Predicting default risk under asymmetric binary link functions

In this article we propose the use of an asymmetric binary link function to extend the proportional hazard model for predicting loan default. The rationale behind this approach is that the symmetry assumption that has been widely used in the literature could be considered as quite restrictive, especially during periods of financial distress. In our approach we allow for a flexible level of asymmetry in the probability of default by the use of the skewed logit distribution. This enable us to estimate the actual level of asymmetry that is associated with the data at hand. We implement our approach to both simulated data and a rich micro dataset of consumer loan accounts. Our results provide clear evidence that ignoring the actual level of asymmetry leads to seriously biased estimates of the slope coefficients, inaccurate marginal effects of the covariates of the model, and overestimation of the probability of default. Regarding the predictive power of the covariates of the model, we have found that loan-specific covariates contain considerably more information about the loan default than macroeconomic covariates, which are often used in practice to carry out macroprudential stress testing.     

Predicting LGD distributions with mixed continuous and discrete ordinal outcomes

The loss given default (LGD) distribution is known to have a complex structure. Consequently, the parametric approach for its prediction by fitting a density function may suffer a loss of predictive power. To overcome this potential drawback, we use the cumulative probability model (CPM) to predict the LGD distribution. The CPM applies a transformed variable to model the LGD distribution. This transformed variable has a semiparametric structure. It models the predictor effects parametrically. The functional form of the transformation is unspecified. Thus, CPM provides more flexibility and simplicity in modeling the LGD distribution. To implement CPM, we collect a sample of defaulted debts from Moody’s Default and Recovery Database. Given this sample, we use an expanding rolling window approach to investigate the out-of-time performance of CPM and its alternatives. Our results confirm that CPM is better than its alternatives, in the sense of yielding more accurate LGD distribution predictions.     

Improved Neymanian analysis for 2K factorial designs with binary outcomes

2^K factorial designs are widely adopted by statisticians and the broader scientific community. In this short note, under the potential outcomes framework, we adopt the partial identification approach and derive the sharp lower bound of the sampling variance of the estimated factorial effects, which leads to an “improved” Neymanian variance estimator that mitigates the overestimation issue suffered by the classic Neymanian variance estimator.     

A semiparametric model for binary response and continuous outcomes under index heteroscedasticity

This paper formulates a likelihood‐based estimator for a double‐index, semiparametric binary response equation. A novel feature of this estimator is that it is based on density estimation under local smoothing. While the proofs differ from those based on alternative density estimators, the finite sample performance of the estimator is significantly improved. As binary responses often appear as endogenous regressors in continuous outcome equations, we also develop an optimal instrumental variables estimator in this context. For this purpose, we specialize the double‐index model for binary response to one with heteroscedasticity that depends on an index different from that underlying the ‘mean response’. We show that such (multiplicative) heteroscedasticity, whose form is not parametrically specified, effectively induces exclusion restrictions on the outcomes equation. The estimator developed exploits such identifying information. We provide simulation evidence on the favorable performance of the estimators and illustrate their use through an empirical application on the determinants, and affect, of attendance at a government‐financed school. Copyright © 2009 John Wiley & Sons, Ltd.     

Predicting general election outcomes: campaigns and changing voter knowledge at the 2017 general election in England

There is a growing literature suggesting that the result for each constituency at British general elections can be predicted using ‘citizen forecasts’ obtained through voter surveys. This may be true for the majority of constituencies where the result at previous contests was a substantial majority for one party’s candidates: few ‘safe seats’ change hands. But is it true in the marginal constituencies, where elections are won and lost? Analysis of such ‘citizen forecast’ data for the Labour-Conservative marginal constituencies in 2017 indicates not. Although respondents were aware of the seats’ relative marginality and of general trends in party support during the campaign, they could not separate out those that were eventually lost by each party from those that were won again, even in seats where the elected party won comfortably.

     

Dose–response modelling for bivariate covariates with and without a spike at zero: theory and application to binary outcomes

In epidemiology and clinical research, there is often a proportion of unexposed individuals resulting in zero values of exposure, meaning that some individuals are not exposed and those exposed have some continuous distribution. Examples are smoking or alcohol consumption. We will call these variables with a spike at zero (SAZ). In this paper, we performed a systematic investigation on how to model covariates with a SAZ and derived theoretical odds ratio functions for selected bivariate distributions. We consider the bivariate normal and bivariate log normal distribution with a SAZ. Both confounding and effect modification can be elegantly described by formalizing the covariance matrix given the binary outcome variable Y. To model the effect of these variables, we use a procedure based on fractional polynomials first introduced by Royston and Altman (1994, Applied Statistics 43: 429–467) and modified for the SAZ situation (Royston and Sauerbrei, 2008, Multivariable model‐building: a pragmatic approach to regression analysis based on fractional polynomials for modelling continuous variables, Wiley; Becher et al., 2012, Biometrical Journal 54: 686–700). We aim to contribute to theory, practical procedures and application in epidemiology and clinical research to derive multivariable models for variables with a SAZ. As an example, we use data from a case–control study on lung cancer.     

Smoothed binary regression quantiles

This paper extends results regarding smoothed median binary regression to general smoothed binary quantile regression, discusses the interpretation of the resulting estimators under alternative assumptions, and shows how they may be used to obtain semiparametric estimates of counterfactual probabilities. The estimators are applied to a model of labour force participation of married women in the USA. We find that the elasticity with respect to non‐labour income is significantly negative only for women that belong to the middle of the conditional willingness‐to‐participate (WTP) distribution. In comparing the quantile models with parametric logit and semiparametric single‐index specifications, we find that the models agree closely for women around the centre of the WTP distribution, but there are considerable disagreements as we move towards the tails of the distribution. Copyright © 2006 John Wiley & Sons, Ltd.     

Non-causality in bivariate binary time series

In this paper we develop a dynamic discrete-time bivariate probit model, in which the conditions for Granger non-causality can be represented and tested. The conditions for simultaneous independence are also worked out. The model is extended in order to allow for covariates, representing individual as well as time heterogeneity. The proposed model can be estimated by Maximum Likelihood. Granger non-causality and simultaneous independence can be tested by Likelihood Ratio or Wald tests. A specialized version of the model, aimed at testing Granger non-causality with bivariate discrete-time survival data is also discussed. The proposed tests are illustrated in two empirical applications.     

Predicting criminal recidivism using neural networks

Prediction of criminal recidivism has been extensively studied in criminology with a variety of statistical models. This article proposes the use of neural network (NN) models to address the problem of splitting the population into two groups — non-recidivists and eventual recidivists — based on a set of predictor variables. The results from an empirical study of the classification capabilities of NN on a well-known recidivism data set are presented and discussed in comparison with logistic regression. Analysis indicates that NN models are competitive with, and may offer some advantages over, traditional statistical models in this domain.     

2007年我国经济周期波动分析与主要宏观经济指标的预测——利用Logistic回归模型的分析

继2006年GDP增长率达到10.7%后,2007年一季度经济继续升温,国内外需求旺盛,工业结构调整步伐加快,能源等支撑条件改善,居民、企业和财政收入继续提高,人民生活水平进一步改善。然而,在消费、出口较快增长的同时,投资增速虽略显温和,但规模仍然较高,新开工项目居高不下,投资增速     

Identification and information in monotone binary models

This paper considers binary response models where errors are uncorrelated with a set of instrumental variables and are independent of a continuous regressor v$v$, conditional on all other variables. It is shown that these exclusion restrictions are not sufficient for identification and that additional identifying assumptions are needed. Such an assumption, introduced by Lewbel [Semiparametric qualitative response model estimation with unknown heteroskedasticity or instrumental variables. Journal of Econometrics 97, 145–177], is that the support of the continuous regressor is large, but we show that it significantly restricts the class of binary phenomena which can be analysed. We propose an alternative additional assumption under which β$\beta$ remains just identified and the estimation unchanged. This alternative assumption does not impose specific restrictions on the data, which broadens the scope of the estimation method in empirical work. The semiparametric efficiency bound of the model is also established and an existing estimator is shown to achieve that bound. The efficient estimator uses a plug-in density estimate. It is shown that plugging in the true density rather than an estimate is inefficient. Extensions to ordered choice models are provided.     

Reference group influence on binary choices dynamics

The recent literature has analyzed binary choices dynamics providing interesting results. Most of these contributions consider interactions within a single group. Nevertheless, in some situations the interaction takes place not only within a single group but also between different groups. In this paper, we investigate the choice dynamics when considering two populations where one serves as a reference group. Considering this influence effect enriches the dynamics. Although the structurally stable resulting dynamics are attracting cycles only, with any positive integer period, the reference group makes the dynamics of the influenced population much more complex. We considered both the possibility that the reference group has the same or the opposite attitude toward the distribution over the choices. We show how the dynamics and the bifurcation structure are modified under the influence of the reference group. Our results illustrate how the propensity to switch choices in the reference groups may, indirectly, affect choices in the first group.     

强化大学英语听力训练中的预测能力

本文通过阐明预测能力在听力理解中重要作用,指出教师在课堂教学中,应重视对学生预测能力的培养,并分析了它的概念,影响预测能力的因素及时机问题,最后提出了实际训练师使用的方法,帮助学生提高英语听力水平,促进英语听力课堂的效果。     

Predicting turning points in the housing market

We identify leading indicators of changes in the housing market and compare their performance in predicting turning points. Being able to predict turning points is of importance to the home building industry, homeowners, and makers of housing policy. Our leading indicators include the Wells Fargo/NAHB Housing Market Index, two of its forward looking components, and an index of consumer sentiment regarding purchasing a home. Our comparison tests include Granger causality and a Bayesian predictor of the probability of a turning point. We find that the measure of consumer sentiment performs relatively well compared to the HMI in predicting home permits, housing starts, and new home sales.     

Predicting bank insolvencies using machine learning techniques

Proactively monitoring and assessing the economic health of financial institutions has always been the cornerstone of supervisory authorities. In this work, we employ a series of modeling techniques to predict bank insolvencies on a sample of US-based financial institutions. Our empirical results indicate that the method of Random Forests (RF) has a superior out-of-sample and out-of-time predictive performance, with Neural Networks also performing almost equally well as RF in out-of-time samples. These conclusions are drawn not only by comparison with broadly used bank failure models, such as Logistic, but also by comparison with other advanced machine learning techniques. Furthermore, our results illustrate that in the CAMELS evaluation framework, metrics related to earnings and capital constitute the factors with higher marginal contribution to the prediction of bank failures. Finally, we assess the generalization of our model by providing a case study to a sample of major European banks.     

Evaluating Pseudo-R2's for binary probit models

Many applied researchers of limited dependent variable models found it disadvantageous that a widely accepted Pseudo-R² does not exist for this type of estimation. The paper provides guidance for researchers in choosing a Pseudo-R² in the binary probit case. The starting point is that R² is best understood in the ordinary least squares (OLS) case with continuous data, which is chosen as the reference situation. It is considered which Pseudo-R² is best able to mimic the OLS-R². The results are surprisingly clear: a measure suggested by McKelvey-Zavoina performs the best under our criterion. However, in the more likely case of low Pseudo-R²'s, a normalization of a measure proposed by Aldrich-Nelson which we suggest is almost as good as the McKelvey-Zavoina, and is in general easier to calculate. We also show that if the underlying R² is predicted using cubic regressions given the Pseudo-R², all measures perform much better.     

A binary tree algorithm on change points detection

In this paper, an algorithm using binary trees is developed to detect the change points of a data set in which the data are assumed to be normally distributed. Usual BIC-type criteria are considered in the binary searching procedures when the number of change points is unknown. The algorithm is also extended to the switching regression models. Simulation study confirms that our algorithm is efficient compared with the ML-method. A real data example also verifies that the proposed procedure is appropriate.