We consider nonparametric generalization of various well-known financial time series models and study estimates of the trend and volatility functions and forecasts based on kernel smoothers as well as on neural networks.
In this paper we deal with the problem of fitting an autoregression of order p to given data coming from a stationary autoregressive process with infinite order. The paper is mainly concerned with the selection of an appropriate order of the autoregressive model. Based on the so-called final prediction error (FPE) a bootstrap order selection can be proposed, because it turns out that one relevant expression occuring in the FPE is ready for the application of the bootstrap principle. Some asymptotic properties of the bootstrap order selection are proved. To carry through the bootstrap procedure an autoregression with increasing but non-stochastic order is fitted to the given data. The paper is concluded by some simulations.
Neural networks are now a well-established tool for solving classification and forecasting problems in financial applications (compare, e.g., Bol et al., 1996, Evans, 1997, Rehkugler and Zimmermann, 1994, Refenes 1995, and Refenes et al. 1996a) though many practioners are still suspicious against too evident success stories. One reason may be that the construction of an appropriate network which provides a reasonable solution to a complex data-analytic problem is rarely made explicit in the literature. In this paper, we try to contribute to filling this gap by discussing in detail the problem of dynamically allocating capital to various components of a currency portfolio in such a manner that the average gain will be larger than for certain benchmark portfolios. We base our solution on feedforward neural networks which are constructed employing various statistical model selection procedures described in, e.g., (Anders, 1997, or Refenes et al., 1996b). Neural networks which are used as the basis of trading strategies in finance should be assessed differently than in technical applications. The task is not to construct a network which provides good forecasts with respect to mean-square error of some quantities of interest or to provide good approximation of some given target values, but to achieve a good performance in economic terms. For portfolio allocation, the main goal is to achieve on the average a large return combined with a small risk. Therefore, we do not consider forecasts of the foreign exchange (FX-) rate time series using neural networks, but we try to get the allocation directly as the output of a network. Furthermore, we do not minimize some estimation or prediction error, but we try to maximize an economically meaningful performance measure, the risk-adjusted return, directly (compare also Heitkamp, 1996). In the subsequent chapter, we describe the details of the portfolio allocation problem. The following two chapters provide some technical information on how the networks were fitted to the available data and how the network inputs and outputs were selected. In chapter 5, finally, we discuss the promising results.
Wicksell's corpuscle problem deals with the estimation of the size distribution of a population of particles, all having the same shape, using a lower imensional sampling probe. This problem was originary formulated for particle systems occurring in life sciences but its solution is of actual and increasing interest in materials science. From a mathematical point of view, Wicksell's problem is an inverse problem where the interesting size distribution is the unknown part of a Volterra equation. The problem is often regarded ill-posed, because the structure of the integrand implies unstable numerical solutions. The accuracy of the numerical solutions is considered here using the condition number, which allows to compare different numerical methods with different (equidistant) class sizes and which indicates, as one result, that a finite section thickness of the probe reduces the numerical problems. Furthermore, the relative error of estimation is computed which can be split into two parts. One part consists of the relative discretization error that increases for increasing class size, and the second part is related to the relative statistical error which increases with decreasing class size. For both parts, upper bounds can be given and the sum of them indicates an optimal class width depending on some specific constants.
It is well-known that some of the classical location problems with polyhedral gauges can be solved in polynomial time by finding a finite dominating set, i.e. a finite set of candidates guaranteed to contain at least one optimal location. In this paper it is first established that this result holds for a much larger class of problems than currently considered in the literature. The model for which this result can be proven includes, for instance, location problems with attraction and repulsion, and location-allocation problems. Next, it is shown that the approximation of general gauges by polyhedral ones in the objective function of our general model can be analyzed with regard to the subsequent error in the optimal objective value. For the approximation problem two different approaches are described, the sandwich procedure and the greedy algorithm. Both of these approaches lead - for fixed epsilon - to polynomial approximation algorithms with accuracy epsilon for solving the general model considered in this paper.
FeNi/FeMn exchange bias samples with a large exchange bias field at room temperature have been prepared on a Cu buffer layer. Upon irradiation with He ions, both the exchange bias field and the coercive field are modified. For low ion doses the exchange bias field is enhanced by nearly a factor of 2. Above a threshold dose of 0.3olsi 10 15 ions/cm 2 , the exchange bias field decreases continuously as the ion dose increases. The ob-served modifications are explained in terms of defect creation acting as pinning sites for domain walls and atomic intermixing.
For the next generation of high data rate magnetic recording above 1 Gbit/s, a better understanding of the switching processes for both recording heads and media will be required. In order to maximize the switch-ing speed for such devices, the magnetization precession after the magnetic field pulse termination needs to be suppressed to a maximum degree. It is demonstrated experimentally for ferrite films that the appropriate adjustment of the field pulse parameters and/or the static applied field may lead to a full suppression of the magnetization precession immediately upon termination of the field pulse. The suppression is explained by taking into account the actual direction of the magnetization with respect to the static field direction at the pulse termination.
We report on an unexpected suppression of the magnetocrystalline anisotropy contribution in epitaxial fcc Co(110) films on Cu(110) below a thickness of dc=(50 +/- 10) Å. For film thicknesses larger than dc the measured anisotropy value agrees with published data. Measurements on films with reduced strain indicate a large strain dependence of dc . A model calculation based on a crystal-field formalism and discussed within the context of band theory, which explicitly takes tetragonal misfit strains into account, reproduces the experimen-tally observed anomalies. Our results indicate that the usually applied phenomenological description of anisotropies, assuming additive free energy terms for each anisotropy contribution, fails in this case.