C55 - Large Data Sets: Modeling and AnalysisReturn

This study investigates whether the response of stock returns to economic policy uncertainty de- pends on the level of government size in the economy. Although there is a consensus in the liter- ature that stock markets react negatively to policy-related uncertainties, the factors that determine the magnitude of this effect have been ignored. This study is the first to demonstrate that the magnitude of this effect depends on the size of the government in the economy. In the study, data for the period 1997Q1-2021Q4 pertaining to 18 countries are used. According to results of fixed-effects estimations with Driscoll-Kraay robust standard errors, economic policy uncer- tainty affects stock returns negatively. In addition, the coefficient of interaction term formed by the variables of policy uncertainty and government size is also negative and significant. These results indicate that the negative response of stock returns to policy uncertainty grows as gov- ernment size increases. The sensitivity analysis results show that the findings are not sensitive to the estimations made by alternative approaches and are therefore robust. The findings of the study contain important implications for policymakers. Investors can also benefit from the results at the point of international asset allocation against future policy-related uncertainties.

With the advancement of machine learning tools, an increasing number of algorithms are being utilized for predicting not only traditional time series data related to financial markets but also those connected to cryptocurrencies. This paper aims to compare various machine learning algorithms used for prediction, in order to identify the one with the greatest practical potential for creating a prediction model of Bitcoin’s price as an investment asset. The analysis focuses on supervised learning algorithms, taking into account the nature of the task involving long time series datasets. The paper also describes the exact process of creating and setting up individual models and their parameters, explaining procedures for obtaining and editing datasets, and shows how to evaluate performance of these models. In addition to the analysis of the main subject of research, which is Bitcoin, the paper also uses an analysis of reference cryptocurrencies such as Ethereum, Litecoin and NEO to compare the resulting performances. The processes consisting of editing the analysed datasets, creating individual prediction models, training and testing the performance of models on historical data, and creating, debugging and implementing individual machine learning models were realised through coding in the Python program.

We provide systematic evidence of the effect of alphabetical sorting on corporate performance based on over a decade of data covering seventeen EU countries in three European language families. We also measure the effects of using English words in a corporate name in a non-English-speaking country, of a corporate name containing a 'national' (patriotic) word, and of simple linguistic properties highlighted in the brand-name marketing literature. Using multiple measures of corporate performance, we find companies sorted low in the alphabet to be less successful in several European countries, particularly in services. 'National' words are associated with substantially higher sales growth in, e.g., Poland, Norway, France and Spain, while the use of English words in company names curbs (fosters) sales growth in Romance- (Slavic-) language countries.