Abstract: Time series of seasonal temperatures recorded in Alaska during the past eighty years are analyzed. A common practice to measure changes in the long-term pattern of temperature series is to fit a deterministic linear trend. A deterministic trend is not a realistic approach and poses some pitfalls from the statistical point of view. A statistical model to fit a latent time-varying level independent of the Pacific climate shift is proposed. The empirical distribution of temperature conditional on the phase of the Pacific Decadal Oscillation is obtained. The results reveal that the switch between the negative and the positive phase leads to differences in temperatures up to 4℃ in a given location and season. Differences across seasons and locations are detected. The effect of the Pacific climate shift is stronger in winter. An overall increase of temperatures is observed in the long term. The estimated trends are not constant but exhibit different patterns that vary in the sign and strength over the sample period. BibTeX | DOI
Abstract: The maximum entropy bootstrap is an algorithm that creates an ensemble for time series inference. Stationarity is not required and the ensemble satisfies the ergodic theorem and the central limit theorem. The meboot R package implements such algorithm. This document introduces the procedure and illustrates its scope by means of several guided applications. BibTeX | JSS
Abstract: This paper reviews the open source R language and environment for statistical computing and graphics. The paper stresses the potential usefulness for Asian universities, researchers and organizations concerned with data analysis. We provide some practical guidelines and information on useful resources when working with R. We also include some examples to illustrate the scope of R as an object-oriented language and introduce the basic syntax. BibTeX | DOI
Abstract: The Clark model is an unobserved components model consisting of a trend and a cycle components. The former is a random walk with stochastic drift where the effect of shocks is permanent, while the latter component is driven by a stationary process characterized by the transitory effect of shocks. In Clark's model the effect of shocks is symmetric throughout the phases of the cycle. Stylized facts reveal the presence of asymmetries where long and smooth expansion periods alternate with sharper and shorter recession periods. In order to account for these asymmetries, researches have drawn attention to models where changes of regime are modelled endogenously as a Markov process. In this thesis, we set up a general framework that encompasses reference models analyzed in the literature. We adopt Clark's structural time series model and extend it allowing for Markov switching regimes in the components and the parameters of the model. We illustrate the contribution from dynamic econometric models to the empirical analysis of gross domestic product time series. In a first stage, the analysis is carried out separately for a trend-cycle unobserved components model and a Markov switching regime model. Next, we conduct two applications to compare results based on exact and approximate maximum likelihood in the context of the Clark model with Markov switching regimes. Upon this framework, we analyze further empirical issues that have been addressed separately in the literature: measure the persistence of shocks and estimation of the contemporaneous correlation between components. Finally, the presence of two transitory components is discussed in a model with asymmetries in the trend. BibTeX | Document | Slides
Abstract: Basándose en la literatura existente, en este trabajo se propone una metodología para el estudio gráfico y analítico del componente estacional en una serie temporal. El objetivo del análisis es determinar si el componente estacional responde a un comportamiento determinista o estocástico. Se muestran un conjunto de aplicaciones con series de la CAPV y del Estado para las que se define un modelo estadístico que recoge las características observadas en el análisis de la serie. BibTeX | Document
Abstract:
This paper describes a parallel implementation of multiple linear regression models that are run
on a general-purpose Graphics Processing Unit.
Seasonal unit root test statistics are obtained from each fitted regression model.
The code has two main applications:
simulation of response surfaces for different combinations of parameters
and bootstrapping the test statistics.
Each of this applications provides a different method to compute
p-values of seasonal unit root tests,
for which available tabulated critical values are limited.
Computing intensive operations are implemented in the CUDA
platform and programming model.
An interface in the R language and environment
is provided for the computation of
Abstract: Despite some practical advantages of the EM algorithm, its use in the context of structural time series models has been limited due to the observed slow convergence. We propose an enhancement of the algorithm by incorporating information from derivative terms that are null in the original design. Simulation experiments show a notable improvement in the convergence of the algorithm, while keeping parameter estimates practically identical to those obtained with the original algorithm. BibTeX | Document
Abstract: We discuss several variations on the implementation of a maximum likelihood procedure to fit the basic structural time series model. We illustrate through an application that several issues that are often neglected by practitioners, such as the parameterization of the model or the concentration of a parameter, may have a significant effect on the results. Difficulties that practitioners sometimes find to replicate parameter estimates reported by others are often due to the omission of these issues. BibTeX | Document
Abstract: We analyse the effect of the Pacific Decadal Oscillation (PDO) on temperatures recorded in Fairbanks (Alaska). We perform a graphical analysis based on seasonal paths and conditional densities obtained through quantile regressions. The results show evidence in agreement with the models and patterns discussed in the literature of climatology. BibTeX | Document
Abstract: We fit an autoregressive time series model with time-varying parameters to study changes in seasonal means of temperature data observed in Fairbanks (Alaska, USA). Other studies have found an increase in temperature using annual or seasonal averages. We develop a statistical model that allows us to have a further insight in this issue using monthly data. The results show that the overall increase in temperature is driven mainly by warmer winter months. The increase in temperature is more prominent in winter months. A decrease in temperature is found in October. BibTeX | Document
Abstract: One of the steps in the Gibbs sampler for the estimation of state-space models requires drawing values for the unobserved state vector. A common approach for this step is the usage of the simulation smoother conditioned to the data. In this document we discuss another approach based on resampling techniques. In particular, we explore Vinod's maximum entropy bootstrap as a technique for drawing the state vector. BibTeX | Document | Animated graphic
Abstract: This document focuses on practical issues showing the use of the partsm R-package. The package allows the user to check for periodicity in the data, fit a periodic autoregressive model of order p, PAR(p), select the periodic autoregressive lag order parameter, test for periodic integration, fit a periodically integrated autoregressive model up to order 2, PIAR, as well as to perform out-of-sample forecasts.
Note: The partsm package was developed as a pedagogical exercise while reading the book Periodicity and Stochastic Trends in Economic Time Series by P.H. Franses. The package replicates some of the results shown in the first eight chapters of the book.
Acknowledgements: Many thanks to Matthieu Stigler for maintaining the package active on CRAN. BibTeX | Document | Package source: CRAN