Search

Little attention has been devoted to the long memory among the different data features considered for clustering time series. Following previous literature, we measure the long memory of a time series through the estimated Hurst exponent. However, we exploit the fact that a constant value for the Hurst exponent h is unrealistic in many practical examples. In this paper, assuming that the time series follows a multifractional Brownian motion, we estimate a time-varying Hurst exponent used as the input for a fuzzy clustering procedure. Motivated by the relevance of long memory in finance, the usefulness of the proposed clustering procedure is shown with an application to stock prices. [ABSTRACT FROM AUTHOR]

Although there are many contributions in the time series clustering literature, few studies still deal with count time series data. This paper aims to develop a fuzzy clustering procedure for count time series data. We propose an Integer GARCH-based Fuzzy C-medoids (INGARCH-FCMd) method for clustering count time series based on a Mahalanobis distance between the parameters estimated by an INGARCH model. We show how the proposed clustering method works by clustering football teams according to the number of scored goals.

This paper develops a new time series clustering procedure allowing for heteroskedasticity, non-normality and model's non-linearity. At this aim, we follow a fuzzy approach. Specifically, considering a Dynamic Conditional Score (DCS) model, we propose to cluster time series according to their estimated conditional moments via the Autocorrelation-based fuzzy C -means (A-FCM) algorithm. The DCS parametric modeling is appealing because of its generality and computational feasibility. The usefulness of the proposed procedure is illustrated using an experiment with simulated data and several empirical applications with financial time series assuming both linear and nonlinear models' specification and under several assumptions about time series density function. [ABSTRACT FROM AUTHOR]

Recently, cryptocurrencies have attracted a growing interest from investors, practitioners and researchers. Nevertheless, few studies have focused on the predictability of them. In this paper we propose a new and comprehensive study about cryptocurrency market, evaluating the forecasting performance for three of the most important cryptocurrencies (Bitcoin, Ethereum and Litecoin) in terms of market capitalization. At this aim, we consider non-Gaussian GARCH volatility models, which form a class of stochastic recursive systems commonly adopted for financial predictions. Results show that the best specification and forecasting accuracy are achieved under the Skewed Generalized Error Distribution when Bitcoin/USD and Litecoin/USD exchange rates are considered, while the best performances are obtained for skewed Distribution in the case of Ethereum/USD exchange rate. The obtain findings state the effectiveness – in terms of prediction performance – of relaxing the normality assumption and considering skewed distributions. [ABSTRACT FROM AUTHOR]