Presents a comparative study for prediction of time series of the Consumer Price Index-CPI using recurrent neural network (RNN). For this, three models are designed for networks with recurrent and are given the changes in "backpropagation" to allow them to incorporate the models ARX (Auto-Regressive with external input) and NARX (Nonlinear Auto Regressive with external input). Furthermore, we present a third architecture, re-fed with the hidden layer, nicknamed ARXI, which is a special case of the Elman Network. Is carried out training for all networks and tests the ability to generalize them (identification stage), in order to select the best architectures of recurrent networks to prediction of the IPC. After this stage, it makes the models validation, by means of the test the extrapolation capacity of the networks, i.e., presented data were not used during the training phase and gets the responses that indicate the capacity to predict future CPI for various times (validation phase). We conclude that NARX networks are those with best performance and that the hybrid system proposed by [5] constitutes an excellent tool when you want to get minimal networks that make a series of perdition satisfactorily.

A Multi-Layer Perceptron (MLP) defines a family of artificial neural networks often used in TS modeling and forecasting. Because of its “black box” aspect, many researchers refuse to use it. Moreover, the optimization (often based on the exhaustive approach where “all” configurations are tested) and learning phases of this artificial intelligence tool (often based on the Levenberg-Marquardt algorithm; LMA) are weaknesses of this approach (exhaustively and local minima). These two tasks must be repeated depending on the knowledge of each new problem studied, making the process, long, laborious and not systematically robust. In this short communication, a pruning process is presented. This method allows, during the training phase, to carry out an inputs selecting method activating (or not) inter-nodes connections in order to verify if forecasting is improved. We propose to use iteratively the popular damped least-squares method to activate inputs and neurons. A first pass is applied to 10% of the learning sample to determine weights significantly different from 0 and delete other. Then a classical batch process based on LMA is used with the new MLP. The validation is done using 25 measured meteorological TS and cross-comparing the prediction results of the classical LMA and the 2-stage LMA.

Foreign exchange rates are among the most important economic indices in the international monetary markets. Applying forecasting models for forecasting in exchange rate markets and assisting investment decision making has become more indispensable in business practices than ever before. For large multinational firms, which conduct substantial currency transfers in the course of business, being able to accurately forecast movements of currency exchange rates can result in substantial improvement in the overall profitability of the firm. However, the literature shows that predicting the exchange rate movements are largely unforecastable due to their high volatility and noise and still are a problematic task. Many researches in time series forecasting have argued that predictive performance improves in combined models, especially when the models in the ensemble are quite different. Hybrid techniques that decompose a time series into its linear and nonlinear components are one of the most popular hybrid models categories, which have been shown to be successful for single models. However, they have yielded mixed results in some situations in comparison with components models used separately; and hence, it is not wise to apply them blindly to any type of data. In this paper, an enhanced version of hybrid neural based models is proposed, incorporating the autoregressive integrated moving average (ARIMA) and artificial neural networks (ANNs) for financial time series forecasting. In proposed model, in contrast to the traditional hybrid ARIMA/ANNs, it can be guaranteed that the performance of the proposed model will not be worse than either of the components models used separately. In additional, empirical results in exchange rate forecasting indicate that the proposed model can be an effective way to improve forecasting accuracy achieved by traditional hybrid ARIMA/ANNs models. Therefore, it can be used as an appropriate alternative model for forecasting in exchange rate markets, especially when higher forecasting accuracy is needed.

Even though exponential smoothing (ES) is publicized as one of the most successful forecasting methods in the time series literature due to its simplicity, its accuracy can be affected by the initialization and optimization procedures followed. It also suffers from some fundamental problems that can be seen clearly when its weighting scheme is studied closely. Exponential smoothing fails to account for the amount of data points that can contribute to the forecast when assigning weights to historical data. ATA smoothing has been proposed as an alternative and is shown to perform better than ES when the accuracies are compared on empirical data. In this paper, the properties of ATA that make it stand out from ES models will be discussed by just comparing the simple versions of both models. Empirical performance of the two simple models will be compared based on popular error metrics.