Statistical Model Building for Neural Networks - Proceedings AFIR 1996 - Nürnberg, Germany
نویسنده
چکیده
Neural networks are a new, very flexible class of statistical and if applied to economic data econometric models. Basically, neural networks are a generalization of nonlinear regression models and can therefore be applied to all kinds of regression problems. Since neural networks do not require the specification of a certain structural form, they are particularly suited for modelling very complex functions as observed on the capital markets. Since neural networks are statistical procedures, they offer the opportunity to perform a thorough statistical analysis in order to build an adequate model. However, this is not the status-quo. Most network practitioners pursue a rather heuristic approach for determining a network architecture which is prone to finding only suboptimal network models, It is the aim of this article to propose a structurized process for modelling neural network architectures, which relies on statistical methods. R & W 6 Les r6seaux neuronaux forment une classe nouvelle et flexible de modbles statistiques et 6conomCtriques. Fondamentalement, les reseaux neuronaux constituent une g6ntralisation des modkles de rtgression non lineaire et peuvent s’appliquer h tout type de problkme de regression. Comme les rkseaux neuronaux ne requibrent la sptkification d’une certaine forme structurelle, ils conviennent particulibrement h la modtlisation de relations complexes telles que celles observks sur les marches financiers. Les rtseaux neuronaux ttant proctdure statistique, il convient d’utiliser cette opportunitk lors de leur modelisation. Toutefois aucun consensus n’Cmerge dans ce domaine. La plupart des praticiens des rtseaux neuronaux adoptent une demarche plutdt heuristique pour dtterminer I’architecture d’un rtseau, ce qui conduit au choix d’un modble non optimal. Le but de cet article est de proposer une procMure structurke pour modtliser l’architecture des rtseaux neuronaux tout en se basant sur des mtthodes statistiques.
منابع مشابه
Forecasting the Exchange Rate: A Comparison Between Econometric and Neural Network Models - Proceedings AFIR 1996 - Nürnberg, Germany
In this paper the performance of four linear models of the exchange rate Spanish peseta/US dollar is compared with that of Artificial Neural Networks. The models are a random walk process and three different specifications based on the purchasing power parity (PPP) theory. The aim is to examine whether potentially highly nonlinear neural network models outperform traditional methods or give at ...
متن کاملPricing of Equity-linked Life Insurance Policies with an Asset Value Guarantee and Periodic Premiums - Proceedings AFIR 1996 - Nürnberg, Germany
In the present paper we establish a quasi-explicit formula for the periodic premium under an equity-linked endowment policy with asset value guarantee in an economy with interest rate risk. R&llml? Le pr6sent article examine 1 'kvaluation d 'une prime periodique d 'assurance pour le produit unit-linked qui contient complCmentairement un rendement garanti. Nous proposons une solution de type qua...
متن کاملTactical Asset Allocation: Predictability of Capital Markets Using Error Correction Models - Proceedings AFIR 1996 - Nürnberg, Germany
How to optimize returns of an international equity or bond portfolio? Which bets should we make between bonds and stocks on a domestic balanced portfolio? Access to capital markets has become increasingly easier for investors. In this context, the Tactical Asset Allocation (TAA), which refers to how a portfolio's funds would be allocated, given the investor's short-term forecasts, is an essenti...
متن کاملOption Pricing Using EGARCH Models - Proceedings AFIR 1996 - Nürnberg, Germany
Various e m p i r i d studies have shown that the time-varying volatility of asset returns can be described by GARCH (generalized autoregressive conditional heteroskedasticity) models. The corresponding GARCH option pricing model of Duan (1995) is capable of depicting the "smile-effect" which often can be found in option prices. In some derivative markets, however, the slope of the smile is not...
متن کاملIntegration of Deep Learning Algorithms and Bilateral Filters with the Purpose of Building Extraction from Mono Optical Aerial Imagery
The problem of extracting the building from mono optical aerial imagery with high spatial resolution is always considered as an important challenge to prepare the maps. The goal of the current research is to take advantage of the semantic segmentation of mono optical aerial imagery to extract the building which is realized based on the combination of deep convolutional neural networks (DCNN) an...
متن کامل