Spelling suggestions: "subject:"cynamic factor analysis"" "subject:"clynamic factor analysis""
1 |
Dynamic Factor Analysis as a Methodology of Business Cycle ResearchKholodilin, Konstantin A. 23 April 2003 (has links)
El objetivo principal de la investigación emprendida en la presente tesis doctoral es elaborar una técnica de construcción de un indicador económico compuesto o un conjunto de dichos indicadores que, correspondiendo al concepto teorético del ciclo económico (comercial), permitirán detectar y predecir los puntos de giro del ciclo económico.Como el punto de partida hemos escogido la definición del ciclo económico propuesta por Burns y Mitchell (1946). Según nuestra opinión, el analisis dinámico factorial es el método idóneo para captar los puntos de giro del ciclo económico en el sentido de Burns y Mitchell. Por un lado, tiene en cuenta los movimientos comunes de varias series macroeconómicas que bajan y suben simultaneamente durante las fases de recesiones y expansiones, respectivamente. Por otro lado, refleja las asimetrías que existen entre las dos fases cíclicas, como, por ejemplo, las tasas de crecimiento y la volatilidad distintas durante las recesiones y expansiones. Ambos rasgos estan subrayados por Burns y Mitchell como características definitivas del ciclo económico.El análisis dinámico factorial en su estado actual exige sin duda ciertas modificaciones y algunas extensiones para obtener las estimaciones insesgadas y consistentes de los indicadores económicos compuestos y para utilizar la información disponible de la mejor manera posible.Nuestra investigación está dirigida, en primer lugar, hacia los economistas prácticos que han optado por utilizar el análisis dinámico factorial para la construcción del indicador del ciclo económico tanto a nivél regional como nacional.La tesis esta compuesta por cinco capítulos donde el primer y el último capítulos son, respectivamente, la introducción y la conclusión. En ellos se exponen los objetivos del estudio y los resultados alcanzados en el curso de la investigación.En el capítulo dos describimos varios metodos de análisis de las fluctuaciones económicas que han sido propuestos durante los últimos 20 años. Por un lado, consideramos los modelos con la dinámica nolineal, concretamente el cambio de regímenes o el Markov switching. Por otro lado, examinamos los modelos lineales del análisis dinámico factorial. Al final del capítulo analizamos el modelo del factor común latente con la dinámica nolineal (con cambios de regímenes) que está construido como una combinación de estos dos metodos principales.En el capítulo tres introducimos un modelo general dinámico multifactorial con la dinámica lineal y nolineal. Este modelo permite captar la dimensión intertemporal (indicador avanzado versus indicador coincidente) de los factores comunes inobservables. Se examinan dos modelos dinámicos alternativos con un factor común inobservable avanzado y un factor común inobservable coincidente. En el primer modelo el factor común coincidente esta influido por el factor común avanzado a través del mecanismo de causalidad de Granger. Mientras que en el segundo modelo los dos factores estan relacionados via la matríz de las probabilidades de transición. Debido a que el factor avanzado contiene información sobre los cambios futuros de las fases cíclicas, ambos modelos permiten hacer predicciones de los puntos de giro del ciclo económico.En el capítulo cuatro elaboramos las técnicas sumplementarias necesarias para resolver algunos problemas de datos que son bastante frecuentes en la actividad de un economista empírico. Los dos problemas más importantes son los cambios estructurales y la falta de observaciones, particularmente cuando los datos que estan disponibles con distintas frecuencias (por ejemplo: los datos mensuales y trimestrales). Estos problemas quiebran la continuidad de la serie temporal y reducen el número de observaciones válidas para el análisis estadístico. Se demuestra que estos problemas se resuelven modificando el modelo de análisis dinámico factorial, con lo que se obtienen estimaciones más eficientes de los parametros del modelo. / The main objective of our research undertaken in this thesis is to elaborate a technique of constructing a composite economic indicator or a set of such indicators which would correspond to the theoretical concept of business cycle and reflect a phenomenon which may be interpreted as the cyclical dynamics of the economy.As a point of departure we have chosen the definition of business cycle proposed by Burns and Mitchell (1946). We believe that the most appropriate method to capture the Burns and Mitchell's cycle would be the dynamic factor analysis.The dynamic factor analysis in its current state requires undoubtedly some refinements and extensions to obtain unbiased and consistent estimates of the composite economic indicators and to use the available information in the best possible way.Our research is mostly oriented towards the practitioners who have opted for using the dynamic factor approach in the construction of the business cycle indicator both at the regional and national levels.The thesis is comprised of five chapters where the first and the last chapters are the introduction and conclusion delineating the objectives of the study and summarizing the results achieved during research.Chapter two describes various approaches to the analysis of economic fluctuations proposed during the last 20 years. On the one hand, it concentrates on models with nonlinear, namely Markov-switching, dynamics, on the other hand, it is concerned with dynamic factor models. Finally, it shows the combined techniques which unify these two principal approaches, thus, modeling common latent factor with regime-switching dynamics.In chapter three we introduce a general multifactor dynamic model with linear and regime-switching dynamics. This model allows capturing the intertemporal (leading versus coincident) dimension of the latent common factors. Two alternative multifactor dynamic models with a leading and a coincident unobserved common factors are examined: a model where the common coincident factor is Granger-caused by the common leading factor and a model where the leading relationship is translated into a set of specific restrictions imposed on the transition probabilities matrix.Chapter four concentrates on the supplementary devices which allow to overcome some data problems which are very frequent in the practitioner's life. Among the most prominent are the structural breaks and missing observations. It is shown that some of these troubles can be coped with by modifying the dynamic common factors models, which leads to more efficient estimates of the parameters of the models.
|
2 |
Comunalidade na liquidez: evidências no mercado brasileiro / Commonality in liquidity: evidence in the brazilian marketCasarin, Fernando 01 August 2011 (has links)
This study aimed to verify the existence of commonality in liquidity in the Brazilian market by delivering common factors of liquidity with an innovative technique (dynamic factor analysis). Also sought to examine the relationship between commonality and return on individual assets. Most studies of commonality are proceeded with data analysis and worked out daily in developed markets like the United States (Chord, Roll and Subrahmanyam (2000) Huberman and Halka (1999), Hasbrouck and Seppi (2001), Henker and Martens (2003 ), Lee (2005) and Brockman, Chung and Perignon (2009)), but some use intraday data on the formation of the sample and, moreover, show the commonality in emerging markets. Brockman and Chung (2002), Zheng and Zhang (2006), and Giouvris Galariotis (2008) are examples of studies in these markets, using a variety of measures and different methodological approaches. There were no Brazilian studies involving the commonality, but a study of foreign Brockman, Chung and Perignon (2009) reported weak evidence in Brazil. The procedure adopted for estimating the dynamic factor analysis (DFA) was based on a study of Frederic (2006) using the software Stata version 11. This survey was conducted with the shares belonging to the Bovespa index (Bovespa) from intraday data every five minute interval in the period from January 4 until April 30, 2010, total assets of 63 theoretical portfolio of first quarter 2010. Due to the limitation of the software, the sample was divided into three groups (group 1, 2 and 3), each composed of 21 companies with 498 5 minute intervals during periods of 83 observations for each trading day, the day 01/04/2010 until 01/11/2010 generating a total of 10,458 observations for each group. Common factors were found from the liquidity variables, which explain in part the common variation in liquidity. After analyzing the factors we proceeded to estimate the regressions by group. For each group had three regressions, only the first return of Ibovespa regressing against the return of the asset. Then we included a factor for liquidity and, after all factors were included in the model.
Among the results of the regressions, the Group 1 stands out, presented the highest coefficient of determination and where the Bovespa index return and Factor 1 were significant, indicating that beyond the market beta the common factor in liquidity also produces impacts on return the company. This study showed that there is commonality in liquidity in the market and also that there is influence of liquidity in the return of individual assets, confirming the evidence found by Brockman, Chung and Perignon (2009). / O presente estudo teve como objetivo verificar a existência de comunalidade na liquidez no mercado brasileiro através da apresentação de fatores comuns de liquidez com uma técnica inovadora (análise fatorial dinâmica). Buscou ainda analisar a relação entre a comunalidade e o retorno dos ativos individuais. A maioria dos estudos de comunalidade são procedidos com análises de dados diários e trabalhados em mercados desenvolvidos como os Estados Unidos (Chordia, Roll e Subrahmanyam (2000) Huberman and Halka (1999), Hasbrouck and Seppi (2001), Henker e Martens (2003), Lee (2005) e Brockman, Chung e Pérignon (2009)), mas alguns utilizam dados intraday na formação da amostra e, além disso, evidenciam a comunalidade também nos mercados emergentes. Brockman and Chung (2002), Zheng e Zhang (2006), Giouvris e Galariotis (2008) são exemplos de estudos nesses mercados, usando uma variedade de medidas e diferentes abordagens metodológicas. Não foram encontradas pesquisas brasileiras envolvendo a comunalidade, mas um estudo estrangeiro de Brockman, Chung e Pérignon (2009) relatou evidências fracas no Brasil. O procedimento adotado para a estimação da análise fatorial dinâmica (AFD) foi baseado no estudo de Frederici (2006) utilizando o software Stata versão 11. Essa pesquisa foi realizada com as ações pertencentes ao índice Bovespa (Ibovespa) a partir de dados intraday a cada intervalo de cinco minutos no período de 04 de Janeiro até 30 de abril de 2010, totalizando 63 ativos da carteira teórica do primeiro quadrimestre de 2010. Devido à limitação do software a amostra foi dividida em três grupos (Grupo 1, 2 e 3), cada um composto por 21 empresas com 498 intervalos de 5 minutos em períodos de 83 observações para cada dia negociado, do dia 04/01/2010 até 11/01/2010 gerando um total de 10458 observações para cada um dos grupos. Foram encontrados fatores comuns a partir das variáveis de liquidez, nos quais explicam parte da variação comum da liquidez. Após a análise dos fatores procedeu-se a estimação das regressões por Grupo. Para cada Grupo foram geradas três regressões, a primeira somente do retorno do Ibovespa regredindo contra o retorno do ativo. Em seguida incluiu-se o fator 1 de liquidez e, após, todos os fatores foram incluídos no modelo. Dentre os resultados das regressões, destaca-se o Grupo 1, cujo modelo estimado apresentou o maior coeficiente de determinação e onde o retorno Ibovespa e o Fator 1 foram significativos, indicando que além do beta de mercado o fator comum da liquidez também produz impactos no retorno da empresa. Este estudo mostrou que existe comunalidade na liquidez no mercado brasileiro e, também, que há influência dos fatores de liquidez no retorno dos ativos individuais, corroborando com as evidências encontradas por Brockman, Chung e Pérignon (2009).
|
3 |
Using Primary Dynamic Factor Analysis on repeated cross-sectional surveys with binary responses / Primär Dynamisk Faktoranalys för upprepade tvärsnittsundersökningar med binära svarEdenheim, Arvid January 2020 (has links)
With the growing popularity of business analytics, companies experience an increasing need of reliable data. Although the availability of behavioural data showing what the consumers do has increased, the access to data showing consumer mentality, what the con- sumers actually think, remain heavily dependent on tracking surveys. This thesis inves- tigates the performance of a Dynamic Factor Model using respondent-level data gathered through repeated cross-sectional surveys. Through Monte Carlo simulations, the model was shown to improve the accuracy of brand tracking estimates by double digit percent- ages, or equivalently reducing the required amount of data by more than a factor 2, while maintaining the same level of accuracy. Furthermore, the study showed clear indications that even greater performance benefits are possible.
|
Page generated in 0.0776 seconds