Credit default swap prediction based on generative adversarial networks
| DOI | https://doi.org/10.1108/DTA-09-2021-0260 |
| Published date | 24 March 2022 |
| Date | 24 March 2022 |
| Pages | 720-740 |
| Subject Matter | Library & information science,Librarianship/library management,Library technology,Information behaviour & retrieval,Metadata,Information & knowledge management,Information & communications technology,Internet |
| Author | Shu-Ying Lin,Duen-Ren Liu,Hsien-Pin Huang |
Credit default swap prediction
based on generative
adversarial networks
Shu-Ying Lin
Department of Finance, Minghsin University of Science and Technology,
Xinfeng, Taiwan, and
Duen-Ren Liu and Hsien-Pin Huang
Institute of Information Management, National Yang Ming Chiao Tung University,
Hsinchu, Taiwan
Abstract
Purpose –Financial price forecast issues are always a concern of investors. However, the financial applications
based on machine learning methods mainly focus on stock market predictions. Few studies have explored credit risk
predictions. Understanding credit risk trends can help investors avoid market risks. The purpose of this study is to
investigate the prediction model that can effectively predict credit default swaps (CDS).
Design/methodology/approach –A novel generative adversarial network (GAN) for CDS prediction is
proposed. The authors take three features into account that are highly relevant to the future trends of CDS:
historical CDS price, news and financial leverage. The main goal of this model is to improve the existing GAN-
based regression model by adding finance and news feature extraction approaches. The proposed model
adopts an attentional long short-term memorynetwork and convolution network to process historical CDS data
and news information, respectively. In addition to enhancing the effectiveness of the GAN model, the authors
also design a data sampling strategy to alleviate the overfitting issue.
Findings –The authors conduct an experiment with a real dataset and evaluate the performance of the
proposed model. The components and selected features of the model are evaluated for their ability to improve
the prediction performance. Theexperimental results show that the proposed model performs better than other
machine learning algorithms and traditional regression GAN.
Originality/value –There are very few studies on prediction models for CDS. With the proposed novel
approach, the authors can improve the performance of CDS predictions. The proposed work can thereby
increase the commercial value of CDS predictions to support trading decisions.
Keywords Credit default swap, Financial prediction, Deep learning, LSTM, Generative adversarial network,
Regression
Paper type Research paper
1. Introduction
Financial price prediction methods can be divided into two types. Traditional economic
models, including autoregressive (AR), moving average (MA), and autoregressive-moving-
average (ARMA), comprise one type (Li and Li, 1996;Siami-Namini and Namin, 2018;Wei
et al., 2014). Most of these models follow some strong statistical assumptions. However,
financial markets in the real world fluctuate; hence, they cannot be captured by statistical
assumptions. The other type of financial price prediction is based on machine learning
methods. Traditional machine learning models (Bhandari et al., 2019) include artificial neural
networks (ANN), support vector machines (SVM), random forests (RF), and gradient boosting
decision trees (GBDT) (Friedman, 2001). In recent years, a newer form of machine learning
called deep learning has become a trend (Pandey and Janghel, 2021). Several studies have
proposed deep neural network models, such as recurrent neural networks (RNN) and long-
DTA
56,5
720
This research was partially supported by the Ministry of Science and Technology of Taiwan under
grant number: MOST 108-2410-H-009-046-MY2.
The current issue and full text archive of this journal is available on Emerald Insight at:
https://www.emerald.com/insight/2514-9288.htm
Received 23 September 2021
Revised 8 December 2021
6 February 2022
Accepted 7 March 2022
Data Technologies and
Applications
Vol. 56 No. 5, 2022
pp. 720-740
© Emerald Publishing Limited
2514-9288
DOI 10.1108/DTA-09-2021-0260
short-term memory neural networks (LSTM) to improve the accuracy of financial time series
prediction (Fischer and Krauss, 2018;Kim and Kang, 2019;Li et al., 2018;Selvin et al., 2017;
Siami-Namini and Namin, 2018).
Most of the studies using machine learning approaches deal with the stock price
prediction in financial markets (Akita et al., 2016;Chen et al., 2017;Ding et al., 2015;Fischer
and Krauss, 2018;Hu et al., 2018;Vargas et al., 2017); Some studies have focused on
identifying financial statement fraud (An and Suh, 2020) and predicting corporate credit
rating (Choi et al., 2020). However, there are few applications for predicting the premium of
credit default swaps (CDS). Therefore, this paper uses CDS as our research target. CDS is a
contingent claim whereby a buyer and seller sign a contract to transfer the credit risk of the
target company. The target company may be a single company, a basket of companies, or a
stock market index. During the life of the contract, the buyer pays a periodic premium to the
seller in exchange for protection of the underlying asset/assets. If the agreed credit event
occurs, the buyer receives compensation or a pre-agreed amount for the loss. At the end of the
swap contract, if no credit events have occurred, the buyer stops paying the premium to the
seller and the seller pays nothing to the buyer.
Merton (1974) proposed a structural form model to show the probability of default, and
assumed that default occurs only when a company cannot pay the interest and principal at
the time of debt maturity. Different from the structural form model, the reduced-form credit
risk models focus on modeling the probability of default as a stochastic process based on
multi-factor model and dynamic analysis of interest rates such as the changes in the term
structure of interest rates, including upward sloping, downward sloping and flattened shapes
based on comparing the short-term and long-term yields (Jarrow and Turnbull, 1995;Liberto,
2019). According to the structural form model, the probability of default depends on three
factors: financial leverage ratio, volatility of assets, and the term structure of a risk-free rate.
Ericsson et al. (2009) used the quote premium of CDS and regression model to explore the
linear relationship between the above three factors and CDS quotes. Following the suggestion
of Ericsson et al. (2009), this study uses daily CDS quotes, which are the price of credit risk.
When using corporate bond yields, it is necessary to collect the risk-free interest rate to
calculate the credit risk premium. Compared to corporate bond yields, CDS quotes are faster
and more accurate in reflecting changes in credit risk. CDS markets are also more active than
corporate bond markets.
In addition to referring to fundamental analysis and technical analysis, investors usually
read textual information when determining whether to purchase financial assets. These items
of information (e.g., official documents, news reports, and social media) influence people’s
decisions to some degree, and thus affect the market. Therefore, in recent years, there has
been considerable research devoted to imitating human behaviors by considering both
numerical values and textual information to construct the prediction model (Akita et al., 2016;
Hu et al., 2018). Accordingly, we apply deep learning methods and take financial news into
account to capture the pattern of real-time and dynamic information.
Some studies analyze the correlation between news media sentiment and financial
markets, especially their impact on stock returns and volatility. For example, Tetlock (2007)
use pessimistic media sentiment as a measure of investors’negative sentiment and find that
the former can predict stock market decline: the higher the degree of negative sentiment, the
lower the market reward and volume. Tetlock et al. (2008) point out that news sentiment is
able to capture implicit information beyond the corporate fundamentals that can be
quantified. Ferguson et al. (2015) also believe that the media’s emotional tone can change the
behavior of investors and affect the stock market. In sum, empirical analysis finds that
positive and negative words in financial news influence investors’decision-making.
Accordingly, we consider the text market sentiment score to forecast the CDS markets.
Credit default
swap
prediction
721
To continue reading
Request your trial