Essays in quantitative economics: Improvements in measurements and macroeconomic analysis
Dissertation
Datum der mündl. Prüfung:2023-12-07
Erschienen:2023-12-22
Betreuer:Prof. Dr. Tatyana Krivobokova
Gutachter:Prof. Dr. Holger Strulik
Gutachter:Prof. Dr. Sebastian Vollmer
Dateien
Name:Thesis_print_version_wo_abstract.pdf
Size:3.69Mb
Format:PDF
Zusammenfassung
Englisch
This thesis highlights opportunities and challenges related to the use of data in empirical economics. The interpretation of data is the primary tool we as economists have at our disposal to calibrate our models and test the hypotheses we derive from them. In academic research, data serves as an intermediary between theory and reality, and thus the growing availability of data offers an abundance of opportunities to answer more and more research questions with more and more precision. However, the increasing availability and complexity of data also entails challenges quantitative economists regularly encounter. Each chapter of this thesis features a different aspect of empirical work in economic research, that can be linked to a specific challenge. Chapter one provides a general introduction to the challenges in modern quantitative economics. Chapter two deals with the curse of dimensionality in the context of economic forecasting and demonstrates how the downsides of more complex data can outweigh the upsides, when using conventional methods not designed to cope with such highly complex data structures. More specifically, factor-based forecasting models, effective in capturing macroeconomic uncertainty have been shown to produce diminished accuracy with increasing input data complexity, i.e. with many additional variables being added. To address this, we propose "blockPCA," an algorithm that identifies variable groups in highly complex data, extracts factors separately from each group, and significantly enhances nowcasting results under high complexity compared to conventional methods. Chapter three discusses challenges related to endogeneity and contributes to the literature on the linkage of increasing automation and increasing market concentration. In this context, we develop a theoretical model in which firms’ markups are endogenous to factor input choices based on technology levels, but are also affected by technology adoption of other domestic and foreign firms. In the empirical analysis, we find that market power, measured as the markup of price over marginal cost, declines on average with higher levels of automation. However, there is substantial heterogeneity, with firms in the highest revenue and markup quintile gaining market power. Moreover, we find that exposure to foreign automation increases competition in the local market. Chapter four addresses challenges related to incomplete data. Many missing entries often make data inapplicable to conventional econometric models. We demonstrate how these missing values can be imputed combining a variety of related data sources, while providing information on the uncertainty of these imputations. From the imputed data we derive time series and panel data on prices and production quantities of industrial robots over time and across countries. The novel price data fills an important gap in the available data landscape on industrial automation and enables directly linking the cost of automation frequently featured in economic models to empirical work. The fifth chapter addresses data biased caused by measurement error in health survey data. Specifically, we investigate the impact of interviewer effects on survey-based physical measures, focusing on blood pressure as a case study. Analyzing three large nationally-representative health surveys in the Global South, the research employs a linear mixed model to quantify the contribution of interviewer effects to the variance of blood pressure measurements. While the overall influence of interviewers on hypertension prevalence at the national level is statistically significant but small, individual extreme interviewers could lead to measurement divergences as high as 12%, particularly affecting estimates at the sub-district level.
Keywords: Data; Complexity; Forecasting; Measurement error; Imputation; Endogeneity; Automation; Market Concentration; Markup; Blood pressure; Hypertension; PCA; Principal Component Analysis; Factor-based model; Nowcasting