|
|
Granger's causality in financial time series
Marčiny, Jakub ; Voříšek, Jan (advisor) ; Lachout, Petr (referee)
The bachelor thesis discusses causality in multiple time series. Granger causality, along with its more general counterparts instantaneous causality and multistep causality, are utilized to study the mutual influence of the individual components of a multiple time series. These concepts are investigated within the framework of vector autoregressive models VAR. After the introduction of basic definitions and facts, the construction of VAR model is described including methods for order selection and verification. Subsequently, causal relations within the model are examined. Finally, empirical analysis of real financial market data is performed using tests procedures programmed with computational software Mathematica.
|
|
|
Modelování spotových cen elektrické energie
Šmíd, Vítězslav ; Honzík, Petr (advisor) ; Hencl, Stanislav (referee)
We describe a single-period vector autoregressive model with parameter restrictions and find a consistent estimator of the parameters. We apply several restricted models to electricity prices in two markets. The datasets are comprised of the settlement prices of day-ahead auctions in which market participants bid on next day's electricity deliveries in 24 separate hourly blocks. We therefore model the data as a time series in R^24. To avoid overfitting we crossvalidate all models using sliding windows of training and testing data. We find that simple models perform better because they are more resilient in volatile periods than more comprehensive models. We suggest that model performance could be improved by incorporating exogenous data, such as weather conditions. Powered by TCPDF (www.tcpdf.org)
|
|
|
Modelování spotových cen elektrické energie
Šmíd, Vítězslav ; Honzík, Petr (advisor) ; Hencl, Stanislav (referee)
We describe a single-period vector autoregressive model with parameter restrictions and find a consistent estimator of the parameters. We apply several restricted models to electricity prices in two markets. The datasets are comprised of the settlement prices of day-ahead auctions in which market participants bid on next day's electricity deliveries in 24 separate hourly blocks. We therefore model the data as a time series in R^24. To avoid overfitting we crossvalidate all models using sliding windows of training and testing data. We find that simple models perform better because they are more resilient in volatile periods than more comprehensive models. We suggest that model performance could be improved by incorporating exogenous data, such as weather conditions. Powered by TCPDF (www.tcpdf.org)
|
|
|
Granger's causality in financial time series
Marčiny, Jakub ; Voříšek, Jan (advisor) ; Lachout, Petr (referee)
The bachelor thesis discusses causality in multiple time series. Granger causality, along with its more general counterparts instantaneous causality and multistep causality, are utilized to study the mutual influence of the individual components of a multiple time series. These concepts are investigated within the framework of vector autoregressive models VAR. After the introduction of basic definitions and facts, the construction of VAR model is described including methods for order selection and verification. Subsequently, causal relations within the model are examined. Finally, empirical analysis of real financial market data is performed using tests procedures programmed with computational software Mathematica.
|
guest ::
