Chulwoo Han

Deep learning networks for stock market analysis and prediction

Deep learning networks are applied to stock market analysis and prediction.A comprehensive analysis with different data representation methods is offered.Five-minute intraday data from the Korean KOSPI stock market is used.The network applied to residuals of autoregressive model improves prediction.Covariance estimation for market structure analysis is improved with the network. We offer a systematic analysis of the use of deep learning networks for stock market analysis and prediction. Its ability to extract features from a large set of raw data without relying on prior knowledge of predictors makes deep learning potentially attractive for stock market prediction at high frequencies. Deep learning algorithms vary considerably in the choice of network structure, activation function, and other model parameters, and their performance is known to depend heavily on the method of data representation. Our study attempts to provides a comprehensive and objective assessment of both the advantages and drawbacks of deep learning algorithms for stock market analysis and prediction. Using high-frequency intraday stock returns as input data, we examine the effects of three unsupervised feature extraction methodsprincipal component analysis, autoencoder, and the restricted Boltzmann machineon the networks overall ability to predict future market behavior. Empirical results suggest that deep neural networks can extract additional information from the residuals of the autoregressive model and improve prediction performance; the same cannot be said when the autoregressive model is applied to the residuals of the network. Covariance estimation is also noticeably improved when the predictive network is applied to covariance-based market structure analysis. Our study offers practical insights and potentially useful directions for further investigation into how deep learning networks can be effectively used for stock market analysis and prediction.

Market overreaction and investment strategies

We investigated the overreaction of the Korean market in response to shocks in the US stock market, and analysed the dynamic relationship between these two markets since 1996. We found that the KOSPI 200 index futures overreacted to the S&P 500 index returns during the period from 2000 to 2009 when the Korean market was in its growth stage. As the Korean market matured and the KOSPI 200 overnight futures were introduced in 2009, the overreaction disappeared. When investors employed the Kelly model or Value-at-Risk to exploit the overreaction, their trading strategies produced significant profits during the growth stage even after considering transaction costs and risk, but the profits attenuated once the overnight futures market was launched in 2009.

An Extended CreditRisk+ Framework for Portfolio Credit Risk Management

Independent sector assumption in the CreditRisk+ has been a major obstacle to implementing the model. Attempts to overcome this limitation have not gained much success. This paper proposes an extension of the original model which accommodates a wide range of sector covariance structures. Existing numerical algorithms designed for the original model can be reused with little modification. Case studies show that our model outperforms other CreditRisk+ variants which allow sector dependency.A simulation version of our model is also introduced, which in turn used to find an optimal portfolio allocation based on the work of Andersson et al. (2001). Simulation error is very small compared to the analytic counterpart and the optimization significantly reduces portfolio credit risk.

Interest Rate Models on Lie Groups

This paper examines an alternative approach to interest rate modeling, in which the nonlinear and random behavior of interest rates is captured by a stochastic differential equation evolving on a curved state space. We consider as candidate state spaces the matrix Lie groups; these offer not only a rich geometric structure, but—unlike general Riemannian manifolds—also allow for diffusion processes to be constructed easily without invoking the machinery of stochastic calculus on manifolds. After formulating bilinear stochastic differential equations on general matrix Lie groups, we then consider interest rate models in which the short rate is defined as linear or quadratic functions of the state. Stochastic volatility is also augmented to these models in a way that respects the Riemannian manifold structure of symmetric positive-definite matrices. Methods for numerical integration, parameter identification, pricing, and other practical issues are addressed through examples.

Modeling severity risk under PD–LGD correlation

In this article, a generic severity risk framework in which loss given default (LGD) is dependent upon probability of default (PD) in an intuitive manner is developed. By modeling the conditional mean of LGD as a function of PD, which also varies with systemic risk factors, this model allows an arbitrary functional relationship between PD and LGD. Based on this framework, several specifications of stochastic LGD are proposed with detailed calibration methods. By combining these models with an extension of CreditRisk+, a versatile mixed Poisson credit risk model that is capable of handling both risk factor correlation and PD–LGD dependency is developed. An efficient simulation algorithm based on importance sampling is also introduced for risk calculation. Empirical studies suggest that ignoring or incorrectly specifying severity risk can significantly underestimate credit risk and a properly defined severity risk model is critical for credit risk measurement as well as downturn LGD estimation.

Impacts of Derivative Markets on Spot Market Volatility and Their Persistence

In this article, we investigate the impacts of futures and options markets on the volatility of the underlying market. Unlike earlier studies, the focus is on their persistence over time. Tests on the Hang Seng index yield several interesting results that often contrast with previous findings. Empirical results suggest that the quality of new information generated by derivative trading determines the impacts on the spot market volatility. The futures market provides new, material information reducing spot market volatility. The Options market, on the other hand, generates noisy information and distorts price, which is followed by an increase in volatility and a decrease in its sensitivity to price change. While the impact of futures persists, that of options mostly disappears as the market matures. Our conjecture is that the futures market is mainly driven by informed, experienced participants, while the options market attracts new, inexperienced investors.

Logit Regression Based Bankruptcy Prediction of Korean Firms

In this article, we develop a bankruptcy prediction model for Korean firms that utilize logit regression. We find that not only financial accounting ratios but equity market inputs and macro-economic variables are also important predictors of bankruptcy. However, unlike the findings of Campbell et al. (2008), using market value of equity in computing total assets did not improve the model. We compare the model with a Merton-type structural model and find that our model demonstrates a higher prediction power in distinguishing distressed firms from healthy firms. Though our model proves to perform better, we are careful to make a conclusion and rather suggest using several models for the purpose of risk management to reduce model risk.

Efficient Value at Risk Estimation for Mortgage-Backed Securities

We develop an efficient Monte Carlo simulation-based methodology for value at risk (VaR) and sensitivity analysis of mortgage-backed securities (MBS) that employs an importance sampling technique developed for quadratic VaR models. Our approach, whose validity is derived from a fundamental result in perturbation analysis, is applicable to any analytic interest rate and prepayment model, and more generally to any path-dependent cashflows that admit analytic gradients. We compare the accuracy and computational performance of our VaR estimators with those obtained via finite-difference gradient approximation schemes.

Partial structural break identification.

We propose an extension of the existing information criterion-based structural break identification approaches. The extended approach helps identify both pure structural change (break) and partial structural change (break). A pure structural change refers to the case when breaks occur simultaneously in all parameters of regression equation, whereas a partial structural change happens when breaks occur in some parameters only. Our approach consistently outperforms other well-known approaches. We also extend the simulation studies of Bai and Perron (2006 and Hall, Osborn and Sakkas (2013) by including more general cases. This provides more comprehensive results and reveals the cases where the existing identification approaches lose power, which should be kept in mind when applying them.

Comparative Analysis of Credit Risk Models for Loan Portfolios

In this article, I compare credit risk models that are used for loan portfolios, both from a theoretical perspective and via simulation studies. My study is distinct from previous studies by including new models, considering sector correlation, and performing comprehensive sensitivity analysis. CreditRisk , CreditMetrics, Basel II internal rating based method, and Mercer Oliver Wymans model are considered. Risk factor distribution and the relationship between risk components and risk factors are the key distinguishing characteristics of each model. CreditRisk , due to its extra degree of freedom, has the highest flexibility to fit various loss distributions. It turns out that sector covariance is the most important risk component for risk management in terms of risk sensitivity. Risk sensitivities not only differ among models but also depend on the input parameters and the quantile at which risk is measured. This implies that risk models can only be judged in terms of the portfolio under consideration, and banks should evaluate them based on their own portfolios.