Pichler, Jan (2023). Three Essays in Financial Economics. (Thesis). Universität Bern, Bern

Preview

Text 23pichler_j.pdf - Thesis Available under License Creative Commons: Attribution-Noncommercial-No Derivative Works (CC-BY-NC-ND 4.0). Download (6MB) | Preview

Abstract

This thesis encompasses three essays, each of which examines the role of information in a specific setting arising in financial economics. Thus, each essay contributes to the literature about the role of information in financial markets and to the debate whether financial markets are efficient or not. The first essay investigates return patterns around news events by analyzing the largest news dataset studied in finance so far: 4.4 million news headlines between January 1996 and December 2019 on firms listed in North America. I use a finance-specific sentiment dictionary to classify these news headlines into positive and negative ones and contrast the results of this approach to classifications based on supervised learning models trained on the market reaction to the news. These supervised learning models include the multinomial Naïve Bayes method and several rudimentary neural networks. This paper contributes to the literature by showing that supervised learning models outperform the sentiment dictionary approach traditionally used in finance. Furthermore, it provides evidence that financial markets anticipate news in the weeks ahead, that the new information is quickly priced in, and that there is no drift afterward. The second essay is a follow-up on the first: It merges the same news dataset with the data on the trades at the New York Stock Exchange (NYSE) between January 2011 and December 2019. This seemingly trivial process is technically challenging because the NYSE Trades and Quotes (TAQ) data contains every trade with a time stamp precision of nanoseconds, i.e., hundreds of millions of trades per trading day, and is several terabytes of raw data. The resulting dataset contains 2.3 million observations and covers the eight hours before and after news publication at the second frequency. I use the models from the first essay to classify the news into positive and negative, show return patterns for the two types of news, and test trading strategies that react instantly to the new information. This study contributes to the literature by providing multiple pieces of evidence that support the efficient market hypothesis at high frequencies. First, I confirm the finding of the first essay that financial markets anticipate new information and show that they react instantly and price in the new information within minutes. Second, the tested trading strategies yield surprisingly low returns. Finally, the average return and volatility patterns around all news are highly consistent with rational pricing: The elevated volatility before news expresses the uncertainty about the content of the news (markets usually know that information is coming because the firms often schedule a news release, but they do not know the content) and I show that holding stocks in the 6.5 hours (one trading day) before news yields an average excess return of 0.1%. The third essay tests multiple measures of profitability and their trend regarding their ability to predict stock returns. Because such predictors challenge the Efficient Market Hypothesis (EMH), they are called anomalies. Reviewing these anomalies is necessary because they sometimes change over time and many disappear when analyzing a different period (especially post-publication) or, even worse, when just applying proper asset pricing tests. I cover the period from June 1980 to December 2021 and mainly analyze six different profitability measures concerning their level, trend, and level relative to the industry’s mean. This paper makes multiple contributions to the literature. First, I show that the trend-of-profitability effect described in Akbas et al. (2017) is mainly driven by the period 2000 to 2006 and has been reversing since then. This finding is also robust against slight changes in their methodology. Second, I confirm that the cleaning of Compustat’s Selling, General and Administrative (SG&A) cost variable by re-adding Research and Development (R&D) expenses, as described in Ball et al (2015), improves not only their profitability measure but also the one used in Fama and French (2015). Third, I show that the difference to the industry’s mean yields strong results in Fama-MacBeth regressions; however, it does not translate into high value-weighted portfolio returns and therefore lags the absolute level of profitability as a predictor for future returns. Fourth, I propose to use a different measure of value compared to the popular book equity-to-market equity ratio, namely Gross Profit (GP) minus SG&A divided by market equity because the first seems to have lost its predictive power while the latter has not and was also a stronger predictor before. This measure of value is an ideal complement to measures of profitability. Collectively, this thesis contributes to the debate on market efficiency and stock return predictability. The first essay finds that financial markets already anticipate news in the weeks ahead, price them in quickly, and that individual news is not a medium- or long-term return predictor. The second essay shows that this pattern can also be found at the intraday level, that most of the new information is priced in immediately, and that the average drift afterward is minimal and only lasts a few minutes. The third essay contrasts the evidence of market efficiency from the first two with long-term return predictability based on profitability measures. While the third essay is not proof against market efficiency because there are theoretical risk-related explanations for excess returns of highly profitable stocks, I consider them at least questionable. Furthermore, I would also like to highlight that two of the risk factors of Fama and French, namely size and value, have had negative returns for over a decade. Should this persist, it raises the question of whether markets were efficient and are not anymore, or if they were not and are now. From a behavioral pers pective, we as humans may be capable of correctly assessing the impact of individual new information but are ignorant about focusing on what is truly relevant in this ever-expanding sea of information.

Item Type:	Thesis
Dissertation Type:	Cumulative
Date of Defense:	4 July 2023
Subjects:	300 Social sciences, sociology & anthropology > 330 Economics
Institute / Center:	03 Faculty of Business, Economics and Social Sciences > Department of Business Management > Institute of Financial Management
Depositing User:	Hammer Igor
Date Deposited:	19 Dec 2023 08:38
Last Modified:	19 Dec 2023 08:45
URI:	https://boristheses.unibe.ch/id/eprint/4798

Actions (login required)

View Item