This thesis discusses a number of non-linear threshold models falling into the Threshold Autoregressive (TAR) and Smooth Transition Autoregressive (STAR) categories, characterised by regime changes when a certain threshold is exceeded. These models find application in many fields, particularly in the analysis of economic and financial time series characterised by non-linear dynamics. BDS and McLeod-Li tests play an important role in detecting the presence of non-linear structures that often remain latent. These tests also support the correct specification of models. The thesis describes the inference procedure for TAR and STAR models using the profile conditional likelihood function, and its asymptotic properties are presented along with a brief study of confidence intervals. Estimations will be made both on simulated data, in order to understand the properties of the models, and on real data, such as exchange rates; the aim is to understand the influence of political-economic events on the time series and their dynamics over the years. Finally, predictions will be made for the evaluation of some of the estimated models in which the mean square prediction error is used to assess their performance.
In questa tesi vengono discussi alcuni modelli non-lineari a soglia rientranti nelle categorie Threshold Autoregressive (TAR) e Smooth Transition Autoregressive (STAR), caratterizzati da cambi di regime al superamento di una certa soglia. Tali modelli trovano applicazione in molti campi, in particolare nell'analisi delle serie storiche economiche e finanziarie caratterizzate da una dinamica non lineare. Un ruolo importante viene ricoperto dai test BDS e McLeod-Li per rilevare la presenza di strutture non-lineari che spesso rimangono latenti. Questi test sono di supporto anche per la corretta specificazione dei modelli. La tesi descrive la procedura di inferenza per i modelli TAR e STAR tramite la profile conditional likelihood function e vengono presentate le relative proprietà asintotiche assieme a un breve studio sugli intervalli di confidenza. Le stime avranno per oggetto sia dati simulati, per comprendere le proprietà dei modelli, sia su dati reali, come i tassi di cambio; lo scopo è di comprendere l'influenza di eventi di natura politico-economica sulle serie storiche e la loro dinamica nel corso degli anni. Infine si effettueranno delle previsioni per la valutazione di alcuni dei modelli stimati in cui si utilizza l'errore quadratico medio previsionale per valutarne le performance.
Modelli a soglia e loro applicazioni ai tassi di cambio
FOSSALUZZA, LUCA
2024/2025
Abstract
This thesis discusses a number of non-linear threshold models falling into the Threshold Autoregressive (TAR) and Smooth Transition Autoregressive (STAR) categories, characterised by regime changes when a certain threshold is exceeded. These models find application in many fields, particularly in the analysis of economic and financial time series characterised by non-linear dynamics. BDS and McLeod-Li tests play an important role in detecting the presence of non-linear structures that often remain latent. These tests also support the correct specification of models. The thesis describes the inference procedure for TAR and STAR models using the profile conditional likelihood function, and its asymptotic properties are presented along with a brief study of confidence intervals. Estimations will be made both on simulated data, in order to understand the properties of the models, and on real data, such as exchange rates; the aim is to understand the influence of political-economic events on the time series and their dynamics over the years. Finally, predictions will be made for the evaluation of some of the estimated models in which the mean square prediction error is used to assess their performance.File | Dimensione | Formato | |
---|---|---|---|
Fossaluzza_Tesi.pdf
accesso aperto
Dimensione
1.11 MB
Formato
Adobe PDF
|
1.11 MB | Adobe PDF | Visualizza/Apri |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14247/25665