| dc.description.abstract | This thesis employs XGBoost and linear regression to reveal whether short-term stock
price trends can explain the variance in textual sentiment in Norwegian equity research
reports covering the Oslo Stock Exchange, investigating whether trend-chasing bias is
present in equity research. The thesis is based on 2,350 equity research reports from the
past 5 years covering the 25 largest companies by market capitalization listed on the Oslo
Stock Exchange, published by Carnegie, DNB Markets, and Pareto Securities.
We present empirical evidence demonstrating an enhancement in the predictive efficacy of
our model with the integration of short-term stock price trend indicators. Specifically,
the incorporation of these indicators resulted in a 2.2% increase in the linear regression
model’s explanatory power compared to our reference model. The full model can account
for 44.7% of the variance in textual sentiment. Further, the XGBoost model improves
predictive accuracy over the linear model and returns the lagged sentiment, investment
bank, financial leverage, and RSI to be the most important variables explaining sentiment,
chronologically ordered by variable importance. The 3-month simple return and MACD
prove to be similar in variable importance with traditional valuation metrics such as the
P/E ratio and firm size. Thus, we find that stock price trend indicators improve the
models capacity to explain the sentiment of an equity research report.
However, our findings cannot state that the given dependency is due to trend-chasing
bias in Norwegian equity research. The textual sentiment is determined by numerous
unobservable variables, making it likely that our model suffers from omitted variable bias,
thus causing endogeneity issues. Further, we cannot determine if a change in textual
sentiment is attributable to a measurable change in the perception of a company, or the
fact that the reports summarize and relay market information. | en_US |
