RAROC (Risk-Adjusted Return on Capital) is widely used in financial _rms as a risk-adjusted profitability measure, but its adoption in non- financial firms remains limited. This paper reviews its shortcomings for non-financial companies and, based on recent empirical findings on the statistical distribution of long-term corporate profits, finds a RAROC-based capital allocation theorem that is applicable even with the restrictions faced by non-financial firms.
This paper investigates inference and diagnostic procedures in linear Gaussian spatial models using the Wendland family of covariance functions, compared with the Matérn family, as a spatial dependence structure. We extend the Spatial Dependency Index (SDI) and derive expressions for the practical range for both covariance families, establishing explicit relationships between the range and the decay parameter. A local influence diagnostic approach is employed to assess the sensitivity of maximum likelihood estimators. To determine cutoff values for identifying potentially influential observations, we propose a Jackknife-after-Bootstrap resampling method-a technique not yet applied in this context. The methodology is illustrated through an application to a soybean yield dataset. Our results indicate that the Wendland covariance family performs satisfactorily under Gaussian spatial model assumptions, offering a competitive alternative to the Matérn family, particularly in terms of computational efficiency due to its compact support.
This research comparatively evaluates several statistical and machine learning models to predict normalized wind power output, when applied to a case study of wind power records from the Energy Reliability Council of Texas (ERCOT). The impact on predictive ability derived from two approaches is quantified: (1) modeling the response directly in the [0,1] interval, versus (2) performing a transformation f to some unbounded space. A correction mechanism is proposed to mitigate the bias introduced by the use of the transformation f. The results suggest that combining suitable transformations (v-logit or probit) with bias correction significantly improves wind power prediction, especially when using flexible models like support vector machines (SVM) and generalized additive models (GAM).
Functional geostatistics enables the spatial prediction of curves, profiles, or time series indexed over space. It assumes that each sampling location is associated with a function rather than a scalar variable. Although a substantial body of literature exists on curve prediction (functional geostatistics and related methods), comparatively fewer contributions have focused on simulation methods capable of generating spatially correlated functional data under controllable dependence structures. Such methods are essential for validating methodological developments, comparing predictors, and quantifying uncertainty. In this work, we propose an extension of the turning bands simulation method to the context of functional geostatistics (TBFG). Curves are represented using a functional basis (e.g., Fourier or spline functions), and the corresponding coefficient vectors are modeled as spatially correlated Gaussian random fields. The turning bands simulation is performed in the coefficient space, after which the curves are reconstructed using the selected functional basis. In the implementation, the one-dimensional processes simulated along the bands employ auxiliary covariance functions whose angular averages reproduce the desired two-dimensional covariance structure. A key feature of TBFG is the inclusion of an explicit spatial range parameter, which allows direct control over the strength of spatial dependence among the simulated curves. Spatial dependence is assessed through the trace-variogram, an extension of the classical variogram defined in the functional space (L2). The simulations show that the range of the tracevariogram increases as the imposed spatial range parameter becomes larger, providing a clear and interpretable diagnostic of the strength of spatial dependence. An appendix includes R code that allows the results to be fully reproduced.
Fitting high-dimensional random-effect models can be computationally demanding. To address this challenge, methodologies such as the Integrated Nested Laplace Approximation (INLA) provide efficient deterministic approximations for latent Gaussian models. We illustrate this approach using the Saber 11 mathematics scores from Colombia, fitting Bayesian spatial mixed-effects models with municipality-level and school-level random effects. The model accounts for associations between mathematics performance and selected student, household, and institution-related covariates, while capturing spatial structure across municipalities and heterogeneity between schools. Penalised Complexity priors were used to specify prior distributions for the variance components, and model comparison was performed using predictive assessment criteria, including Automatic Leave-Group-Out Cross-Validation. Based on the reported comparison criteria, the model including both municipality-level spatial effects and school-level random effects showed the best predictive performance among the candidate models.
