| Remaining Sample Size | Observations Removed | |
|---|---|---|
| Total number of articles | 43,132,204 | |
| Drop articles out of date range | 37,256,210 | 5,875,994 |
| Drop articles with no associated stock | 33,062,855 | 4,193,355 |
| Drop articles with more than one stock tagged | 32,143,209 | 919,646 |
| Drop articles without match with the CSMAR database | 22,543,726 | 9,599,483 |
Repository for the processed data. Folder for data information. Number of articles by hour, date, and year. Number of stocks mentioned from Jan 2015 to Jul 2019.
Dictionary of parameters: https://github.com/xiubooth/Textual-Analysis/blob/main/params/params.py
Repository for trained models, hyper-parameters, and L/S portfolio positions, weights and returns
| Long-EW | Short-EW | LS-EW | Long-VW | Short-VW | LS-VW | Index | |
|---|---|---|---|---|---|---|---|
| Average daily return | 0.264% | 0.159% | 0.423% | 0.309% | 0.106% | 0.416% | -0.001% |
| Sharpe ratio | 1.72 | 1.26 | 11.01 | 2.12 | 1.06 | 8.68 | 0.07 |
| Turnover | 66.6% | 70.5% | 68.6% | 65.9% | 68.7% | 67.3% | / |
Full summary of results for analysis of sentiment charged words.
Top 10 frequent words: 涨(0.0619), 跌(0.0536), 发展(0.0433), 胜(0.0241), 平安(0.0216), 高新(0.0184), 建设(0.0177), 动力(0.0167), 健康(0.0158), 创业(0.0135)
Bottom 10 sentiment words: 跌(-0.0536), 垃圾(-0.0071), 杀(-0.0032), 反弹(-0.0027), 下跌(-0.0027), 差(-0.0017), 机会(-0.0016), 大跌(-0.0014), 问题(-0.0014), 受益(-0.0013)
Top 10 sentiment words: 涨(0.0233), 发展(0.0109), 胜(0.0064), 创业(0.0062), 建设(0.0060), 高新(0.0058), 健康(0.0049), 幸福(0.0044), 创新(0.0036), 动力(0.0036)
Repository for trained models, hyper-parameters, and L/S portfolio positions, weights and returns
| Long-EW | Short-EW | LS-EW | Long-VW | Short-VW | LS-VW | Index | |
|---|---|---|---|---|---|---|---|
| Average daily return | 0.059% | 0.234% | 0.293% | -0.044% | 0.268% | 0.224% | -0.001% |
| Sharpe ratio | 0.63 | 1.52 | 6.92 | -0.60 | 1.80 | 3.09 | 0.07 |
| Turnover | 86.4% | 91.5% | 89.0% | 90.4% | 92.4% | 91.4% | / |
| Long-EW | Short-EW | LS-EW | Long-VW | Short-VW | LS-VW | Index | |
|---|---|---|---|---|---|---|---|
| Average daily return | 0.005% | 0.307% | 0.312% | -0.047% | 0.231% | 0.183% | -0.001% |
| Sharpe ratio | 0.17 | 1.77 | 6.81 | -0.67 | 1.65 | 2.70 | 0.07 |
| Turnover | 88.3% | 91.5% | 89.9% | 91.3% | 92.9% | 92.1% | / |
-
The articles whose
returnsorthree-day returnscannot be obtained are removed from the data during pre-processing (many of which corresponds to delisted stocks). This corresponds to purposefully avoiding future delisting -- the information that we could not know in advance. -
It is not feasible to directly short stocks in the Chinese market. We can short on ETFs instead.
-
A simple strategy that counts the occurrence of
涨and跌(which have the highest occurrence and the most positive/negative sentiments)
- Look at the performance of the strategies on the CSI 300 constituent stocks.