Proposal for Turning Point Detection Method Using Financial Text and Transformer

Abstract

In this study, we demonstrate whether analysts’ sentiment toward individual stocks is useful for stock market analysis. This can be achieved by creating a polarity index in analyst reports using natural language processing. In this study, we calculated anomaly scores for the created polarity index using anomaly detection algorithms. The results show that the proposed method is effective in detecting the turning point of the polarity index.

Appendices

Appendix A Evaluation Metrics for Time Series Deep Learning Models

In this section, we define the comparison metrics used in the comparison experiments.

$$\begin{aligned} MAE = \frac{1}{N}\sum _{i=1}^{N}|y_i - \hat{y_i}| \end{aligned}$$

(3)

$$\begin{aligned} MSE = \frac{1}{N}\sum _{i=1}^{N}(y_i - \hat{y_i})^2 \end{aligned}$$

(4)

$$\begin{aligned} RMSE = \sqrt{\frac{1}{N}\sum _{i=1}^{N}(y_i - \hat{y_i})^2} \end{aligned}$$

(5)

$$\begin{aligned} MSLE = \frac{1}{N}\sum _{i=1}^{N} (log(y_i + 1) - log(\hat{y_i} + 1))^2 \end{aligned}$$

(6)

where $\hat{y_i}$ is the predicted value at i and $\hat{y_i}$ is the measured value at i. The closer to zero each evaluation index is, the better it is.

Appendix B Empirical Results for Industry-Specific Polarity Index

This appendix shows the anomaly scores of the polarity index by industry. The selection of industries was done randomly among 33 industries (Figs. 7, 8 and 9).

Fig. 7.

Turning Point Detection for Construction Polarity Index

Fig. 8.

Turning Point Detection for Wholesale Polarity Index

Fig. 9.

Turning Point Detection for Other Products Polarity Index