依據新聞情緒建構市場擇時策略:以台灣市場為例
No Thumbnail Available
Date
2021
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
隨著網路資訊普及，投資人能夠即時獲取財經新聞，並能對所獲取的資訊有即時反應，因此，本研究以 2014 年 4 月至 2015 年 12 月，台灣加權指數為樣本，透過文字探勘之方式，量化新聞詞語，計算並結構化新聞情緒，將新聞情緒分為正向、負向新聞情緒，探討新聞情緒對投資人進行交易決策之影響。實證結果顯示，日平均新聞情緒之遞延效果兩日後無顯著與報酬有關，負面新聞情緒較原始新聞情緒對報酬率影響程度較大，而隔夜的新聞情緒會因投資人的過度反應而與報酬產生負向關係，將結構化後新聞情緒分數轉為交易指標濾網後，能有效提升程式之績效，其中空方績效提升較多方大。
The prevalence of information online enables investors to browse real-time financial news and react immediately to the obtained information. This study samples the Taiwan Weighted Stock Index between April 2014 and December 2015 by using text mining to quantify words and phrases in news articles. Sentiments reflected by each article are calculated and structured to divide them into positive and negative sentiments. The goal is to explore the effect of news sentiments on trade decisions made by investors. Empirical results reveal that daily average news sentiments exert a delayed effect on the return on investment; however, this effect becomes nonsignificant after 2 days. Negative news sentiments exert a greater effect on the return rate than do original news sentiments. Overnight news sentiments are negatively correlated with the return because of overreaction in investors. After structured news sentiment scores are converted to trade indicator filters, program performance can be improved effectively, and the level of improvement to short position performance is higher than that of long position performance.
The prevalence of information online enables investors to browse real-time financial news and react immediately to the obtained information. This study samples the Taiwan Weighted Stock Index between April 2014 and December 2015 by using text mining to quantify words and phrases in news articles. Sentiments reflected by each article are calculated and structured to divide them into positive and negative sentiments. The goal is to explore the effect of news sentiments on trade decisions made by investors. Empirical results reveal that daily average news sentiments exert a delayed effect on the return on investment; however, this effect becomes nonsignificant after 2 days. Negative news sentiments exert a greater effect on the return rate than do original news sentiments. Overnight news sentiments are negatively correlated with the return because of overreaction in investors. After structured news sentiment scores are converted to trade indicator filters, program performance can be improved effectively, and the level of improvement to short position performance is higher than that of long position performance.
Description
Keywords
文字探勘, 新聞情緒, 程式交易, Text mining, News sentiment, Program trading