Διδακτορικές διατριβές

Μόνιμο URI για αυτήν τη συλλογήhttps://pyxida.aueb.gr/handle/123456789/34

Περιήγηση

A B C D E F G H I J K L M N O P Q R S T U V W X Y Z

Α Β Γ Δ Ε Ζ Η Θ Ι Κ Λ Μ Ν Ξ Ο Π Ρ Σ Τ Υ Φ Χ Ψ Ω

Τώρα δείχνει 1 - 1 από 1

Textual analysis in finance: the cases of mergers and initial public offerings
(2021) Katsafados, Apostolos G.; Κατσαφάδος, Απόστολος; Athens University of Economics and Business, Department of Accounting and Finance; Επίσκοπος, Αθανάσιος; Ανδρουτσόπουλος, Ίων (Ιωάννης); Δράκος, Κωνσταντίνος; Γεωργούτσος, Δημήτριος; Σπύρου, Σπύρος; Τσεκρέκος, Ανδριανός; Λελεδάκης, Γεώργιος
This thesis is divided into seven chapters. Their common feature is that they all revolve around the use of textual analysis, and by extension its application in the finance sector. The first chapter provides the introduction of this thesis and points out why the focus on textual analysis is important. Next, in the second chapter, a relatively brief but thorough review of the literature is presented to crystalize the bases, the constants, and the trends of the research activity in this area. The reason is that in such a case the position of this thesis in relation to the literature, the contribution to it, as well as the empirical findings can better be understood.The third chapter uses textual analysis to identify merger participants, either bidders or targets, in the U.S. banking sector. Based on Loughran and McDonald’s lists of positive and negative words, we compute the sentiment of the bank’s annual reports (10-Ks). In our empirical analysis, we use logistic regression to gauge the probability of a bank participating in a merger event. First, we show that a greater amount of positive words in a bank’s 10-K is linked with a greater possibility of becoming a bidder. Second, we find that a higher frequency of negative words in a bank’s 10-K is associated with a higher possibility of becoming a target. Our inferences remain robust even if we include various bank-specific control variables in our logistic regressions models. The fourth chapter examines the issue of the previous chapter from a different perspective. Unlike the usage of econometric methodologies to explore the significance of the coefficients under an explanatory framework, here our aim is prediction by using machine learning models, including ideas from deep learning models. More specifically, we endeavor to examine whether there is any predictive ability of textual information from annual reports (10-Ks) when predicting bank mergers. We prove that textual data enhance the predictive accuracy of the models both for bidders and targets. By and large, the combination of both textual features and financial variables as input in the models achieves the highest scores. On the one side, the findings for targets indicate that random forest (RF) is the best among others in terms of out-of-sample accuracy. In that case, we use textual features with both unigrams and bigrams using term frequency-inverse document frequency (TF-IDF) weighting scheme along with financial variables. On the other side, deep learning models perform the highest accuracy score at bidder prediction task. In particular, we use the centroid of word embeddings combined with the financial variables. Notably, our finance-specific word embeddings perform better than the generic ones. Again TF-IDF weighting scheme seems to improve the overall predictive outcome. Our findings show that textual disclosure manages to mitigate the opacity of the banks. The fifth chapter tries to get insights into the predictive power of textual data derived from the prospectuses (S-1 filings) in predicting IPO underpricing. In particular, we use several machine learning models to proceed to our prediction tasks. First of all, our research differentiates from prior literature as it predicts not only if an IPO will be underpriced or not, under a binary classification framework, but also it foresees the magnitude of underpricing. At both of these tasks, we find that textual features can efficiently complement financial variables. In reality, machine learning models that use both textual features and financial variables as inputs achieve greater performance compared to models employing a single type of input. Also, we explore methodological ways with which financial variables can be effectively combined with the numerous textual features. Overall, our findings offer empirical evidence on how textual information is able to reduce the ex-ante valuation uncertainty of IPO firms.The sixth chapter adds to the literature that attempts to explain the IPO underpricing, especially based on the tone of the IPO prospectus. We prove that a higher fraction of uncertain text in S-1 filings as an internal source of uncertainty is related to higher underpricing. However, the main merit of our study is that we focus on the policy uncertainty index as a source of external uncertainty, in addition to the textual sentiment. We surprisingly find that higher policy uncertainty prior to the filing date of S-1 is connected with a lower underpricing. Interestingly, we show that high policy uncertainty influences the firm’s decision to go public. In fact, policy uncertainty is negatively linked to IPO volume. We further document that only firms with good quality continue going public despite the high policy uncertainty, thus experiencing a lower underpricing. The seventh chapter provides the main inferences of this thesis as well as offers several suggestions for future research.

Περιήγηση

Πλοήγηση Διδακτορικές διατριβές ανά Θέμα "Bank mergers"