Tesi etd-10062024-215155
Link copiato negli appunti
Tipo di tesi
Corso Ordinario Secondo Livello
Autore
SAMMARINI, ANITA
URN
etd-10062024-215155
Titolo
A Monte Carlo Analysis of VARs and Local Projections in a High-Dimensional Heteroskedastic Setting
Struttura
Classe Scienze Sociali
Corso di studi
SCIENZE ECONOMICHE E MANAGERIALI - SCIENZE ECONOMICHE E MANAGERIALI
Commissione
Tutor Prof. ROVENTINI, ANDREA
Relatore Prof. RAGUSA, GIUSEPPE
Presidente Prof. IRALDO, FABIO
Membro Dott.ssa CANTARELLI, PAOLA
Membro Prof. BARONTINI, ROBERTO
Membro Prof. BOTTAZZI, GIULIO
Membro Prof. CINQUINI, LINO
Membro Prof. TENUCCI, ANDREA
Membro Prof.ssa VAINIERI, MILENA
Membro Prof. MINA, ANDREA
Membro Prof. MONETA, ALESSIO
Membro Prof. TURCHETTI, GIUSEPPE
Membro Dott. GIACHINI, DANIELE
Membro Prof. DI MININ, ALBERTO
Relatore Prof. RAGUSA, GIUSEPPE
Presidente Prof. IRALDO, FABIO
Membro Dott.ssa CANTARELLI, PAOLA
Membro Prof. BARONTINI, ROBERTO
Membro Prof. BOTTAZZI, GIULIO
Membro Prof. CINQUINI, LINO
Membro Prof. TENUCCI, ANDREA
Membro Prof.ssa VAINIERI, MILENA
Membro Prof. MINA, ANDREA
Membro Prof. MONETA, ALESSIO
Membro Prof. TURCHETTI, GIUSEPPE
Membro Dott. GIACHINI, DANIELE
Membro Prof. DI MININ, ALBERTO
Parole chiave
- VAR
- LP
- heteroskedasticity
- time series
Data inizio appello
25/11/2024;
Disponibilità
completa
Riassunto analitico
While LPs and VARs are known to asymptotically estimate the same impulse re-
sponse functions, their relative performance in small sample setting lacks in-depth
exploration and understanding. To address this, the present study is a comparative
analysis of the two methodologies based on Montecarlo simulations characterized by
data generating processes varying in the nature of the variance-covariance matrix of
residuals and sample size. Importantly, this works aims at assessing the influence
of conditional heteroskedasticity in time series data on the estimation of structural
impulse responses accounting for GARCH residuals in DGPs. The findings reveal
that LPs demonstrate superior robustness in drastically underspecified models, as
evidenced by higher coverage rates, despite exhibiting larger average lengths and
variability. In cases of overspecification with a large number of lags, both VARs and
LPs perform comparably, although LPs show slight disadvantages due to their data-
intensive nature. Small sample settings affect both methods, leading to decreased
coverage rates, increased biases, and standard deviations. Notably, LPs are more
sensitive to sample size variations, particularly evident in LP(24). Furthermore,
results concerning the impact of conditional heteroskedasticity on IRF estimation
reveal that when the GARCH process assigns greater weight to past squared resid-
uals, both models experience substantial shifts in their profiles, with LPs displaying
more sensitivity to this condition. However, when past conditional variances receive
greater weight, the performance of both models is comparable to the case with in-
dependent and identically distributed residuals. This research contributes to the
understanding of IRF estimation methodologies in small sample settings, shedding
light on their relative strengths and weaknesses.
sponse functions, their relative performance in small sample setting lacks in-depth
exploration and understanding. To address this, the present study is a comparative
analysis of the two methodologies based on Montecarlo simulations characterized by
data generating processes varying in the nature of the variance-covariance matrix of
residuals and sample size. Importantly, this works aims at assessing the influence
of conditional heteroskedasticity in time series data on the estimation of structural
impulse responses accounting for GARCH residuals in DGPs. The findings reveal
that LPs demonstrate superior robustness in drastically underspecified models, as
evidenced by higher coverage rates, despite exhibiting larger average lengths and
variability. In cases of overspecification with a large number of lags, both VARs and
LPs perform comparably, although LPs show slight disadvantages due to their data-
intensive nature. Small sample settings affect both methods, leading to decreased
coverage rates, increased biases, and standard deviations. Notably, LPs are more
sensitive to sample size variations, particularly evident in LP(24). Furthermore,
results concerning the impact of conditional heteroskedasticity on IRF estimation
reveal that when the GARCH process assigns greater weight to past squared resid-
uals, both models experience substantial shifts in their profiles, with LPs displaying
more sensitivity to this condition. However, when past conditional variances receive
greater weight, the performance of both models is comparable to the case with in-
dependent and identically distributed residuals. This research contributes to the
understanding of IRF estimation methodologies in small sample settings, shedding
light on their relative strengths and weaknesses.
File
| Nome file | Dimensione |
|---|---|
| Tesi.pdf | 8.27 Mb |
Contatta l'autore |
|