Browsing by Author "Odendaal, Hanjo"
Now showing 1 - 1 of 1
Results Per Page
Sort Options
- ItemNews, sentiment and the real economy(Stellenbosch : Stellenbosch University, 2020-12) Odendaal, Hanjo; Kirsten, Johann; Reid, Monique B.; Stellenbosch University. Faculty of Economic and Management Sciences. Dept. of Economics.ENGLISH SUMMARY : In this dissertation, text analysis is presented as a complement to traditional survey-based methods used to capture sentiment. This is achieved by firstly constructing media-based sentiment indices from a large variety of news sources for South Africa and presenting these indices as a feasible way to replicate the results of the traditional survey-based consumer confidence index (CCI). The findings of the cointegration and Granger-causality tests support the hypothesis that news-based indices could possibly be used to address shortcomings commonly experienced in the survey-based alternative. The second contribution towards the literature is the evaluation of the adequacy of media-based indices as a predictor of personal consumption. The predictive power of media sentiment indices are evaluated in a Bayesian forecasting horse race alongside the CCI. The conclusion revealed that the inclusion of media-based sentiment indices as predictors in a model can decrease forecasting errors of personal consumption expenditure. The forecasting errors decreased in the cases of both short and long (up to 2 years) forecasting horizons. The results substantiate the theory that news media sentiment contains information on the coincidental and future state of the economy which is not captured in the CCI. The results suggest that media based indices could function as both a complement or alternative to the CCI in consumption forecasting. The final contribution of the thesis showcases the effectiveness of utilizing domain-specific dictionaries to capture sentiment. In the last chapter, domain-specific dictionaries are constructed, in an automated fashion, using Random Forests. These domain-specific indices successfully capture economic sentiment more accurately than the widely used Loughran dictionary. This framework reduces the resources required to extract information from media reports into a sentiment dictionary, while also maintaining a level of transparency. Collectively, the results presented in this dissertation offer some initial support for the use of text analysis in South Africa as an alternative way of capturing softer economic indicators such as economic sentiment.