- Tasks:
(sort in chronological order)
| Reference | Paper | Source | Model Type | Evaluation Metric(s) | Time Span | Contributions | Venue |
|---|---|---|---|---|---|---|---|
| Theil et al. (2018) | Word Embeddings-Based Uncertainty Detection in Financial DisclosuresChristoph | CRSP & EDGAR | Jrip | Precision, Recall, F1, Accuracy, p-value | 1994-2015 | Expand the L&M uncertainty dictionary by adding scematically close terms. Investigated manual filtering by expert and specific versus general-domain embedding model. New dataset of annotated annual reports. | ECONLP-2018 |
| Theil et al. (2019) | PRoFET: Predicting the Risk of Firms from Event Transcripts | They present a new dataset of 90K earnings call transcripts (open source) | BiLSTM | Linear correlation coefficient Pearson’s r, the Non-linear rank correlation coefficients Spearman’s ρ and Kendall’s τ, and the MSE | 2002–2017 | They introduce PRoFET, the first neural model for volatility prediction jointly exploiting both semantic language representations and a comprehensive set of financial features | IJCAI-19 |
| Theil et al. (2020) | Explaining Financial Uncertainty through Specialized Word Embeddings | EDGAR | Regression | t-statistic of coefficients | 1994-2015 | Developed the first word embedding models accounting for industry-specific vocabulary. Statistically explain drift in returns and volality of stock price. | ACM Transactions on Data Science-20 |
| Qin and Yang et al. (2019) | What You Say and How You Say It Matters: Predicting Stock Volatility Using Verbal and Vocal Cues | Dataset | MDRM | MSE | S&P 500 companies in 2017 | First work to extend the earnings conference call analysis as a multimodal problem by incorporating textual and audio information in the same model. | ACL-19 |
| Yang et al. (2020) | HTML: Hierarchical Transformer-based Multi-task Learning for Volatility Prediction | Code | HTML (Hiererachical Transformer with Multi-task Learning) | MSE | S&P 500 companies in 2017 | Yang et al. (2020) is the first approach that can deal with the long-form financial documents for volatility prediction. This paper demonstrates very significant improvements in prediction accuracy, in the range 17% - 49% compared to the current state-of-the-art. | WWW-20 |
| Ye et al. (2020) | Financial Risk Prediction with Multi-Round Q&A Attention Network | 6494 text-only samples | Bidirectional LSTM | MSE | 2015-2018 | Instead of word-level or document level feature extraction, they focus on dialogues in the conference. Through extracting features of each round of dialogue, the model predicts the financial volatility. Compared to the most common baseline, past volatility, proposed model achieves 47% improvement in 3-days span, 31% in 7-days span, and 23% in 15-days span. | IJCAI-20 |
| Li et al. (2020) | MAEC: A Multimodal Aligned Earnings Conference Call Dataset for Financial Risk Prediction | MAEC Dataset | Multimodal Alignment | MSE | S&P 1500 companies from 2015 to 2018 | This is a data resource paper that is more than six times larger than those currently available to the research community. | CIKM-20 |
(sort in chronological order)
| Reference | Paper | Source | Model Type | Evaluation Metric(s) | Time Span | Contributions | Venue |
|---|---|---|---|---|---|---|---|
| Yang et al. (2020) | Financial Risk Analysis for SMEs with Graph-based Supply Chain Mining | Alipay Ant SME Lending (NOT open-sourced) | Spatial-temporal aware Graph Neural Network | Supply Chain Mining task, Loan Default Prediction task | \ | Analyze financial risk through mining the supply chain between Small and Medium-size Enterprises | IJCAI-20 |
| Bisi et al. (2020) | Risk-Averse Trust Region Optimization for Reward-Volatility Reduction | \ | \ | Reward Volatility, Most of proof is given in theoretical derivation. | \ | In many cases, the risk is measured not only on a long-term perspective, but also on a step-wise reward. (like on a daily base) | IJCAI-20 |
| Cheng et al. (2020) | Risk Guarantee Prediction in Networked-Loans | A real-world dataset from a major institution in East Asia (NOT open-sourced) | Graph convolutional network, Graph recurrent network | On datasets in 2014,2015 and 2016, got 84.6%, 84,2% and 85% accuracy respectively. | 01/01/2013-31/12/2016 | Detect and predict the risk in a guaranteed loan | IJCAI-20 |
| Ting-Wei et al. (2020) | “The Cat is Out of the Bag” Explainable Risk Ranking with Financial Reports | \ | WRR(self-constructed) (RankSVM + FastText + HAN) | Spearman’s Rho (ρ), Kendall’s Tau (τ) | 01/01/1996-31/12/2013 | Propose an eXplainable Risk Ranking (XRR) model that uses multilevel encoders and attention mechanisms to analyze financial risks among companies | AAAI-20 |
| Masson & Montariol (2020) | Detecting Omissions of Risk Factors in Company Annual Reports | DoRe Corpus (Masson and Paroubek, 2020) | FlauBERT | Accuracy, F1, Precision, Recall | 2009-2019 | First to look at risk omission in annual reports | FinNLP-2020 |