Temporal Implicit Multimodal Networks for Investment and Risk Management
Abstract
1 Introduction
![](/cms/10.1145/3643855/asset/9a0ef183-4d15-49ca-b4eb-0f05b1d3c19d/assets/images/medium/tist-2023-04-0169-f01.jpg)
![](/cms/10.1145/3643855/asset/93e527a3-644d-4c4c-9f3c-6d122a4bf413/assets/images/medium/tist-2023-04-0169-f02.jpg)
2 Related Work
2.1 Financial Time-series Forecasting
2.2 Network Learning for Financial Time-series
3 Temporal Implicit Multimodal Network Model
![](/cms/10.1145/3643855/asset/58ed2a71-48af-4a2f-9ecc-d3f2cda031f9/assets/images/medium/tist-2023-04-0169-f03.jpg)
3.1 Temporal Implicit Network Learning
3.2 Dynamic Network Encoding
3.3 Temporal Encoding
3.4 Multimodal Fusion
3.5 Forecasting and Loss Functions
4 Experiments
4.1 Datasets
Textual News Data.
Numerical Stock Price Data.
IN-NY | IN-NA | BE-NY | BE-NA | |
---|---|---|---|---|
No. articles | 221,513 | 1,377,098 | ||
No. assets (stocks) | 374 | 402 | 2,240 | 2,514 |
No. data points | 470,118 | 505,314 | 2,815,680 | 3,160,098 |
![](/cms/10.1145/3643855/asset/c225d5aa-61b8-41b2-9137-fae22840301b/assets/images/medium/tist-2023-04-0169-f04.jpg)
4.2 Tasks and Metrics
4.3 Baselines and Settings
4.4 Results
IN-NY | IN-NA | BE-NY | BE-NA | |||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
RMSE | MAE | SMAPE | RMSE | MAE | SMAPE | RMSE | MAE | SMAPE | RMSE | MAE | SMAPE | |
Mean Forecasting | ||||||||||||
GRU-1 | 0.0653 | 0.0136 | 1.2705 | 0.0259 | 0.0147 | 1.2744 | 0.0768 | 0.0194 | 1.3169 | 0.1937 | 0.0385 | 1.3380 |
TST-1 | 0.0657 | 0.0145 | 1.4798 | 0.0319 | 0.0156 | 1.4492 | 0.0753 | 0.0193 | 1.4898 | 0.1968 | 0.0362 | 1.3724 |
NBEATS | 0.0651 | 0.0136 | 1.3847 | 0.0261 | 0.0155 | 1.2794 | 0.0700 | 0.0185 | 1.4031 | 0.1904 | 0.0326 | 1.3555 |
DARNN | 0.0651 | 0.0137 | 1.3871 | 0.0262 | 0.0148 | 1.3678 | 0.0724 | 0.0179 | 1.3645 | 0.1950 | 0.0322 | 1.3634 |
MTGNN | 0.0652 | 0.0156 | 1.2493 | 0.0428 | 0.0169 | 1.3234 | 0.0703 | 0.0179 | 1.3801 | 0.2323 | 0.0451 | 1.3826 |
FAST | 0.0680 | 0.0148 | 1.4507 | 0.0347 | 0.0174 | 1.3350 | 0.0825 | 0.0199 | 1.4007 | 0.1985 | 0.0395 | 1.3669 |
SE | 0.0706 | 0.0201 | 1.3244 | 0.0429 | 0.0233 | 1.3208 | 0.0869 | 0.0226 | 1.3520 | 0.1980 | 0.0410 | 1.3363 |
GRU-2 | 0.0652 | 0.0140 | 1.2695 | 0.0257 | 0.0146 | 1.2547 | 0.0756 | 0.0192 | 1.3016 | 0.1973 | 0.0368 | 1.3296 |
TST-2 | 0.0656 | 0.0143 | 1.4014 | 0.0329 | 0.0165 | 1.2910 | 0.0768 | 0.0199 | 1.4064 | 0.1962 | 0.0377 | 1.3634 |
TIME | 0.0652 | 0.0115 | 1.0424 | 0.0231 | 0.0115 | 1.0520 | 0.0703 | 0.0164 | 1.2696 | 0.1929 | 0.0320 | 1.2796 |
Volatility Forecasting | ||||||||||||
GRU-1 | 0.1957 | 0.0437 | 0.5357 | 0.0820 | 0.0463 | 0.5517 | 0.2256 | 0.0556 | 0.6336 | 0.5977 | 0.1137 | 0.7841 |
TST-1 | 0.1909 | 0.0442 | 0.5231 | 0.1012 | 0.0499 | 0.5583 | 0.2383 | 0.0629 | 0.6098 | 0.5928 | 0.1181 | 0.6897 |
NBEATS | 0.1571 | 0.0363 | 0.4879 | 0.0722 | 0.0397 | 0.4921 | 0.2250 | 0.0556 | 0.5917 | 0.5926 | 0.1099 | 0.6862 |
DARNN | 0.1848 | 0.0381 | 0.4696 | 0.0754 | 0.0409 | 0.4941 | 0.2294 | 0.0594 | 0.5925 | 0.5963 | 0.1171 | 0.6851 |
MTGNN | 0.1551 | 0.0414 | 0.6033 | 0.1157 | 0.0577 | 0.6244 | 0.2275 | 0.0561 | 0.5937 | 0.5963 | 0.1189 | 0.7110 |
FAST | 0.2125 | 0.0479 | 0.5623 | 0.1170 | 0.0574 | 0.6272 | 0.2722 | 0.0747 | 0.7218 | 0.6018 | 0.1327 | 0.7737 |
SE | 0.2129 | 0.0488 | 0.5758 | 0.1213 | 0.0585 | 0.6270 | 0.2703 | 0.0742 | 0.7044 | 0.6018 | 0.1317 | 0.7102 |
GRU-2 | 0.1946 | 0.0443 | 0.5595 | 0.0806 | 0.0458 | 0.5484 | 0.2234 | 0.0588 | 0.6453 | 0.5995 | 0.1145 | 0.7672 |
TST-2 | 0.1957 | 0.0450 | 0.5389 | 0.1063 | 0.0541 | 0.5970 | 0.2443 | 0.0662 | 0.6487 | 0.5963 | 0.1282 | 0.7532 |
TIME | 0.1550 | 0.0327 | 0.4080 | 0.0722 | 0.0364 | 0.4271 | 0.2200 | 0.0546 | 0.5840 | 0.5922 | 0.1093 | 0.6805 |
Correlation Forecasting | ||||||||||||
GRU-1 | 0.5054 | 0.4383 | 1.3498 | 0.4999 | 0.4326 | 1.4708 | 0.5083 | 0.4391 | 1.4381 | 0.4905 | 0.4210 | 1.5441 |
TST-1 | 0.5069 | 0.4414 | 1.3748 | 0.4987 | 0.4319 | 1.4460 | 0.5068 | 0.4391 | 1.4410 | 0.4891 | 0.4205 | 1.5678 |
NBEATS | 0.5064 | 0.4395 | 1.3507 | 0.4986 | 0.4322 | 1.4571 | 0.5074 | 0.4387 | 1.4339 | 0.4890 | 0.4202 | 1.5550 |
DARNN | 0.5069 | 0.4419 | 1.3761 | 0.4991 | 0.4327 | 1.4602 | 0.5083 | 0.4399 | 1.4372 | 0.4897 | 0.4213 | 1.5773 |
MTGNN | 0.5110 | 0.4435 | 1.3740 | 0.5002 | 0.4329 | 1.4533 | 0.5085 | 0.4405 | 1.4483 | 0.5035 | 0.4238 | 1.5704 |
FAST | 0.5086 | 0.4436 | 1.3888 | 0.4992 | 0.4328 | 1.4640 | 0.5085 | 0.4407 | 1.4541 | 0.4893 | 0.4207 | 1.5661 |
SE | 0.5126 | 0.4431 | 1.3985 | 0.5047 | 0.4348 | 1.4746 | 0.5161 | 0.4433 | 1.4416 | 0.4902 | 0.4198 | 1.5630 |
GRU-2 | 0.5060 | 0.4391 | 1.3670 | 0.5003 | 0.4321 | 1.4609 | 0.5088 | 0.4387 | 1.4224 | 0.4898 | 0.4209 | 1.5598 |
TST-2 | 0.5063 | 0.4408 | 1.3673 | 0.4989 | 0.4329 | 1.4624 | 0.5068 | 0.4393 | 1.4439 | 0.4894 | 0.4209 | 1.5675 |
TIME | 0.4167 | 0.3396 | 1.0260 | 0.4197 | 0.3472 | 1.1291 | 0.4781 | 0.4075 | 1.3107 | 0.4778 | 0.4062 | 1.4731 |
5 Ablation Studies
RMSE | MAE | SMAPE | |
---|---|---|---|
Mean Forecasting | |||
w/o. TimeVect | 0.0716 | 0.0120 | 1.0633 |
w. single net. | 0.0653 | 0.0116 | 1.0545 |
w/o. inner wt. | 0.0655 | 0.0116 | 1.0510 |
R = 10% | 0.0656 | 0.0117 | 1.0787 |
R = 80% | 0.0662 | 0.0117 | 1.0507 |
no backcast loss | 0.0662 | 0.0117 | 1.0573 |
no mean loss | 0.0994 | 0.0653 | 1.7131 |
no vol. loss | 0.0656 | 0.0119 | 1.1032 |
no corr. Loss | 0.0694 | 0.0121 | 1.0813 |
TIME | 0.0652 | 0.0115 | 1.0424 |
Volatility Forecasting | |||
w/o. TimeVect | 0.1617 | 0.0341 | 0.4187 |
w. single net. | 0.1553 | 0.0338 | 0.4266 |
w/o. inner wt. | 0.1560 | 0.0328 | 0.4137 |
R = 10% | 0.1555 | 0.0347 | 0.4416 |
R = 80% | 0.1622 | 0.0332 | 0.4185 |
no backcast loss | 0.1558 | 0.0333 | 0.4268 |
no mean loss | 0.1561 | 0.0363 | 0.4633 |
no vol. loss | 0.2419 | 0.1051 | 1.6133 |
no corr. Loss | 0.1599 | 0.0327 | 0.4096 |
TIME | 0.1550 | 0.0327 | 0.4080 |
Correlation Forecasting | |||
w/o. TimeVect | 0.4230 | 0.3475 | 1.0550 |
w. single net. | 0.4233 | 0.3465 | 1.0480 |
w/o. inner wt. | 0.4184 | 0.3414 | 1.0328 |
R = 10% | 0.4355 | 0.3608 | 1.0913 |
R = 80% | 0.4227 | 0.3463 | 1.0440 |
no backcast loss | 0.4223 | 0.3462 | 1.0505 |
no mean loss | 0.4422 | 0.3674 | 1.1086 |
no vol. loss | 0.4478 | 0.3733 | 1.1267 |
no corr. Loss | 0.5467 | 0.4833 | 1.9203 |
TIME | 0.4167 | 0.3396 | 1.0260 |
6 Case Studies
6.1 Portfolio Allocation
6.2 Value-at-Risk
6.3 Experiment Results
IN-NY | IN-NA | |||||
---|---|---|---|---|---|---|
% Br. | \(\mathcal {R}\) | \(\mathcal {R}^{\prime }\) | % Br. | \(\mathcal {R}\) | \(\mathcal {R}^{\prime }\) | |
GRU-1 | 5.1% | 1.7 | 1.2 | 3.4% | 1.9 | 0.9 |
TST-1 | 11.7% | 1.2 | 1.1 | 3.4% | 1.0 | 1.0 |
NBEATS | 5.1% | 1.4 | 1.5 | 2.4% | 1.7 | 0.8 |
DARNN | 8.5% | 1.7 | 1.1 | 4.7% | 1.3 | 0.8 |
MTGNN | 6.3% | 1.8 | 1.1 | 9.4% | 0.6 | 0.4 |
FAST | 17.9% | 0.1 | 0.1 | 13.7% | 0.3 | 0.4 |
SE | 12.6% | 0.1 | 0.1 | 5.7% | 0.3 | 0.5 |
GRU-2 | 6.3% | 2.5 | 1.7 | 3.4% | 1.8 | 0.6 |
TST-2 | 13.5% | 1.3 | 1.2 | 5.3% | 0.8 | 1.1 |
TIME | 4.3% | 2.5 | 2.2 | 2.2% | 2.7 | 2.9 |
7 Interpretability
![](/cms/10.1145/3643855/asset/67a14ebb-a19f-4a29-b597-1538dfe7e892/assets/images/medium/tist-2023-04-0169-f05.jpg)
8 Conclusion and Future Work
Footnotes
References
Index Terms
- Temporal Implicit Multimodal Networks for Investment and Risk Management
Recommendations
Investment and Risk Management with Online News and Heterogeneous Networks
Stock price movements in financial markets are influenced by large volumes of news from diverse sources on the web, e.g., online news outlets, blogs, social media. Extracting useful information from online news for financial tasks, e.g., forecasting stock ...
Multimodal Multi-Task Financial Risk Forecasting
MM '20: Proceedings of the 28th ACM International Conference on MultimediaStock price movement and volatility prediction aim to predict stocks' future trends to help investors make sound investment decisions and model financial risk. Companies' earnings calls are a rich, underexplored source of multimodal information for ...
Uncertainty, Capital Investment, and Risk Management
We use forward-looking and exogenous measures of output price uncertainty to examine the effect of price uncertainty on firm-level capital investment, risk management, and debt issuance. The effects of uncertainty vary significantly by firm size. When ...
Comments
Information & Contributors
Information
Published In
![cover image ACM Transactions on Intelligent Systems and Technology](/cms/asset/39b3a3ac-629c-4d20-8d33-a8fed02a9503/3613561.cover.jpg)
Publisher
Association for Computing Machinery
New York, NY, United States
Publication History
Check for updates
Author Tags
Qualifiers
- Research-article
Funding Sources
- National Research Foundation, Singapore
- Monetary Authority of Singapore Postgraduate Scholarship
Contributors
Other Metrics
Bibliometrics & Citations
Bibliometrics
Article Metrics
- 0Total Citations
- 1,148Total Downloads
- Downloads (Last 12 months)1,148
- Downloads (Last 6 weeks)144
Other Metrics
Citations
View Options
Login options
Check if you have access through your login credentials or your institution to get full access on this article.
Sign in