- Title
- Mining social media as a measure of equity market sentiment
- Creator
- Mahmoudi, Nader
- Relation
- University of Newcastle Research Higher Degree Thesis
- Resource Type
- thesis
- Date
- 2020
- Description
- Research Doctorate - Doctor of Philosophy (PhD)
- Description
- Investor sentiment is an important topic in behavioural finance where endogeneity and spurious correlation are critical issues given unequivocal measures of investor sentiment are not available. Measures of investor sentiment employed in empirical work are estimated usually as market-level proxies that are invariant across individual firms. These measures are not appropriate when empirical issues involve firm-specific issues such a mergers and acquisitions (M&A). Properly constructed measures of sentiment should capture the beliefs of individual investors at the firm-level. For this reason, big data that captures the beliefs of individuals is likely to provide a fruitful source of information in measuring investor sentiment. In particular, social media posts comprise unstructured opinion content that may be highly valuable in the measurement of sentiment. Studies that seek to examine the impact of sentiment in financial markets have been affected by inaccurate sentiment measurement and the use of inappropriate data. Chapter 2 applies state-of-the-art techniques from the domain-general sentiment analysis literature to construct a more accurate decision support system that generates demonstrable improvement in investor sentiment classification performance compared with previous studies. The inclusion of emojis is shown to lead to significant improvements in sentiment classification in traditional algorithms. Moreover, deep neural networks with domain-specific word embeddings outperform the traditional approaches for the classification of investor sentiment. The approach to sentiment classification outlined in this chapter can be applied in future empirical tests that examine the impact of investor sentiment on issues within financial markets. Despite the widespread consensus that the inclusion of emojis improves the efficiency of natural language processing tasks, the existing literature focuses only on the general meanings transferred by emojis without considering the merit of incorporating emojis in specific contexts, such as investor sentiment classification. Chapter 3 reports that a classifier that incorporates domain-specific emoji vectors can improve the performance of investor sentiment classification accuracy. In addition, a time-series measure of investor sentiment developed using this classifier also exhibits additional marginal explanatory power on returns and volatility. Given the importance of domain specific emoji vectors in investor sentiment classification of social media data, this chapter compares two competing approaches to lexicon development and proposes an emoji lexicon that can be used by future researchers. Noise trader models of market behaviour propose that investor sentiment affects market responses to corporate announcements. As beliefs can be cross-sectionally heterogeneous, firm-specific investor sentiment may differ from aggregate levels of investor sentiment. Therefore, previous studies, which focus exclusively on market level investor sentiment measures, are likely to have under-stated the economic magnitude of the role that sentiment plays in corporate announcement returns. Chapter 3 provides the first theoretical model for the relative importance of firm- and market- level investor sentiment in asset pricing. This chapter also demonstrates empirically that firm-level investor sentiment has marginal explanatory power beyond market-level investor sentiment for merger and acquisition announcement returns, turnover and long-run reversals. It is also shown that firm-level investor sentiment dominates when there is high divergence in investor sentiment across firms, whereas the effect of market-level investor sentiment increases during “hot” market periods.
- Subject
- emojis; emoji lexicon; noise traders; return reversals; sentiment classification; stock market; StockTwits; uncertainty; word embeddings; deep neural network; domain-specific; domain-independent; firm-level investor sentiment; heterogeneity; investor; market-level investor sentiment; mergers and acquisition
- Identifier
- http://hdl.handle.net/1959.13/1428875
- Identifier
- uon:38664
- Rights
- Copyright 2020 Nader Mahmoudi
- Language
- eng
- Full Text
- Hits: 1851
- Visitors: 2389
- Downloads: 606
Thumbnail | File | Description | Size | Format | |||
---|---|---|---|---|---|---|---|
View Details Download | ATTACHMENT01 | Thesis | 5 MB | Adobe Acrobat PDF | View Details Download | ||
View Details Download | ATTACHMENT02 | Abstract | 540 KB | Adobe Acrobat PDF | View Details Download |