This June issue of The Journal of Risk Model Validation once again consists of five papers and contains a wide range of topics.
The first of the five papers, “Forecasting scenarios from the perspective of a reverse stress test using second-order cone programming”, is by Katsuhiro Tanaka. Loosely speaking, second-order cone programming is the inclusion of quadratic inequality constraints in conventional quadratic programming, which was originally based on linear inequality constraints. This paper uses the concept of an acceptable range, which is determined by the Mahalanobis distance, to set up an optimization and a framework for stress testing. Second-order cone programming is then applied to this acceptable range.
“Goodness-of-fit for discrete-choice models of borrower default” by Arden Hall is the issue’s second paper. It analyzes the effectiveness of goodness-of-fit tests based on ranks, such as the Kolomogorov–Smirnov test, the Gini coefficient and the receiver operating characteristic (ROC) curve, in terms of validating discrete-choice models of default. Simulation evidence is presented that shows that the tests perform well, even when the probability of default is incorrectly measured. Additional robust ways of assessing goodness-of-fit are also discussed.
Jiaming Liu and Chong Wu are the authors of our third paper: “A gradient-boosting decision-tree approach for firm failure prediction: an empirical model evaluation of Chinese listed companies”. This paper looks at corporate failure prediction. The authors employ a gradient boosting decision tree (GBDT) method to improve firm failure prediction. They argue that it gives a more intuitive explanation of the role of factors and a better analysis of the relative importance of each financial variable. The proposed model is compared with four other popular ensemble methods. The authors’ experimental results show that the GBDT outperforms the other ensemble methods in terms of accuracy, precision, F -score and area under the curve (AUC). Liu and Wu set out to provide a full validation of the GBDT method and argue that it is useful in controlling risk in financial risk management.
Our fourth paper is “Modeling impacts of stock jumps on real estate investment trust returns with application to value-at-risk” by Fen-Ying Chen. Unlike the majority of the existing literature, which has empirically concentrated on the relationship between real estate investment trust (REIT) prices and the stock market, this paper claims to be the first to directly model the effects of stock jumps on REIT returns associated with an alternative dynamic process. The link between this and model validation/stress testing is that the empirical results show the magnitudes of jumps in expected returns and the volatility impact on REITs’ value-at-risk.
“Simple models in finance: a mathematical analysis of the probabilistic recognition heuristic” by Martín Egozcue, Luis Fuentes García, Konstantinos V. Katsikopoulos and Michael Smithson is the fifth and final paper in the issue. Thomas Gray once said: “Where ignorance is bliss, ’tis folly to be wise”. This paper addresses that point. It is well known, state the authors, that laypeople and practitioners often resist using complex mathematical models, such as the ones proposed by economics or finance, and instead use fast and frugal strategies to make decisions. In this paper the authors study one such strategy: the recognition heuristic. This heuristic states that people infer that an object they recognize has a higher value of a criterion of interest than an object they do not recognize. They extend previous studies by including a general model of the recognition heuristic that considers probabilistic recognition, and they undertake a mathematical analysis to provide a number of results. Finally, they discuss whether having less information could be convenient for making more accurate decisions, thus giving substance to that contemporary phrase “too much information”.
Trinity College, University of Cambridge
Forecasting scenarios from the perspective of a reverse stress test using second-order cone programming
This paper proposes a model for forecasting scenarios from the perspective of a reverse stress test using interest rate, equity and foreign exchange data.
This paper demonstrates that the rank-order tests are unreliable for assessing models to be used to predict probabilities.
A gradient-boosting decision-tree approach for firm failure prediction: an empirical model evaluation of Chinese listed companies
In this paper, the authors employ a gradient-boosting decision-tree method to improve firm failure prediction and explain how to better analyze the relative importance of each financial variable.
Modeling impacts of stock jumps on real estate investment trust returns with application to value-at-risk
This paper aims to model the impact of extreme stock jumps on REIT returns.
In this paper, the authors present a general model of the recognition heuristic that assumes that objects’ recognition is random.