The following essay includes three original chapters. Chapter 1 provides an early warning method for tactical asset allocation to deal with volatility clustering and fat-tail distributions of financial returns. The early warning signals are given by a two-state Markov switching model with high-volatility and low-volatility states and time-varying transition probabilities. The transition is driven by exogenous factors namely early-warnings. Using Bayesian inference and Gibbs Sampling for posterior approximation, I apply the model to daily returns of the S&P500 Index sampled from December 2000 up to the present. The paper suggests that a measure of the gap between implied volatility and realised volatility has important leading properties in forecasting switches between volatility regimes. The paper argues that trading strategies need to be calibrated on the predominant volatility state. Therefore, trend-following signals deliver significant gains in low volatility, while short-term reversal strategies yield positive performance in the high volatility regime. In Chapter 2, a data-enriched term-structure model for interest rates is presented following a general term-structure architecture based on non-arbitrage arguments. According to the Vasicek's model, the yield curve is completely driven by the unobserved instantaneous spot rate. Nevertheless, many alternative sources of information can be used to estimate it. Our state-space model includes the traditional Vasicek's measurement equation with the spot rate as a state process and it is augmented with further measurement equations on bonds at different maturities. In order to make the evaluation of the yield curve more reactive to market events, the model incorporates additional exogenous variables related to monetary policy, equity market volatility and macroeconomic fundamentals, such as inflation and output gap. All the measurement equations are driven by the state process and the augmentation allows for improving estimation of the latent spot rate. We propose a Bayesian framework for model inference based on an efficient posterior approximation procedure. The computational efficiency follows from the linearity of the model, which allows for Kalman filtering and smoothing recursions. The paper presents an innovative and interpretable way to estimate the risk premium as a combination of explicit sources of market risk. The model is applied to the US yield curve sampled at a monthly frequency from December 2000 up to the present. The augmentation and Bayesian inference allow for generating projections of the yield curve in line with different market scenarios. Such a simulation provides great support to decisions of tactical asset allocation since it allows for evaluating the coherence of the portfolio allocation with the predominant regime implied in the market yield curve. Chapter 3 presents a flexible way for modelling volatility spillovers. Popular extensions of heterogeneous autoregressive models have been proposed for describing realized volatility. However, they leave substantial information unmodelled in residuals. This paper proposes a methodology for dynamic modelling and forecasting realized variance vectors based on a Time-Varying Bayesian Markov Switching VAR Model. The approach allows for flexible dependence patterns and investigates spillover effects across volatilities of different asset classes. The application of the model to a global portfolio investing in bonds, equities and currencies provides that spillovers follow a regime-based representation, with state transitions driven by the 3-month US T-Bill rate, whose dynamics is strictly connected to market expectations about future monetary policy decisions. Arising from estimated regime-based volatility spillovers, a dynamic trading algorithm on the S&P500 Index is provided.
Il seguente progetto include 3 capitoli. Il Capitolo 1 introduce un modello di allocazione tattica di portafoglio delineato al fine di gestire la propensione dei rendimenti a manifestare una volatilità non costante nel tempo ed un comportamento probabilistico poco compatibile con una distribuzione gaussiana. I segnali di trading derivano da un modello Markov Switching caratterizzato da regimi di alta e bassa volatilità con probabilità di transizione stocastiche. La transizione tra i 2 stati non è pertanto constante ma è guidata da fattori esogeni che manifestano importanti proprietà previsionali in merito ai potenziali cambi di regime. Il paper prevede l'impiego di metodologie Bayesiane e dell'algoritmo di Gibbs Sampling per la stima e l'approssimazione delle distribuzioni a posteriori dei parametri. L'applicazione del modello ai rendimenti giornalieri dell'indice S&P500 dal 2000 ad oggi dimostra come il differenziale tra la volatilità implicita nel prezzo delle opzioni e quella realizzata abbia importanti proprietà previsionali nell'anticipare cambi di regime di volatilità. Alla luce di tale risultato, il paper argomenta come le strategie di trading debbano essere calibrate sulla base del regime di volatilità predominante. Il Capitolo 2 propone un modello di struttura a termine per i tassi di interesse costruito attorno ai principi di non-arbitraggio. In linea con quanto evidenziato dal paper di Vasicek, la curva dei rendimenti è interamente guidata dalla dinamica del tasso spot. In quanto non osservabile, tale tasso istantaneo si configura come variabile latente. Il paper propone un modello State-Space in cui l'equazione di misurazione di Vasicek include il tasso spot come processo di stato. Il modello incorpora variabili esogene addizionali legate alla politica monetaria, alla volatilità del mercato azionario e ai fondamentali macroeconomici quali inflazione e output gap. Tutte le equazioni di misurazione sono guidate dal tasso spot come processo di stato. Il paper prevede l'impiego di metodologie Bayesiane e dell'algoritmo di Gibbs Sampling per la stima e l'approssimazione delle distribuzioni a posteriori dei parametri. Il paper propone un modo innovativo e interpretabile per stimare il parametro di risk-premium, formalizzato come combinazione di esplicite fonti di rischio di mercato. Il modello è applicato alla curva dei rendimenti americani registrati su frequenza mensile dal 2000 ad oggi. L'approccio Bayesiano permette la generazione di curve dei rendimenti in linea con diversi scenari macroeconomici e di valutare la coerenza delle strategie in essere rispetto al regime di mercato predominante. Il Capitolo 3 presenta una flessibile modellizzazione degli effetti contagio tra gli shocks di volatilità delle diverse attività finanziarie. Il paper propone una metodologia per modellare e prevedere vettori di varianze realizzate basata su di un modello Markov Switching VAR con probabilità di transizione non contanti nel tempo. Tale approccio permette di indagare in merito agli effetti contagio che possono intaccare i profili di volatilità delle diverse attività finanziarie detenute all'interno di una strategia di investimento. L'applicazione del modello ad un portafoglio globale che investe in titoli governativi, indici azionari e valute evidenzia come gli effetti contagio seguano una rappresentazione a regimi, in cui le probabilità di transizione da uno stato all'altro sono guidate dal tasso di interesse a 3 mesi americano, la cui dinamica è strettamente connessa alle aspettative di politica monetaria. Il paper propone un algoritmo di trading che in modo dinamico determina l'esposizione ottimale all'indice S&P500, condizionatamente alla struttura correlativa predominante tra le volatilità delle attività finanziarie detenute in portafoglio.
Three Essays in Bayesian Financial Econometrics
TROVATO, ANDREA
2025
Abstract
The following essay includes three original chapters. Chapter 1 provides an early warning method for tactical asset allocation to deal with volatility clustering and fat-tail distributions of financial returns. The early warning signals are given by a two-state Markov switching model with high-volatility and low-volatility states and time-varying transition probabilities. The transition is driven by exogenous factors namely early-warnings. Using Bayesian inference and Gibbs Sampling for posterior approximation, I apply the model to daily returns of the S&P500 Index sampled from December 2000 up to the present. The paper suggests that a measure of the gap between implied volatility and realised volatility has important leading properties in forecasting switches between volatility regimes. The paper argues that trading strategies need to be calibrated on the predominant volatility state. Therefore, trend-following signals deliver significant gains in low volatility, while short-term reversal strategies yield positive performance in the high volatility regime. In Chapter 2, a data-enriched term-structure model for interest rates is presented following a general term-structure architecture based on non-arbitrage arguments. According to the Vasicek's model, the yield curve is completely driven by the unobserved instantaneous spot rate. Nevertheless, many alternative sources of information can be used to estimate it. Our state-space model includes the traditional Vasicek's measurement equation with the spot rate as a state process and it is augmented with further measurement equations on bonds at different maturities. In order to make the evaluation of the yield curve more reactive to market events, the model incorporates additional exogenous variables related to monetary policy, equity market volatility and macroeconomic fundamentals, such as inflation and output gap. All the measurement equations are driven by the state process and the augmentation allows for improving estimation of the latent spot rate. We propose a Bayesian framework for model inference based on an efficient posterior approximation procedure. The computational efficiency follows from the linearity of the model, which allows for Kalman filtering and smoothing recursions. The paper presents an innovative and interpretable way to estimate the risk premium as a combination of explicit sources of market risk. The model is applied to the US yield curve sampled at a monthly frequency from December 2000 up to the present. The augmentation and Bayesian inference allow for generating projections of the yield curve in line with different market scenarios. Such a simulation provides great support to decisions of tactical asset allocation since it allows for evaluating the coherence of the portfolio allocation with the predominant regime implied in the market yield curve. Chapter 3 presents a flexible way for modelling volatility spillovers. Popular extensions of heterogeneous autoregressive models have been proposed for describing realized volatility. However, they leave substantial information unmodelled in residuals. This paper proposes a methodology for dynamic modelling and forecasting realized variance vectors based on a Time-Varying Bayesian Markov Switching VAR Model. The approach allows for flexible dependence patterns and investigates spillover effects across volatilities of different asset classes. The application of the model to a global portfolio investing in bonds, equities and currencies provides that spillovers follow a regime-based representation, with state transitions driven by the 3-month US T-Bill rate, whose dynamics is strictly connected to market expectations about future monetary policy decisions. Arising from estimated regime-based volatility spillovers, a dynamic trading algorithm on the S&P500 Index is provided.| File | Dimensione | Formato | |
|---|---|---|---|
|
AndreaTrovato_FINALE_PDFA.pdf
accesso aperto
Licenza:
Tutti i diritti riservati
Dimensione
2.99 MB
Formato
Adobe PDF
|
2.99 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/20.500.14242/355129
URN:NBN:IT:UNIVE-355129