A Semiparametric Gaussian Copula Regression Model for Predicting Financial Risks from Earnings Calls

Earnings call summarizes the financial performance of a company, and it is an important indicator of the future financial risks of the company.

Predicting the risks of publicly listed companies is of great interests not only to the traders and analysts on the Wall Street, but also virtually anyone who has investments in the market (Kogan et al., 2009).

Fung et al.

In NLP, many statistical machine learning methods that capture the dependencies among random variables, including topic models (Blei et al., 2003; Lafferty and Blei, 2005; Wang et al., 2012), always have to make assumptions with the underlying distributions of the random variables, and make use of informative priors.

We use three datasets4 of transcribed quarterly earnings calls from the US.

Volatility is an important measure of the financial risk, and in this work, we focus on predicting the future volatility following the earnings teleconfer-

6.1 Experimental Setup

In the experimental section, we notice that the proposed semiparametric Gaussian copula model has obtained promising results in various setups on three datasets in this text regression task.

In this work, we have demonstrated that the more complex quarterly earnings calls can also be used to predict the measured volatility of the stocks in the limited future.

Appears in 15 sentences as: SVM (18)

In *A Semiparametric Gaussian Copula Regression Model for Predicting Financial Risks from Earnings Calls*

- 0 Our results significantly outperform standard linear regression and strong SVM baselines.Page 2, “Introduction”
- (2003) are among the first to study SVM and text mining methods in the market prediction domain, where they align financial news articles with multiple time series to simulate the 33 stocks in the Hong Kong Hang Seng Index.Page 2, “Related Work”
- (2009) model the SEC-mandated annual reports, and performs linear SVM regression with e-insensitive loss function to predict the measured volatility.Page 2, “Related Work”
- Traditional discriminative models, such as linear regression and linear SVM , have been very popular in various text regression tasks, such as predicting movie revenues from reviews (Joshi et al., 2010), understanding the geographic lexical variation (Eisenstein et al., 2010), and predicting food prices from menus (Chahuneau et al., 2012).Page 2, “Related Work”
- The baselines are standard squared-loss linear regression, linear kernel SVM, and nonlinear (Gaussian) kernel SVM .Page 6, “Experiments”
- We use the Statistical Toolbox’s linear regression implementation in Matlab, and LibSVM (Chang and Lin, 2011) for training and testing the SVM models.Page 6, “Experiments”
- The hyperparameter C in linear SVM, and the 7 and C hyperparameters in Gaussian SVM are tuned on the training set using 10-fold cross-validation.Page 6, “Experiments”
- On the pre-2009 dataset, we see that the linear regression and linear SVM perform reasonably well, but the Gaussian kernel SVM performs less well, probably due to overfitting.Page 6, “Experiments”
- Similar performances are also obtained in the 2009 dataset, where the result of linear SVM baseline falls behind.Page 6, “Experiments”
- On the post—2009 dataset, none of results from the linear and nonlinear SVM models can match up with the linear regression model, but our proposed copula model still improves over all baselines by a large margin.Page 6, “Experiments”
- From the experiments on the pre-2009 dataset, we see that when the amount of training data is small (25%), both SVM models have obtained very impressive results.Page 6, “Experiments”

See all papers in *Proc. ACL 2014* that mention SVM.

See all papers in *Proc. ACL* that mention SVM.

Back to top.

Appears in 13 sentences as: regression model (13)

In *A Semiparametric Gaussian Copula Regression Model for Predicting Financial Risks from Earnings Calls*

- Our proposed semiparametric copula regression model takes a different perspective.Page 2, “Copula Models for Text Regression”
- Then we describe the proposed semiparametric Gaussian copula text regression model .Page 3, “Copula Models for Text Regression”
- We formulate the copula regression model as follows.Page 3, “Copula Models for Text Regression”
- The algorithmic implementation of our semiparametric Gaussian copula text regression model is shown in Algorithm 1.Page 5, “Copula Models for Text Regression”
- In the first experiment, we compare the proposed semiparametric Gaussian copula regression model to three baselines on three datasets with all features.Page 6, “Experiments”
- On the post—2009 dataset, none of results from the linear and nonlinear SVM models can match up with the linear regression model , but our proposed copula model still improves over all baselines by a large margin.Page 6, “Experiments”
- To understand the learning curve of our proposed copula regression model , we use the 25%, 50%, 75% subsets from the training data, and evaluate all four models.Page 6, “Experiments”
- Interestingly, the proposed copula regression model has dominated all methods for both metrics throughout all proportions of the “post-2009” earnings calls dataset, where instead of financial crisis, the economic recovery is the main theme.Page 7, “Experiments”
- Finally, we investigate the robustness of the proposed semiparametric Gaussian copula regression model by varying the amount of features in the covariate space.Page 7, “Experiments”
- On the pre-2009 dataset, we see that the gaps between the best-perform copula model and the second-best linear regression model are consistent throughout all feature sizes.Page 7, “Experiments”
- On the 2009 dataset, we see that the performance of Gaussian copula is aligned with the linear regression model in terms of Spearman’s correlation, where the former seems to perform better in terms of Kendall’s tau.Page 7, “Experiments”

See all papers in *Proc. ACL 2014* that mention regression model.

See all papers in *Proc. ACL* that mention regression model.

Back to top.

Appears in 11 sentences as: linear regression (11)

In *A Semiparametric Gaussian Copula Regression Model for Predicting Financial Risks from Earnings Calls*

- To evaluate the performance of our approach, we compare with a standard squared loss linear regression baseline, as well as strong basehnes such as hnear and non:hnear supportPage 1, “Introduction”
- 0 Our results significantly outperform standard linear regression and strong SVM baselines.Page 2, “Introduction”
- Traditional discriminative models, such as linear regression and linear SVM, have been very popular in various text regression tasks, such as predicting movie revenues from reviews (Joshi et al., 2010), understanding the geographic lexical variation (Eisenstein et al., 2010), and predicting food prices from menus (Chahuneau et al., 2012).Page 2, “Related Work”
- The baselines are standard squared-loss linear regression , linear kernel SVM, and nonlinear (Gaussian) kernel SVM.Page 6, “Experiments”
- We use the Statistical Toolbox’s linear regression implementation in Matlab, and LibSVM (Chang and Lin, 2011) for training and testing the SVM models.Page 6, “Experiments”
- On the pre-2009 dataset, we see that the linear regression and linear SVM perform reasonably well, but the Gaussian kernel SVM performs less well, probably due to overfitting.Page 6, “Experiments”
- On the post—2009 dataset, none of results from the linear and nonlinear SVM models can match up with the linear regression model, but our proposed copula model still improves over all baselines by a large margin.Page 6, “Experiments”
- On the pre-2009 dataset, we see that the gaps between the best-perform copula model and the second-best linear regression model are consistent throughout all feature sizes.Page 7, “Experiments”
- On the 2009 dataset, we see that the performance of Gaussian copula is aligned with the linear regression model in terms of Spearman’s correlation, where the former seems to perform better in terms of Kendall’s tau.Page 7, “Experiments”
- model over squared loss linear regression model are increasing, when working with larger feature spaces.Page 8, “Experiments”
- Focusing on the three financial crisis related datasets, the proposed model significantly outperform the standard linear regression method in statistics and strong discriminative support vector regression baselines.Page 9, “Conclusion”

See all papers in *Proc. ACL 2014* that mention linear regression.

See all papers in *Proc. ACL* that mention linear regression.

Back to top.

Appears in 5 sentences as: overfitting (5)

- On the other hand, once such assumptions are removed, another problem arises — they might be prone to errors, and suffer from the overfitting issue.Page 2, “Copula Models for Text Regression”
- Therefore, coping with the tradeoff between expressiveness and overfitting , seems to be rather important in statistical approaches that capture stochastic dependency.Page 2, “Copula Models for Text Regression”
- This is of crucial importance to modeling text data: instead of using the classic bag-of-words representation that uses raw counts, we are now working with uniform marginal CDFs, which helps coping with the overfitting issue due to noise and data sparsity.Page 4, “Copula Models for Text Regression”
- On the pre-2009 dataset, we see that the linear regression and linear SVM perform reasonably well, but the Gaussian kernel SVM performs less well, probably due to overfitting .Page 6, “Experiments”
- The second issue is about overfitting .Page 9, “Discussions”

See all papers in *Proc. ACL 2014* that mention overfitting.

See all papers in *Proc. ACL* that mention overfitting.

Back to top.

Appears in 4 sentences as: feature space (3) feature spaces (1)

- For example, when bag-of-word-unigrams are present in the feature space , it is easier if one does not explicitly model the stochastic dependencies among the words, even though doing so might hurt the predictive power, while the variance from the correlations among the random variables is not explained.Page 2, “Related Work”
- By doing this, we are essentially performing probability integral transform— an important statistical technique that moves beyond the count-based bag-of-words feature space to marginal cumulative density functions space.Page 3, “Copula Models for Text Regression”
- model over squared loss linear regression model are increasing, when working with larger feature spaces .Page 8, “Experiments”
- By applying the Probability Integral Transform to raw features in the copula model, we essentially avoid comparing apples and oranges in the feature space , which is a common problem in bag-of-features models in NLP.Page 9, “Discussions”

See all papers in *Proc. ACL 2014* that mention feature space.

See all papers in *Proc. ACL* that mention feature space.

Back to top.

Appears in 4 sentences as: proposed model (4)

- Christensen (2005) shows that sorting and balanced binary trees can be used to calculate the correlation coefficients with complexity of 0(nlog Therefore, the computational complexity of MLE for the proposed model is O(n logPage 4, “Copula Models for Text Regression”
- main questions we ask are: how is the proposed model different from standard text regres-siorflclassification models?Page 9, “Discussions”
- Focusing on the three financial crisis related datasets, the proposed model significantly outperform the standard linear regression method in statistics and strong discriminative support vector regression baselines.Page 9, “Conclusion”
- By varying the size of the training data and the dimensionality of the covariates, we have demonstrated that our proposed model is relatively robust across different parameter settings.Page 9, “Conclusion”

See all papers in *Proc. ACL 2014* that mention proposed model.

See all papers in *Proc. ACL* that mention proposed model.

Back to top.

Appears in 4 sentences as: significantly outperform (2) significantly outperforms (2)

- In experiments, we show that our model significantly outperforms strong linear and nonlinear discriminative baselines on three datasets under various settings.Page 1, “Abstract”
- By varying different experimental settings on three datasets concerning different periods of the Great Recession from 2006-2013, we empirically show that our approach significantly outperforms the baselines by a wide margin.Page 2, “Introduction”
- 0 Our results significantly outperform standard linear regression and strong SVM baselines.Page 2, “Introduction”
- Focusing on the three financial crisis related datasets, the proposed model significantly outperform the standard linear regression method in statistics and strong discriminative support vector regression baselines.Page 9, “Conclusion”

See all papers in *Proc. ACL 2014* that mention significantly outperform.

See all papers in *Proc. ACL* that mention significantly outperform.

Back to top.

Appears in 3 sentences as: bag-of-words (3)

- By performing probability integral transform, our approach moves beyond the standard count-based bag-of-words models in NLP, and improves previous work on text regression by incorporating the correlation among local features in the form of semiparametric Gaussian copula.Page 1, “Abstract”
- By doing this, we are essentially performing probability integral transform— an important statistical technique that moves beyond the count-based bag-of-words feature space to marginal cumulative density functions space.Page 3, “Copula Models for Text Regression”
- This is of crucial importance to modeling text data: instead of using the classic bag-of-words representation that uses raw counts, we are now working with uniform marginal CDFs, which helps coping with the overfitting issue due to noise and data sparsity.Page 4, “Copula Models for Text Regression”

See all papers in *Proc. ACL 2014* that mention bag-of-words.

See all papers in *Proc. ACL* that mention bag-of-words.

Back to top.

Appears in 3 sentences as: feature set (1) feature sets (1) Feature sets: (1)

- This nice property essentially allows us to fuse distinctive lexical, syntactic, and semantic feature sets naturally into a single compact model.Page 3, “Copula Models for Text Regression”
- Feature sets:Page 6, “Experiments”
- To do this, we sample equal amount of features from each feature set , and concatenatePage 7, “Experiments”

See all papers in *Proc. ACL 2014* that mention Feature sets:.

See all papers in *Proc. ACL* that mention Feature sets:.

Back to top.

Appears in 3 sentences as: machine learning (3)

- Copula models (Schweizer and Sklar, 1983; Nelsen, 1999) are often used by statisticians (Genest and Favre, 2007; Liu et al., 2012; Masarotto and Varin, 2012) andecononfiMB(Chenandfbn,2006)u)Mudythe bivariate and multivariate stochastic dependency among random variables, but they are very new to the machine learning (Ghahramani et al., 2012; Han et al., 2012; Xiang and Neville, 2013; Lopez-paz et al., 2013) and related communities (Eick-hoff et al., 2013).Page 1, “Introduction”
- In NLP, many statistical machine learning methods that capture the dependencies among random variables, including topic models (Blei et al., 2003; Lafferty and Blei, 2005; Wang et al., 2012), always have to make assumptions with the underlying distributions of the random variables, and make use of informative priors.Page 2, “Copula Models for Text Regression”
- This mixed form of formal statement and informal speech brought difficulties to machine learning algorithms.Page 5, “Datasets”

See all papers in *Proc. ACL 2014* that mention machine learning.

See all papers in *Proc. ACL* that mention machine learning.

Back to top.