Thesis Evaluación y tendencia de la contribución de modelos cuantitativos predictivos aplicados en el negocio financiero. Caso aplicado a un family office en Chile.
Loading...
Date
2014-04
Authors
Journal Title
Journal ISSN
Volume Title
Program
Ingeniería Civil Industrial
Departament
Campus
Campus Santiago Vitacura
Abstract
El objetivo del presente estudio se centró en evaluar la capacidad de modelos predictivos en el negocio financiero de un family office chileno.
El análisis se canalizó en testear la hipótesis de si la capacidad predictiva de los modelos fue efectiva y si esta se mantuvo para los últimos 10 años, separando dicho análisis en 3 tipos de modelos, de causalidad, de series de tiempo y de velas japonesas. Para el caso del modelo de causalidad se utilizó una regresión logarítmica que estimó niveles de tipo de cambio de equilibrio con proyección a 90 días, para el tipo de cambio local. En el caso de las series de tiempo se aplicó un modelo simple de un rezago para el peso chileno (clp), el oro, el real brasileño, el S&P500 (SPY) y el BCU5. Para caso del análisis de las velas japonesas se aplicó al peso chileno (clp), real brasileño, peso mexicano y S&P500 (SPY).
Los resultados encontrados para el caso del modelo de causalidad mostraron que este no fue rentable en el negocio de posiciones ya que para los últimos 10 años generó perdidas persistentes y cuando hubo utilidades estas sobrepasaban el plazo anual del presupuesto proyectado, además de poseer un significativo error estadístico. Por otra parte, los modelos de series de tiempo mostraron alta capacidad predictiva donde para los negocios de peso chileno, oro, real brasileño y bolsa americana (S&P500), los retornos esperados diarios se movieron entre un 0.18% y un 0.79%, con señales de compra y venta con una efectividad entre un 77% y un 99%. En el caso del BCU-5 el retorno esperado diario fue significativamente más bajo el cual se situó entre un 0.03% y un 0.06%. Las velas japonesas mostraron un excelente desempeño en el mercado de moneda local (peso chileno), real brasileño, peso mexicano y bolsa americana (S&P500), con retornos diarios esperados que se movieron entre 0.35% y 0.62% y con un % de efectividad que rondó entre un 82% y un 98%.
Dado los resultados anteriores, la administración del family office reafirmó la utilidad de aplicar periódicamente los modelos de series de tiempo en conjunto con las velas japonesas, con la certeza de que no hay evidencia de que estos modelos hayan perdido capacidad predictiva en los últimos 10 años.
The objective of this study focused on evaluating the ability of predictive models in the financial business of a Chilean family office. Analysis was channeled in testing the hypothesis of whether the predictive ability of the models was effective and whether this was maintained for the past 10 years, separating the analysis into 3 types of models, causality, time series and candlestick. In the case of a logarithmic model of causality levels regression estimated equilibrium exchange rate projection to 90 days for the local exchange rate was used. In the case of a simple time series model of a lag for the Chilean peso (CLP ), gold , Brazilian real, the S&P500(SPY) and BCU5 applied . For the case of Japanese candlestick analysis was applied to the chilean peso (CLP ), brazilian real, mexican peso and S&P500 (SPY ). The results for the case of causal model showed that this was not profitable positions in business since for the last 10 years generated persistent losses and profits when these were greater than the annual budget of the projected period besides having a significant error statistic. Moreover, models of time series showed high predictive ability, where for businesses Brazilian Real Chilean peso, gold and U.S. stock market (S&P500), the expected returns daily were between 0.18 % and 0.79% , with signs buying and selling with an effectiveness between 77 % and 99% . In the case of BCU -5 daily return was significantly lower than expected which was between 0.02 % and 0.03 %. The candlestick showed excellent performance in market local currency (chilean peso), brazilian real, mexican peso and american equity (S&P500), with expected daily returns that were between 0.35 % and 0.62 % and a % of effectiveness between 82 % and 98 %. Given the above results, the administration of the family office periodically reaffirmed the usefulness of applying time series models together with the candlestick, in the certainty that there is no evidence that these predictive models have lost in the last 10 years.
The objective of this study focused on evaluating the ability of predictive models in the financial business of a Chilean family office. Analysis was channeled in testing the hypothesis of whether the predictive ability of the models was effective and whether this was maintained for the past 10 years, separating the analysis into 3 types of models, causality, time series and candlestick. In the case of a logarithmic model of causality levels regression estimated equilibrium exchange rate projection to 90 days for the local exchange rate was used. In the case of a simple time series model of a lag for the Chilean peso (CLP ), gold , Brazilian real, the S&P500(SPY) and BCU5 applied . For the case of Japanese candlestick analysis was applied to the chilean peso (CLP ), brazilian real, mexican peso and S&P500 (SPY ). The results for the case of causal model showed that this was not profitable positions in business since for the last 10 years generated persistent losses and profits when these were greater than the annual budget of the projected period besides having a significant error statistic. Moreover, models of time series showed high predictive ability, where for businesses Brazilian Real Chilean peso, gold and U.S. stock market (S&P500), the expected returns daily were between 0.18 % and 0.79% , with signs buying and selling with an effectiveness between 77 % and 99% . In the case of BCU -5 daily return was significantly lower than expected which was between 0.02 % and 0.03 %. The candlestick showed excellent performance in market local currency (chilean peso), brazilian real, mexican peso and american equity (S&P500), with expected daily returns that were between 0.35 % and 0.62 % and a % of effectiveness between 82 % and 98 %. Given the above results, the administration of the family office periodically reaffirmed the usefulness of applying time series models together with the candlestick, in the certainty that there is no evidence that these predictive models have lost in the last 10 years.
Description
Keywords
Modelos predictivos, Family office, Modelos de series de tiempo, Planificación estratégica, Evaluación de inversión, Toma de decisiones
