Business

Online Appendix: High Frequency Newswire Textual Sentiment Analysis

Online Appendix: High Frequency Newswire Textual Sentiment Analysis: Evidence From International Stock Markets During The European Financial Crisis

Andreas S. Chouliaras

Luxembourg School of Finance

December 31, 2015

Abstract:

This is the internet appendix for “High Frequency Newswire Textual Sentiment Analysis: Evidence from International Stock Markets during the European Financial Crisis”.

The paper “High Frequency Newswire Textual Sentiment Analysis: Evidence from International Stock Markets during the European Financial Crisis” to which this Appendix applies is available at: http://ssrn.com/abstract=2572597.

Online Appendix: High Frequency Newswire Textual Sentiment Analysis: Evidence From International Stock Markets During The European Financial Crisis – Introduction

Table 1: Portugal Stock Market – 30 minutes. The model I employ the study the effect of the content of news in high frequency stock returns is the following: Screenshot_6 where Mt takes the value of, the positive (Gt), the negative (Bt), the pessimism (Pt) and the news count (Nt), of the previous 30-minute interval, as defined in Section 3.5 of the paper. I control for five lags of returns (i.e. five 30-minute lagged returns for every stock market) to deal with autocorrelation in the returns. The regressions I perform are robust, using the Huber-White sandwich estimators (Huber (1967), White (1980)) to deal with autocorrelation, heteroskedasticity, heterogeneity and lack of normality.

High Frequency

Table 2: Ireland stock market – 30 minutes. The model I employ the study the effect of the content of news in high frequency stock returns is the following: Screenshot_6 where Mt takes the value of, the positive (Gt), the negative (Bt), the pessimism (Pt) and the news count (Nt), of the previous 30-minute interval, as defined in Section 3.5 of the paper. I control for five lags of returns (i.e. five 30-minute lagged returns for every stock market) to deal with autocorrelation in the returns. The regressions I perform are robust, using the Huber-White sandwich estimators (Huber (1967), White (1980)) to deal with autocorrelation, heteroskedasticity, heterogeneity and lack of normality.

High Frequency

Table 3: Italy stock market – 30 minutes. The model I employ the study the effect of the content of news in high frequency stock returns is the following: Screenshot_6 where Mt takes the value of, the positive (Gt), the negative (Bt), the pessimism (Pt) and the news count (Nt), of the previous 30-minute interval, as defined in Section 3.5 of the paper. I control for five lags of returns (i.e. five 30-minute lagged returns for every stock market) to deal with autocorrelation in the returns. The regressions I perform are robust, using the Huber-White sandwich estimators (Huber (1967), White (1980)) to deal with autocorrelation, heteroskedasticity, heterogeneity and lack of normality.

High Frequency

Table 4: Greece stock market – 30 minutes. The model I employ the study the effect of the content of news in high frequency stock returns is the following: Screenshot_6 where Mt takes the value of, the positive (Gt), the negative (Bt), the pessimism (Pt) and the news count (Nt), of the previous 30-minute interval, as defined in Section 3.5 of the paper. I control for five lags of returns (i.e. five 30-minute lagged returns for every stock market) to deal with autocorrelation in the returns. The regressions I perform are robust, using the Huber-White sandwich estimators (Huber (1967), White (1980)) to deal with autocorrelation, heteroskedasticity, heterogeneity and lack of normality.

High Frequency

See full PDF below.