This dissertation presents panel structure models. Panel structure models are econometric models designed to explore clustering structure in large panel data and to estimate group-specific regression parameters. The main reference of this research is "Estimation of Panel Data Models with Parameter Heterogeneity when Group Membership is Unknown", a paper of Lin C.C and Ng S. published in 2012 on the Journal of Econometric Methods. The paper proposes news models to deal with unobserved heterogeneity in which individuals form a number of homogeneous groups in a heterogeneous population. Said differently, the regression coefficients are different across different groups but are the same within each groups. The first model proposed , called Two-step Pseudo Threshold Approach, uses the parameter estimates to form the threshold variables and the unknown threshold value to form groups, which similarity is defined in terms of slope coefficients. The second method, called Conditional K-means, assigns units to clusters based on the deviation between the individual and the conditional mean of the group, therefore the measure of similarity is defined in terms of conditional mean. To determine the optimal number of groups, two different procedures were used: the sequential homogeneity test of Pesaran and Yamagata (2008) and a modified BIC criterion. Finally, these models were applied to the study of the American stock market, taking as a representative sample all the constituents of the Standard & Poor 500 index over a period of 10 years, from October 2007 to June 2017.
Questa tesi ha lo scopo di presentare i modelli chiamati Panel Structure Models. Questi ultimi sono modelli econometrici progettati per esplorare la struttura di gruppo in grandi dati panel e per stimare i parametri di regressione specifici del gruppo. Il riferimento principale di questa ricerca è " Estimation of Panel Data Models with Parameter Heterogeneity when Group Membership is Unknown ", un paper di Lin C.C e Ng S. pubblicato nel 2012 sul Journal of Econometric Methods. Il paper propone nuovi modelli per controllare l'eterogeneità non osservata, in cui gli individui formano un numero di gruppi omogenei in una popolazione eterogenea. Detto diversamente, i coefficienti di regressione sono diversi tra i diversi gruppi ma sono uguali all'interno di ciascun gruppo. Il primo modello proposto, chiamato Two-step Pseudo Threshold Approach, utilizza le stime dei parametri per determinare le variabili soglia e il cut-off sulla base delle quali saranno determinati i gruppi. Perciò, in questo approccio la similarità è definita in termini di coefficienti di pendenza. Il secondo metodo, chiamato Conditional K-means, assegna le unità ai cluster in base alla deviazione tra l'individuo e la media condizionata del gruppo, quindi la misura della similarità è definita in termini di media condizionata. Per determinare il numero ottimale di gruppi, sono state utilizzate due diverse procedure: il test di omogeneità sequenziale di Pesaran e Yamagata (2008) e una forma modificata del Criterio di Informazione Bayesiano (BIC). Infine, questi modelli sono stati applicati allo studio del mercato azionario americano, prendendo come campione rappresentativo tutti i titoli dell'indice Standard & Poor 500 su un periodo di 10 anni, da ottobre 2007 a giugno 2017.
Panel Structure Models for Group Specific Estimation: An Empirical Application on US Stock Market
SCARONE, MARTINA
2016/2017
Abstract
This dissertation presents panel structure models. Panel structure models are econometric models designed to explore clustering structure in large panel data and to estimate group-specific regression parameters. The main reference of this research is "Estimation of Panel Data Models with Parameter Heterogeneity when Group Membership is Unknown", a paper of Lin C.C and Ng S. published in 2012 on the Journal of Econometric Methods. The paper proposes news models to deal with unobserved heterogeneity in which individuals form a number of homogeneous groups in a heterogeneous population. Said differently, the regression coefficients are different across different groups but are the same within each groups. The first model proposed , called Two-step Pseudo Threshold Approach, uses the parameter estimates to form the threshold variables and the unknown threshold value to form groups, which similarity is defined in terms of slope coefficients. The second method, called Conditional K-means, assigns units to clusters based on the deviation between the individual and the conditional mean of the group, therefore the measure of similarity is defined in terms of conditional mean. To determine the optimal number of groups, two different procedures were used: the sequential homogeneity test of Pesaran and Yamagata (2008) and a modified BIC criterion. Finally, these models were applied to the study of the American stock market, taking as a representative sample all the constituents of the Standard & Poor 500 index over a period of 10 years, from October 2007 to June 2017.È consentito all'utente scaricare e condividere i documenti disponibili a testo pieno in UNITESI UNIPV nel rispetto della licenza Creative Commons del tipo CC BY NC ND.
Per maggiori informazioni e per verifiche sull'eventuale disponibilità del file scrivere a: unitesi@unipv.it.
https://hdl.handle.net/20.500.14239/9997