dc.description.abstract | We attempt to capture the sentiment in CEO speeches held at German companies’ AGMs and to assess whether this sentiment is associated with significant market reactions subsequent to the AGM. For that purpose, we gather the CEO speeches held at German DAX and MDAX companies’ annual shareholder meetings from 2008 to 2016 by manually collecting transcripts from the companies’ internet webpages. Our initial sample consists of 356 CEO speeches by 58 companies. We evaluate further documents, such as company charters, shareholder meeting invitations, and audio or video material from the companies’ webpages, in order to confirm that the CEO speeches are indeed initially held in German. Based on this additional analysis, we exclude 18 speeches resulting in a final sample of 338 speeches. Before we can segment the reports into vectors of word counts, we have to convert the documents, which are typically available in PDF file format, to TXT format. In this process, we also replace typographic ligatures and employ UTF-8 character encoding on all files in order to allow for German-specific characters such as ‘Ä’,’Ü’,’Ö’, or ‘ß’. All characters are transformed into lower case and tokenized afterwards, whereby we define a token as any subsequent order of at least three alphabetic characters. In order to exclude potential spelling errors, we exclude tokens that do not occur in at least one percent of the speeches. After that, we apply a stop-word list on the reports to filter out words that might have important semantic functions, but rarely contribute information. Hereafter, the documents are transformed to word count vectors using the Rapidminer software. In a final step, the CEO speeches’ numbers of negative and positive words are counted with respect to the word lists of the BPW, SENTIWS and LIWC dictionaries. | |