The last decades have shown an increasing trend in the creation and collection of data, and Big Data is one of the new categories within this field. Big Data is, however, first valuable when it has shown its usability. This paper tested the usability of Big Data in trading strategies by including it in a prediction model. The usability was tested by measuring its ability to outperform a passive trading strategy and a base trading strategy. The Base trading strategy did only predict stocks on VIX data. The Big Data were sentiment data collected and generated from Twitter, News articles and Google Search Volume. The traded asset were Dow Jones Industrial index and its underlying firms. Three approaches were tested with the named data, by creating a 1) Index-specific sentiment, 2) Marketgeneral sentiment and 3) Firm-specific sentiments. A robustness check, consisting of 750 simulations were done for linear regression models and their respective trading strategy. The majority of trading strategies for the different Big Data sources exceeded the return of the Base strategies. The best return was generated by using News volume of the words Bullish and Bearish, which yielded an annual abnormal return of 15% for a Short/Long strategy after accounting for trading-cost. This study suggests that Big Data, in the form of sentiment, can potentially improve the yielded return, for prediction models of stock prices.
|Educations||MSc in Finance and Investments, (Graduate Programme) Final Thesis|
|Number of pages||158|