C53 - Forecasting Models; Simulation MethodsReturn

We provide an extensive study assessing whether a prior wavelet-based denoising step enhances the forecast accuracy of standard forecasting models. Many combinations of attribute values of the thresholding (denoising) algorithm are explored together with several traditional forecasting models used in economic time series forecasting. The results are evaluated using M3 competition yearly time series. We conclude that the performance of a forecasting model combined with the prior denoising step is generally not recommended, which implies that a straightforward generalisation of some of the results available in the literature (which found the denoising step to be beneficial) is not possible. Even if cross-validation is used to select the value of the threshold, a superior performance of the forecasting model with the prior denoising step does not generally follow.

An interesting approach to economic and financial time series forecasting consists of decomposing an input time series additively into several components, each component capturing the dynamics of a different frequency range. Consequently, each component is modelled and forecasted separately, the predictions being summed up to form an overall forecast of the input time series. The present paper considers one very important aspect of the forecasting procedure. More specifically, it provides a better understanding of how an additive decomposition of the input time series into several components can be obtained using the wavelet transform and how boundary conditions in the individual components should be properly treated. Even though these aspects are presented as a part of the wavelet theory in several books on wavelets, their implementation is prone to misinterpretations in the literature on applied time series forecasting, possibly due to the complexity of the wavelet transform. Since our exposition is focused predominantly on these aspects, it provides a concise explanation which may be helpful to practitioners. The maximal overlap discrete wavelet transform is employed, other types of wavelet transforms also being briefly discussed.

Enormous development of firm valuation from many aspects can be seen in the recent period. One of the main fields is scoring, which provides a probability verdict about the future development of a firm: its probability of default. This article focuses on introducing macroeconomic modelling using VEC models to predict the future level of default in the Czech economy. Our results have proven a general connection between corporate defaults and the macroeconomic condition of the economy, which is going through a convergence process. The specific findings are new and have not been observed yet. A connection between the GDP and defaults revealed a positive relationship, which is probably a consequence of the convergence process, a development of the economy in many new fields. We have also found a long-term equilibrium among unemployment, loans, price of oil and defaults. We have revealed a higher level of defaults can be expected in 2013, which is connected with the economic contraction in the prediction period.

By modifying and generalizing the wavelet-based approach of approximately simulating univariate long-memory processes that is available in the literature, we propose a methodology for simulating a bivariate stationary process, whose components exhibit different relationships at different scales. We derive the formulas for the autocovariance and cross-covariance sequences of the simulated bivariate process. We provide a setting for the parameters of the simulation which might generate a bivariate time series resembling that of stock log returns. Using this setting, we study the properties of our methodology via Monte Carlo simulation.

In this paper, I propose the use of fast Fourier transform (FFT) as a convenient tool for combining forecast densities of vector autoregressive models in a hybrid Bayesian manner. While a vast amount of papers comprises combinations based on normal approximations, Monte Carlo methods were fully utilized here, which made the analysis computationally demanding. For the sake of minimization of computational time, the FFT algorithm was used to combine the densities of poorly simulated partial models. As a result, a minor loss of quality in the final combined model was allowed, in contrast with the reduction in the necessary simulation time. However, it turns out in the end that the FFT-based approach exceeds 'brute-force' simulation in all aspects. The suggested method is demonstrated on an ex ante prediction of the Czech GDP and on a pair of artificial examples.