 2286608, p = 0.05, two tailed two sample Wilcoxon rank sum test, Bonferroni correction applied). Once again however, we find no evidence of a statistically significant difference between the returns from the Wikipedia edit based strategies, and the random strategies (mean R = 0.12; W = 1516626, p = 0.19, = 0.05, two tailed two sample Wilcoxon rank sum test, Bonferroni correction applied). Our analysis shows that the returns of Wikipedia page view based strategies are significantly higher than the returns of random strategies for this period (mean R = 1.10; W = 2286608, p = 0.05, two tailed two sample Wilcoxon rank sum test, Bonferroni correction applied). Once again however, we find no evidence of a statistically significant difference between the returns from the Wikipedia edit based strategies, and the random strategies (mean R = 0.12; W = 1516626, p = 0.19, = 0.05, two tailed two sample Wilcoxon rank sum test, Bonferroni correction applied).The lack of relationship found for the data on Wikipedia edits may simply reflect the substantial difference in the volume of data available for views and for edits, despite the much larger number of pages considered in this second analysis. For example, across the whole period, the Wikipedia articles on financial topics had an average of 1,351,796 views each, but replica Hermes yellow gold necklace only 431 edits. Of these pages, the most viewed page had 14,449,973 views, in comparison to 4832 edits. The least viewed page had 2,033 views, whereas 43 of the 285 pages in question had no edits at all. For the purposes of this study, we therefore do not consider edit data further.Strategy returns in different yearsThe period of time we investigate here includes a particularly large drop in the DJIA in 2008. We therefore investigate what the returns from these trading strategies would have been for each individual year in our study period. Again, we consider the returns of strategies based on changes in views of the 285 financially related Wikipedia pages, again with t = 3 weeks. In Figure 3, the distribution of returns from the trading strategies are shown for each of the four years for which we have full Wikipedia page view data (blue) alongside returns from random strategies for that year (grey). The distribution of returns from the trading strategies, again with t = 3 weeks, are shown for each of the four years for which we have full Wikipedia page view data (blue) alongside returns from random strategies for that year (grey). Whilst returns differ from year to year (mean return for each year in standard deviations of random strategy returns for the given year: 2008, 0.89; 2009, 0.19; 2010, 0.19; 2011, 0.55; 2 = 129.49, df = 3, p W = 2156094, p W = 1584915, p = 0.001; 2010: W = 1585336, p = 0.001; 2011: W = 1915511, p = 0.05; all two tailed two sample Wilcoxon rank sum tests, using comparisons to the random strategy distributions for the given year).We find that returns do differ from year to year (mean return for each year in standard deviations of random strategy returns for the given year: 2008, 0.89; 2009, 0.19; 2010, 0.19; 2011, 0.55; 2 = 129.49, df = 3, p W = 2156094, p W = 1584915, p = 0.001; 2010: W = 1585336, p = 0.001; 2011: W = 1915511, p = 0.05; all two tailed two sample Wilcoxon rank sum tests, using comparisons to the distribution of random strategy returns for the given year).The effect of tWe investigate the effect of changes in t on the returns from the trading strategies. Again, we consider portfolios of trading strategies based on changes in views of the 285 financially related Wikipedia pages. The mean return from trading strategies, expressed in standard deviations of random strategy returns, is shown in Figure S1 (see Supplementary Information) for t = 1 to 10 weeks. We find that the mean return of the strategies does differ significantly for the different values of t we tested (2 = 93.26, df = 9, p t between 1 and 10 weeks (all Ws > 1950000, all psMean return of the DJIA following increases and decreases in Wikipedia viewsTo complement the trading strategy analysis, we carry out a further analysis of weekly DJIA returns following increases and decreases in views of Wikipedia articles on financial topics.For each of the 285 Wikipedia articles on financial topics, we identify all weeks t within our study period in which the volume of page views increased in week t such that n(t, t) > 0, using t = 3. Across this set of weeks, we calculate the mean return of the DJIA during week t + 1, log(p(t + 2)) log(p(t + 1)). Similarly, we calculate the mean return of the DJIA during week t + 1 for the set of weeks in which the volume of page views decreased in week t such that n(t, t)Between these two sets of weeks, we find a significant difference in the mean return of the DJIA during week t + 1 (W = 78012, p = 0.05, two tailed two sample Wilcoxon rank sum test). Following a decrease in views of Wikipedia pages relating to financial topics, we find a mean DJIA weekly return of 0.0027 a return significantly greater than 0 (V = 39592, p = 0.05, two tailed one sample Wilcoxon signed rank test). In contrast, following an increase in views of Wikipedia pages relating to financial topics in week t, we find a mean DJIA weekly return of 0.0021, significantly less than 0 (V = 2222, p = 0.05, two tailed one sample Wilcoxon signed rank test). The results of this analysis are therefore in line with the relationship between changes in views of Wikipedia articles on financial topics and subsequent movements in the DJIA suggested by the trading strategy analysis.
AbstractFinancial crises result from a catastrophic combination of actions. Vast stock market datasets offer us a window into some of the actions that have led to these crises. Here, we investigate whether data generated through Internet usage contain traces of attempts to gather information before trading decisions were taken. We present evidence in line with the intriguing suggestion that data on changes in how often financially related Wikipedia pages were viewed may have contained early signs of stock market moves. Our results suggest that online data may allow us to gain new insight into early information gathering stages of decision making.IntroductionThe complex behaviour of our society emerges from decisions made by many individuals. In certain combinations, these numerous decisions can lead to sudden catastrophe, as demonstrated during crowd disasters and financial crises. Stock market data provide extremely detailed records of decisions that traders have made, in an area in which disasters have a widespread impact. As a result, these stock market records have generated considerable scientific attention1,2,3,4,5,6,7,8,9,10,11,12,13,14.Human decision making does not, however, consist solely of the final execution of a chosen action, such as a trade recorded at a stock exchange. Instead, within the constraints of available resources, we often begin by gathering information to help us identify what the consequences of possible actions might be15.With Internet provision becoming so widespread, online resources have become the first port of call in many quests for new information. As a rule, providers of such online resources collect extensive data on their usage, adding to a range of new large scale measurements of collective human behaviour16,17,18,19,20,21. In this way, the ubiquity of the Internet in everyday life has not only changed the way in which people collect information to make decisions, but has opened up new avenues for scientists to investigate the early information gathering stages of decision making processes.Previous studies have demonstrated that analysis of search data can provide insight into current or even subsequent behaviour in the real world. For example, changes in the frequency with which users look for certain terms on search engines such as Google and Yahoo! have been correlated with changes in the numbers of reports of flu infections across the USA22, the popularity of films, games and music on their release23, unemployment rates24,25, tourist numbers25, and trading volumes in the US stock markets26,27. A recent study showed that Internet users from countries with a higher per capita gross domestic product (GDP) search for proportionally more information about the future than information about the past, in comparison with Internet users from countries with a lower per capita GDP28.In work most closely related to the study presented here, Preis, Moat and Stanley outline an analysis of historic data which suggests that changes in search volume for financially relevant search terms can be linked to stock market moves29. A further study analysed data from Twitter and considered the emotions of traders, rather than their information gathering processes, suggesting that changes in the calmness of Twitter messages could be linked to changes in stock market prices30.In this study, we investigate whether data on the usage of the popular online encyclopaedia Wikipedia31,32,33,34 can be linked to subsequent decisions made in the stock markets. Specifically, can we find any evidence that changes in the numbers of views or edits to articles relating to companies and other financial topics on Wikipedia may provide insight into the information gathering process of investors?ResultsTo investigate the relationship between changes in large scale information gathering behaviour on Wikipedia and market participants' trading decisions, we consider data on how often pages on the English language Wikipedia have been viewed, and how often pages on the English language Wikipedia have been edited. Wikipedia entries can be both viewed and edited by any Internet user.We calculate two measures of Wikipedia user activity: the average number of page views and the average number of page edits that have taken place for a given Wikipedia page in week t, where we define weeks as ending on a Sunday. All names of Wikipedia pages used and further details on data pre processing are provided in the Supplementary Information. To quantify changes in information gathering behaviour, we choose one measure of Wikipedia user activity n(t), either page view or page edit volume, and calculate the difference between the page view or page edit volume for week t, to the average copy cartier earrings page view or page edit volume for the previous t weeks: n(t, t) = n(t) N(t 1, t) with N(t 1, t) = (n(t 1) + n(t 2) + + n(t t))/t, where t is measured in units of weeks.We begin our comparison of changes in Wikipedia usage to subsequent stock market movements in this historic data by implementing a hypothetical investment strategy that uses data on either Wikipedia page views or Wikipedia page edits to trade on the Dow Jones Industrial Average (DJIA), following the approach introduced by Preis, Moat, and Stanley29. In this hypothetical strategy, we sell the DJIA at the closing price p(t + 1) on the first trading day of week t + 1 if the volume of views or edits has increased in week t such that n(t, t) > 0. We then close the position by buying the DJIA at price p(t + 2) at the end of the first trading day of the following week t + 2. Note that mechanisms exist which make it possible to sell stocks on a financial market without first owning them. If instead the volume of views or edits has decreased or remained the same in week t such that n(t, t) 0, then we buy the DJIA at the closing price p(t + 1) on the first trading day of week t + 1, and sell the DJIA at price p(t + 2) at the end of the first trading day of the coming week t + 2 to close the position.We calculate the cumulative return R of a strategy by taking the natural log of the ratio of the final portfolio value to the initial portfolio Hermes necklace imitation value. If we take a short position selling at the closing price p(t + 1) and buying back at price p(t + 2) then the change in the cumulative return R for a strategy is log(p(t + 1)) log(p(t + 2)). If we take a long position buying at the closing price p(t + 1) and selling at price p(t + 2) then the change in the cumulative return R is log(p(t + 2)) log(p(t + 1)). In this way, buy and sell actions have symmetric impacts on the cumulative return R of a strategy. In addition, we neglect transaction fees, since the maximum number of transactions per year when using this strategy is only 104, allowing one closing and one opening transaction per week. We note that inclusion of transaction fees would of course diminish any profit if this hypothetical strategy were to be used in the real world. However, this assumption does not have consequences for conclusions about the relationship between user activity on Wikipedia and movements in the DJIA.We compare the returns from the Wikipedia data based strategies to the returns from a random strategy. In the random strategy, a decision is made each week to buy or sell the DJIA. The probability that the DJIA will be bought rather than sold is always 50%, and the decision is unaffected by decisions in previous weeks. This random strategy leads to no significant profit or loss. For the statistical comparisons reported in the following sections, we use 10,000 independent realisations of this random strategy for the period between 10th December 2007 and 30th April 2012. We find no evidence that the overall return from these 10,000 realisations is significantly positive or significantly negative (mean return = 0.0002, V = 25012353, p = 0.97, = 0.05, two tailed one sample Wilcoxon signed rank test of symmetry of distribution of returns around 0). We use a non parametric test to check this point, as the distribution of returns deviates significantly from the normal distribution (D = 0.1716, p = 0.05, Kolmogorov Smirnov test). Similarly, the remainder of the analyses of return distributions reported here also use non parametric tests. Throughout the rest of the results, the cumulative returns R of all non random strategies are stated in terms of standard deviations above or below the mean cumulative return of the random strategy.Views and edits of Wikipedia articles about companies listed in the DJIAFigure 1 shows the distributions of returns from two portfolios of 30 hypothetical strategies, trading weekly on the DJIA. These trading strategies are based on changes in how often the 30 Wikipedia pages describing the companies in the DJIA were viewed (blue) and edited (red) during the period December 2007 April 2012, with t = 3 weeks. The distribution of returns from 10,000 independent realisations of a random strategy is also shown (gray). The distribution of returns from 10,000 independent realizations of a random strategy is also shown (gray). Data is displayed using a kernel density estimate and the ggplot2 library36, with a Gaussian kernel and bandwidth calculated using Silverman's rule of thumb37. Whereas we show in the text that random strategies lead to no significant profit or loss, we find that the returns of Wikipedia article view based strategies for this period are significantly higher than the returns of the random strategies (mean R = 0.50; W = 199690, p = 0.005, = 0.05, two tailed two sample Wilcoxon rank sum test, Bonferroni correction applied). There is however no statistically significant difference between the returns from the Wikipedia edit based strategies and the random strategies (mean R = 0.09; W = 140781, p > 0.9, = 0.05, two tailed two sample Wilcoxon rank sum test, Bonferroni correction applied).We find that there are significant differences between these three distributions (2 = 10.21, df = 2, p = 0.006, = 0.05; Kruskal Wallis rank sum test). Our analysis shows that the returns of Wikipedia page view based strategies for this period are significantly higher than the returns of the random strategies (mean R = 0.50; W = 199690, p = 0.005, = 0.05, two tailed two sample Wilcoxon rank sum test, Bonferroni correction applied). There is however no statistically significant difference between the returns from the Wikipedia edit based strategies and the random strategies (mean R = 0.09; W = 140781, p > 0.9, = 0.05, two tailed two sample Wilcoxon rank sum test, Bonferroni correction applied).Views and edits of Wikipedia articles about financial topicsWe investigate whether these results extend to Wikipedia articles on more general financial topics. To address this question, we make use of the fact that Wikipedia contains lists of pages relating to specific topics. Here, we examine view and edit data for 285 pages relating to general economic concepts, as listed in the subsection "General economic concepts" on the English language Wikipedia page "Outline of economics".Figure 2 shows the results of an analysis of the distribution of returns from two portfolios of 285 hypothetical strategies, trading weekly on the DJIA. These strategies are based on changes in how often these 285 financially related Wikipedia pages were viewed (blue) and edited knock off Hermes necklace (red) during the same period, again with t = 3 weeks. As before, we find that there is a significant difference between the returns generated by the random strategies, the Wikipedia view based strategies and the Wikipedia edit based strategies (2 = 307.88, df = 2, p = 0.05; Kruskal Wallis rank sum test). As before, the returns of Wikipedia page view based strategies are significantly higher than the returns of random strategies for this period (mean R = 1.10; W = 2286608, p = 0.05, two tailed two sample Wilcoxon rank sum test, Bonferroni correction applied). Once again however, we find no evidence of a statistically significant difference between the returns from the Wikipedia edit based strategies, and the random strategies (mean R = 0.12; W = 1516626, p = 0.19, = 0.05, two tailed two sample Wilcoxon rank sum test, Bonferroni correction applied). Our analysis shows that the returns of Wikipedia page view based strategies are significantly higher than the returns of random strategies for this period (mean R = 1.10; W = 2286608, p = 0.05, two tailed two sample Wilcoxon rank sum test, Bonferroni correction applied). Once again however, we find no evidence of a statistically significant difference between the returns from the Wikipedia edit based strategies, and the random strategies (mean R = 0.12; W = 1516626, p = 0.19, = 0.05, two tailed two sample Wilcoxon rank sum test, Bonferroni correction applied).The lack of relationship found for the data on Wikipedia edits may simply reflect the substantial difference in the volume of data available for views and for edits, despite the much larger number of pages considered in this second analysis. For example, across the whole period, the Wikipedia articles on financial topics had an average of 1,351,796 views each, but replica Hermes yellow gold necklace only 431 edits. Of these pages, the most viewed page had 14,449,973 views, in comparison to 4832 edits. The least viewed page had 2,033 views, whereas 43 of the 285 pages in question had no edits at all. For the purposes of this study, we therefore do not consider edit data further.Strategy returns in different yearsThe period of time we investigate here includes a particularly large drop in the DJIA in 2008. We therefore investigate what the returns from these trading strategies would have been for each individual year in our study period. Again, we consider the returns of strategies based on changes in views of the 285 financially related Wikipedia pages, again with t = 3 weeks. In Figure 3, the distribution of returns from the trading strategies are shown for each of the four years for which we have full Wikipedia page view data (blue) alongside returns from random strategies for that year (grey). The distribution of returns from the trading strategies, again with t = 3 weeks, are shown for each of the four years for which we have full Wikipedia page view data (blue) alongside returns from random strategies for that year (grey). Whilst returns differ from year to year (mean return for each year in standard deviations of random strategy returns for the given year: 2008, 0.89; 2009, 0.19; 2010, 0.19; 2011, 0.55; 2 = 129.49, df = 3, p W = 2156094, p W = 1584915, p = 0.001; 2010: W = 1585336, p = 0.001; 2011: W = 1915511, p = 0.05; all two tailed two sample Wilcoxon rank sum tests, using comparisons to the random strategy distributions for the given year).We find that returns do differ from year to year (mean return for each year in standard deviations of random strategy returns for the given year: 2008, 0.89; 2009, 0.19; 2010, 0.19; 2011, 0.55; 2 = 129.49, df = 3, p W = 2156094, p W = 1584915, p = 0.001; 2010: W = 1585336, p = 0.001; 2011: W = 1915511, p = 0.05; all two tailed two sample Wilcoxon rank sum tests, using comparisons to the distribution of random strategy returns for the given year).The effect of tWe investigate the effect of changes in t on the returns from the trading strategies. Again, we consider portfolios of trading strategies based on changes in views of the 285 financially related Wikipedia pages. The mean return from trading strategies, expressed in standard deviations of random strategy returns, is shown in Figure S1 (see Supplementary Information) for t = 1 to 10 weeks. We find that the mean return of the strategies does differ significantly for the different values of t we tested (2 = 93.26, df = 9, p t between 1 and 10 weeks (all Ws > 1950000, all psMean return of the DJIA following increases and decreases in Wikipedia viewsTo complement the trading strategy analysis, we carry out a further analysis of weekly DJIA returns following increases and decreases in views of Wikipedia articles on financial topics.For each of the 285 Wikipedia articles on financial topics, we identify all weeks t within our study period in which the volume of page views increased in week t such that n(t, t) > 0, using t = 3. Across this set of weeks, we calculate the mean return of the DJIA during week t + 1, log(p(t + 2)) log(p(t + 1)). Similarly, we calculate the mean return of the DJIA during week t + 1 for the set of weeks in which the volume of page views decreased in week t such that n(t, t)Between these two sets of weeks, we find a significant difference in the mean return of the DJIA during week t + 1 (W = 78012, p = 0.05, two tailed two sample Wilcoxon rank sum test). Following a decrease in views of Wikipedia pages relating to financial topics, we find a mean DJIA weekly return of 0.0027 a return significantly greater than 0 (V = 39592, p = 0.05, two tailed one sample Wilcoxon signed rank test). In contrast, following an increase in views of Wikipedia pages relating to financial topics in week t, we find a mean DJIA weekly return of 0.0021, significantly less than 0 (V = 2222, p = 0.05, two tailed one sample Wilcoxon signed rank test). The results of this analysis are therefore in line with the relationship between changes in views of Wikipedia articles on financial topics and subsequent movements in the DJIA suggested by the trading strategy analysis.
The Wall