Econometric Analysis of Media Communication

This dissertation conducts an econometric analysis of media communication, examining its immediate impact in the short term and its broader effects in the long term. The first study focuses on the immediate, short-term effects of TV ads, analyzing minute-frequency data to estimate consumer reactions, such as increased online searches, within minutes of ad exposure. This analysis confronts challenges like diurnal pattern removal through Kernel smoothing and computationally intensive estimation of ad effect distribution from a large dataset (approximately 525,000 rows). The second part compares different methods for seasonal adjustment and estimates the impact of an affiliate campaign using a finite distributed lag model. The last three parts of the thesis examine long-term advertising effects utilizing daily data. This segment addresses issues such as accurate seasonality estimation, variable selection, and modelling of non-linear response curves through a combination of non-linear optimization within a genetic algorithm framework and Fourier transformation for seasonality decomposition. The outcomes of these complex methods are compared with results from neural network analyses on the same dataset. Findings from these studies indicate a significant immediate consumer response to TV ads, typically within one minute, highlighting the potential for rapid creative testing and early campaign optimization. As for the long-term effects, the econometrics model outperformed the neural network model by achieving better performance on the test set and provides a realistic estimation of factors contributing to business performance. This model enables the evaluation of different investment strategies and, therefore, finding optimal investment strategies for enhanced ROI or brand building based on estimated response curves.