Neural and wavelet network models for financial distress classification

This work analyzes the use of linear discriminant models, multi-layer perceptron neural networks and wavelet networks for corporate financial distress prediction. Although simple and easy to interpret, linear models require statistical assumptions that may be unrealistic. Neural networks are able to discriminate patterns that are not linearly separable, but the large number of parameters involved in a neural model often causes generalization problems. Wavelet networks are classification models that implement nonlinear discriminant surfaces as the superposition of dilated and translated versions of a single "mother wavelet" function. In this paper, an algorithm is proposed to select dilation and translation parameters that yield a wavelet network classifier with good parsimony characteristics. The models are compared in a case study involving failed and continuing British firms in the period 1997-2000. Problems associated with over-parameterized neural networks are illustrated and the Optimal Brain Damage pruning technique is employed to obtain a parsimonious neural model. The results, supported by a re-sampling study, show that both neural and wavelet networks may be a valid alternative to classical linear discriminant models.

Using continuous ground-based radar and lidar measurements for evaluating the representation of clouds in four operational models

The ability of four operational weather forecast models [ECMWF, Action de Recherche Petite Echelle Grande Echelle model (ARPEGE), Regional Atmospheric Climate Model (RACMO), and Met Office] to generate a cloud at the right location and time (the cloud frequency of occurrence) is assessed in the present paper using a two-year time series of observations collected by profiling ground-based active remote sensors (cloud radar and lidar) located at three different sites in western Europe (Cabauw. Netherlands; Chilbolton, United Kingdom; and Palaiseau, France). Particular attention is given to potential biases that may arise from instrumentation differences (especially sensitivity) from one site to another and intermittent sampling. In a second step the statistical properties of the cloud variables involved in most advanced cloud schemes of numerical weather forecast models (ice water content and cloud fraction) are characterized and compared with their counterparts in the models. The two years of observations are first considered as a whole in order to evaluate the accuracy of the statistical representation of the cloud variables in each model. It is shown that all models tend to produce too many high-level clouds, with too-high cloud fraction and ice water content. The midlevel and low-level cloud occurrence is also generally overestimated, with too-low cloud fraction but a correct ice water content. The dataset is then divided into seasons to evaluate the potential of the models to generate different cloud situations in response to different large-scale forcings. Strong variations in cloud occurrence are found in the observations from one season to the same season the following year as well as in the seasonal cycle. Overall, the model biases observed using the whole dataset are still found at seasonal scale, but the models generally manage to well reproduce the observed seasonal variations in cloud occurrence. Overall, models do not generate the same cloud fraction distributions and these distributions do not agree with the observations. Another general conclusion is that the use of continuous ground-based radar and lidar observations is definitely a powerful tool for evaluating model cloud schemes and for a responsive assessment of the benefit achieved by changing or tuning a model cloud

A statistical mechanical approach for the computation of the climatic response to general forcings

The climate belongs to the class of non-equilibrium forced and dissipative systems, for which most results of quasi-equilibrium statistical mechanics, including the fluctuation-dissipation theorem, do not apply. In this paper we show for the first time how the Ruelle linear response theory, developed for studying rigorously the impact of perturbations on general observables of non-equilibrium statistical mechanical systems, can be applied with great success to analyze the climatic response to general forcings. The crucial value of the Ruelle theory lies in the fact that it allows to compute the response of the system in terms of expectation values of explicit and computable functions of the phase space averaged over the invariant measure of the unperturbed state. We choose as test bed a classical version of the Lorenz 96 model, which, in spite of its simplicity, has a well-recognized prototypical value as it is a spatially extended one-dimensional model and presents the basic ingredients, such as dissipation, advection and the presence of an external forcing, of the actual atmosphere. We recapitulate the main aspects of the general response theory and propose some new general results. We then analyze the frequency dependence of the response of both local and global observables to perturbations having localized as well as global spatial patterns. We derive analytically several properties of the corresponding susceptibilities, such as asymptotic behavior, validity of Kramers-Kronig relations, and sum rules, whose main ingredient is the causality principle. We show that all the coefficients of the leading asymptotic expansions as well as the integral constraints can be written as linear function of parameters that describe the unperturbed properties of the system, such as its average energy. Some newly obtained empirical closure equations for such parameters allow to define such properties as an explicit function of the unperturbed forcing parameter alone for a general class of chaotic Lorenz 96 models. We then verify the theoretical predictions from the outputs of the simulations up to a high degree of precision. The theory is used to explain differences in the response of local and global observables, to define the intensive properties of the system, which do not depend on the spatial resolution of the Lorenz 96 model, and to generalize the concept of climate sensitivity to all time scales. We also show how to reconstruct the linear Green function, which maps perturbations of general time patterns into changes in the expectation value of the considered observable for finite as well as infinite time. Finally, we propose a simple yet general methodology to study general Climate Change problems on virtually any time scale by resorting to only well selected simulations, and by taking full advantage of ensemble methods. The specific case of globally averaged surface temperature response to a general pattern of change of the CO2 concentration is discussed. We believe that the proposed approach may constitute a mathematically rigorous and practically very effective way to approach the problem of climate sensitivity, climate prediction, and climate change from a radically new perspective.

The International Urban Energy Balance Models Comparison Project: First Results from Phase 1

A large number of urban surface energy balance models now exist with different assumptions about the important features of the surface and exchange processes that need to be incorporated. To date, no com- parison of these models has been conducted; in contrast, models for natural surfaces have been compared extensively as part of the Project for Intercomparison of Land-surface Parameterization Schemes. Here, the methods and first results from an extensive international comparison of 33 models are presented. The aim of the comparison overall is to understand the complexity required to model energy and water exchanges in urban areas. The degree of complexity included in the models is outlined and impacts on model performance are discussed. During the comparison there have been significant developments in the models with resulting improvements in performance (root-mean-square error falling by up to two-thirds). Evaluation is based on a dataset containing net all-wave radiation, sensible heat, and latent heat flux observations for an industrial area in Vancouver, British Columbia, Canada. The aim of the comparison is twofold: to identify those modeling ap- proaches that minimize the errors in the simulated fluxes of the urban energy balance and to determine the degree of model complexity required for accurate simulations. There is evidence that some classes of models perform better for individual fluxes but no model performs best or worst for all fluxes. In general, the simpler models perform as well as the more complex models based on all statistical measures. Generally the schemes have best overall capability to model net all-wave radiation and least capability to model latent heat flux.

Algorithmic statistical analysis of electrophysiological data for the investigation of structure-activity relationship in single neurons

We are developing computational tools supporting the detailed analysis of the dependence of neural electrophysiological response on dendritic morphology. We approach this problem by combining simulations of faithful models of neurons (experimental real life morphological data with known models of channel kinetics) with algorithmic extraction of morphological and physiological parameters and statistical analysis. In this paper, we present the novel method for an automatic recognition of spike trains in voltage traces, which eliminates the need for human intervention. This enables classification of waveforms with consistent criteria across all the analyzed traces and so it amounts to reduction of the noise in the data. This method allows for an automatic extraction of relevant physiological parameters necessary for further statistical analysis. In order to illustrate the usefulness of this procedure to analyze voltage traces, we characterized the influence of the somatic current injection level on several electrophysiological parameters in a set of modeled neurons. This application suggests that such an algorithmic processing of physiological data extracts parameters in a suitable form for further investigation of structure-activity relationship in single neurons.

Model inter-comparison between statistical and dynamic model assessments of the long-term stability of blanket peat in Great Britain (1940-2099)

We compared output from 3 dynamic process-based models (DMs: ECOSSE, MILLENNIA and the Durham Carbon Model) and 9 bioclimatic envelope models (BCEMs; including BBOG ensemble and PEATSTASH) ranging from simple threshold to semi-process-based models. Model simulations were run at 4 British peatland sites using historical climate data and climate projections under a medium (A1B) emissions scenario from the 11-RCM (regional climate model) ensemble underpinning UKCP09. The models showed that blanket peatlands are vulnerable to projected climate change; however, predictions varied between models as well as between sites. All BCEMs predicted a shift from presence to absence of a climate associated with blanket peat, where the sites with the lowest total annual precipitation were closest to the presence/absence threshold. DMs showed a more variable response. ECOSSE predicted a decline in net C sink and shift to net C source by the end of this century. The Durham Carbon Model predicted a smaller decline in the net C sink strength, but no shift to net C source. MILLENNIA predicted a slight overall increase in the net C sink. In contrast to the BCEM projections, the DMs predicted that the sites with coolest temperatures and greatest total annual precipitation showed the largest change in carbon sinks. In this model inter-comparison, the greatest variation in model output in response to climate change projections was not between the BCEMs and DMs but between the DMs themselves, because of different approaches to modelling soil organic matter pools and decomposition amongst other processes. The difference in the sign of the response has major implications for future climate feedbacks, climate policy and peatland management. Enhanced data collection, in particular monitoring peatland response to current change, would significantly improve model development and projections of future change.

Impacts of pollution and climate change on ombrotrophic Sphagnum species in the UK: analysis of uncertainties in two empirical niche models

A significant challenge in the prediction of climate change impacts on ecosystems and biodiversity is quantifying the sources of uncertainty that emerge within and between different models. Statistical species niche models have grown in popularity, yet no single best technique has been identified reflecting differing performance in different situations. Our aim was to quantify uncertainties associated with the application of 2 complimentary modelling techniques. Generalised linear mixed models (GLMM) and generalised additive mixed models (GAMM) were used to model the realised niche of ombrotrophic Sphagnum species in British peatlands. These models were then used to predict changes in Sphagnum cover between 2020 and 2050 based on projections of climate change and atmospheric deposition of nitrogen and sulphur. Over 90% of the variation in the GLMM predictions was due to niche model parameter uncertainty, dropping to 14% for the GAMM. After having covaried out other factors, average variation in predicted values of Sphagnum cover across UK peatlands was the next largest source of variation (8% for the GLMM and 86% for the GAMM). The better performance of the GAMM needs to be weighed against its tendency to overfit the training data. While our niche models are only a first approximation, we used them to undertake a preliminary evaluation of the relative importance of climate change and nitrogen and sulphur deposition and the geographic locations of the largest expected changes in Sphagnum cover. Predicted changes in cover were all small (generally <1% in an average 4 m2 unit area) but also highly uncertain. Peatlands expected to be most affected by climate change in combination with atmospheric pollution were Dartmoor, Brecon Beacons and the western Lake District.

Evaluating the potential for statistical decadal predictions of sea surface temperatures with a perfect model approach

We explore the potential for making statistical decadal predictions of sea surface temperatures (SSTs) in a perfect model analysis, with a focus on the Atlantic basin. Various statistical methods (Lagged correlations, Linear Inverse Modelling and Constructed Analogue) are found to have significant skill in predicting the internal variability of Atlantic SSTs for up to a decade ahead in control integrations of two different global climate models (GCMs), namely HadCM3 and HadGEM1. Statistical methods which consider non-local information tend to perform best, but which is the most successful statistical method depends on the region considered, GCM data used and prediction lead time. However, the Constructed Analogue method tends to have the highest skill at longer lead times. Importantly, the regions of greatest prediction skill can be very different to regions identified as potentially predictable from variance explained arguments. This finding suggests that significant local decadal variability is not necessarily a prerequisite for skillful decadal predictions, and that the statistical methods are capturing some of the dynamics of low-frequency SST evolution. In particular, using data from HadGEM1, significant skill at lead times of 6–10 years is found in the tropical North Atlantic, a region with relatively little decadal variability compared to interannual variability. This skill appears to come from reconstructing the SSTs in the far north Atlantic, suggesting that the more northern latitudes are optimal for SST observations to improve predictions. We additionally explore whether adding sub-surface temperature data improves these decadal statistical predictions, and find that, again, it depends on the region, prediction lead time and GCM data used. Overall, we argue that the estimated prediction skill motivates the further development of statistical decadal predictions of SSTs as a benchmark for current and future GCM-based decadal climate predictions.

International evidence on the predictability of returns to securitized real estate assets: econometric models versus neural networks

The performance of various statistical models and commonly used financial indicators for forecasting securitised real estate returns are examined for five European countries: the UK, Belgium, the Netherlands, France and Italy. Within a VAR framework, it is demonstrated that the gilt-equity yield ratio is in most cases a better predictor of securitized returns than the term structure or the dividend yield. In particular, investors should consider in their real estate return models the predictability of the gilt-equity yield ratio in Belgium, the Netherlands and France, and the term structure of interest rates in France. Predictions obtained from the VAR and univariate time-series models are compared with the predictions of an artificial neural network model. It is found that, whilst no single model is universally superior across all series, accuracy measures and horizons considered, the neural network model is generally able to offer the most accurate predictions for 1-month horizons. For quarterly and half-yearly forecasts, the random walk with a drift is the most successful for the UK, Belgian and Dutch returns and the neural network for French and Italian returns. Although this study underscores market context and forecast horizon as parameters relevant to the choice of the forecast model, it strongly indicates that analysts should exploit the potential of neural networks and assess more fully their forecast performance against more traditional models.

A new modelling framework for statistical cumulus dynamics

We propose a new modelling framework suitable for the description of atmospheric convective systems as a collection of distinct plumes. The literature contains many examples of models for collections of plumes in which strong simplifying assumptions are made, a diagnostic dependence of convection on the large-scale environment and the limit of many plumes often being imposed from the outset. Some recent studies have sought to remove one or the other of those assumptions. The proposed framework removes both, and is explicitly time-dependent and stochastic in its basic character. The statistical dynamics of the plume collection are defined through simple probabilistic rules applied at the level of individual plumes, and van Kampen's system size expansion is then used to construct the macroscopic limit of the microscopic model. Through suitable choices of the microscopic rules, the model is shown to encompass previous studies in the appropriate limits, and to allow their natural extensions beyond those limits.

The statistical properties of tropical ice clouds generated by the West-African and Australian monsoons from ground-based radar-lidar observations

In the present paper we characterize the statistical properties of non-precipitating tropical ice clouds (deep ice anvils resulting from deep convection and cirrus clouds) over Niamey, Niger, West Africa, and Darwin, northern Australia, using ground-based radar–lidar observations from the Atmospheric Radiation Measurement (ARM) programme. The ice cloud properties analysed in this paper are the frequency of ice cloud occurrence, cloud fraction, the morphological properties (cloud-top height, base height, and thickness), the microphysical and radiative properties (ice water content, visible extinction, effective radius, terminal fall speed, and concentration), and the internal cloud dynamics (in-cloud vertical air velocity). The main highlight of the paper is that it characterizes for the first time the probability density functions of the tropical ice cloud properties, their vertical variability and their diurnal variability at the same time. This is particularly important over West Africa, since the ARM deployment in Niamey provides the first vertically resolved observations of non-precipitating ice clouds in this crucial area in terms of redistribution of water and energy in the troposphere. The comparison between the two sites also provides an additional observational basis for the evaluation of the parametrization of clouds in large-scale models, which should be able to reproduce both the statistical properties at each site and the differences between the two sites. The frequency of ice cloud occurrence is found to be much larger over Darwin when compared to Niamey, and with a much larger diurnal variability, which is well correlated with the diurnal cycle of deep convective activity. The diurnal cycle of the ice cloud occurrence over Niamey is also much less correlated with that of deep convective activity than over Darwin, probably owing to the fact that Niamey is further away from the deep convective sources of the region. The frequency distributions of cloud fraction are strongly bimodal and broadly similar over the two sites, with a predominance of clouds characterized either by a very small cloud fraction (less than 0.3) or a very large cloud fraction (larger than 0.9). The ice clouds over Darwin are also much thicker (by 1 km or more statistically) and are characterized by a much larger diurnal variability than ice clouds over Niamey. Ice clouds over Niamey are also characterized by smaller particle sizes and fall speeds but in much larger concentrations, thereby carrying more ice water and producing more visible extinction than the ice clouds over Darwin. It is also found that there is a much larger occurrence of downward in-cloud air motions less than 1 m s−1 over Darwin, which together with the larger fall speeds retrieved over Darwin indicates that the life cycle of ice clouds is probably shorter over Darwin than over Niamey.

Modeling data with structural and temporal correlation using lower level and higher level multilevel models

Novel imaging techniques are playing an increasingly important role in drug development, providing insight into the mechanism of action of new chemical entities. The data sets obtained by these methods can be large with complex inter-relationships, but the most appropriate statistical analysis for handling this data is often uncertain - precisely because of the exploratory nature of the way the data are collected. We present an example from a clinical trial using magnetic resonance imaging to assess changes in atherosclerotic plaques following treatment with a tool compound with established clinical benefit. We compared two specific approaches to handle the correlations due to physical location and repeated measurements: two-level and four-level multilevel models. The two methods identified similar structural variables, but higher level multilevel models had the advantage of explaining a greater proportion of variation, and the modeling assumptions appeared to be better satisfied.

Evaluation of ice cloud representation in the ECMWF and UK Met Office models using CloudSat and CALIPSO data

Ice cloud representation in general circulation models remains a challenging task, due to the lack of accurate observations and the complexity of microphysical processes. In this article, we evaluate the ice water content (IWC) and ice cloud fraction statistical distributions from the numerical weather prediction models of the European Centre for Medium-Range Weather Forecasts (ECMWF) and the UK Met Office, exploiting the synergy between the CloudSat radar and CALIPSO lidar. Using the last three weeks of July 2006, we analyse the global ice cloud occurrence as a function of temperature and latitude and show that the models capture the main geographical and temperature-dependent distributions, but overestimate the ice cloud occurrence in the Tropics in the temperature range from −60 °C to −20 °C and in the Antarctic for temperatures higher than −20 °C, but underestimate ice cloud occurrence at very low temperatures. A global statistical comparison of the occurrence of grid-box mean IWC at different temperatures shows that both the mean and range of IWC increases with increasing temperature. Globally, the models capture most of the IWC variability in the temperature range between −60 °C and −5 °C, and also reproduce the observed latitudinal dependencies in the IWC distribution due to different meteorological regimes. Two versions of the ECMWF model are assessed. The recent operational version with a diagnostic representation of precipitating snow and mixed-phase ice cloud fails to represent the IWC distribution in the −20 °C to 0 °C range, but a new version with prognostic variables for liquid water, ice and snow is much closer to the observed distribution. The comparison of models and observations provides a much-needed analysis of the vertical distribution of IWC across the globe, highlighting the ability of the models to reproduce much of the observed variability as well as the deficiencies where further improvements are required.

Response theory for equilibrium and non-equilibrium statistical mechanics: causality and generalized Kramers-Kronig relations

We consider the general response theory recently proposed by Ruelle for describing the impact of small perturbations to the non-equilibrium steady states resulting from Axiom A dynamical systems. We show that the causality of the response functions entails the possibility of writing a set of Kramers-Kronig (K-K) relations for the corresponding susceptibilities at all orders of nonlinearity. Nonetheless, only a special class of directly observable susceptibilities obey K-K relations. Specific results are provided for the case of arbitrary order harmonic response, which allows for a very comprehensive K-K analysis and the establishment of sum rules connecting the asymptotic behavior of the harmonic generation susceptibility to the short-time response of the perturbed system. These results set in a more general theoretical framework previous findings obtained for optical systems and simple mechanical models, and shed light on the very general impact of considering the principle of causality for testing self-consistency: the described dispersion relations constitute unavoidable benchmarks that any experimental and model generated dataset must obey. The theory exposed in the present paper is dual to the time-dependent theory of perturbations to equilibrium states and to non-equilibrium steady states, and has in principle similar range of applicability and limitations. In order to connect the equilibrium and the non equilibrium steady state case, we show how to rewrite the classical response theory by Kubo so that response functions formally identical to those proposed by Ruelle, apart from the measure involved in the phase space integration, are obtained. These results, taking into account the chaotic hypothesis by Gallavotti and Cohen, might be relevant in several fields, including climate research. In particular, whereas the fluctuation-dissipation theorem does not work for non-equilibrium systems, because of the non-equivalence between internal and external fluctuations, K-K relations might be robust tools for the definition of a self-consistent theory of climate change.

Downscaling from GCM precipitation: a benchmark for dynamical and statistical downscaling methods

A precipitation downscaling method is presented using precipitation from a general circulation model (GCM) as predictor. The method extends a previous method from monthly to daily temporal resolution. The simplest form of the method corrects for biases in wet-day frequency and intensity. A more sophisticated variant also takes account of flow-dependent biases in the GCM. The method is flexible and simple to implement. It is proposed here as a correction of GCM output for applications where sophisticated methods are not available, or as a benchmark for the evaluation of other downscaling methods. Applied to output from reanalyses (ECMWF, NCEP) in the region of the European Alps, the method is capable of reducing large biases in the precipitation frequency distribution, even for high quantiles. The two variants exhibit similar performances, but the ideal choice of method can depend on the GCM/reanalysis and it is recommended to test the methods in each case. Limitations of the method are found in small areas with unresolved topographic detail that influence higher-order statistics (e.g. high quantiles). When used as benchmark for three regional climate models (RCMs), the corrected reanalysis and the RCMs perform similarly in many regions, but the added value of the latter is evident for high quantiles in some small regions.