Essays on modern finance and financial networks

Three main assumptions underpin modern financial theories: (1) price behaviour is both independently and randomly volatile, (2) investors are rational, making decisions to maximise expected utility, and (3) markets respond to information efficiently. However, modern financial theories have been subjected to constant scrutiny by critics, both at a theoretical and empirical level. Large numbers of studies conducted by experimental economists and market psychologists have documented departures from these assumptions. These departures include phenomena such as market failures, crashes, bubbles, crises, spreading phenomena, and the stability and fragility of financial markets. Financial markets are governed by systemic shifts and display non-equilibrium properties that make modelling their behaviour highly complex and controversial. These arguments set the scene for opening richer contexts to develop new models for investigating the complex behaviour of financial markets. In this thesis, the overarching purpose of research is to adopt a network approach for a better understanding and improved modelling of the collective behaviour of global stock markets. Chapter outlines the strand of literature that recommends network theory to model the functional structure of different markets, and thus, more accurately, the scope or degree of influence across multiple financial elements. Studies focusing on (dis)similarity-based network methodologies have shown that despite the great success and importance of this literature, a serious debate has emerged about the practicality of edge definitions based on a correlation measure. According to this view, correlation-based financial networks only fully describe the financial system if the system behaves linearly and, additionally, if an assumption of multivariate normal distribution holds. These conditions do not match the complexity of those systems and the solid evidence of the nonlinearity of financial markets. Simultaneously, the z-score normalisation, which is intrinsic in the correlation function, ignores the means (expected return) and standard deviations (risk) when modelling interconnectivities. These oversights raise the significant question of whether security return networks can be realistically modelled and interpreted by market correlations. Chapter 3 introduces the standard Euclidean distance as an alternative to the correlation measure for modelling connectivity across multiple financial indices. Such an extension relaxes the linearity assumption of financial systems and incorporates indices' risk and return profile along with their correlation property (Lemma 3.3.1). This metric is used to explain the collective behaviour of the MSCI world market and compare the results with other correlation networks. Findings show that realised volatility accounts for 71% of the observed topology, whereas correlation explains only 29% of the vii market structure. No evidence was found supporting the importance of expected return. Power-law exponents and degree distributions reveal that the centrality of hub nodes is considerably higher in the Euclidean as opposed to correlation networks. Accordingly, the importance and influence of central countries (like US and Japan hubs) in spreading high volatility is considerably higher than what correlation networks report. Chapter 4 takes this idea of network modelling and adds it into a multifractal study of the financial return series. Multifractal analysis provides powerful tools to understand the complex, nonlinear nature of time series in diverse fields. Currently, multifractal analysis forms one of the main directions of financial market analysis. However, the main body of existing literature investigates single return series (i.e., market portfolio index) and its multifractal properties to form conclusions on the (in)efficiency of markets. This is while the collective behaviour of financial time series requires analysis of the interplay between all financial elements rather than by individually studying the constituent indices. To describe this relationship between the various elements of the financial system, a state-dependent CEN model is used along with suitable network indicators that explain the complex dynamics of return series at both the individual and aggregate level. Next, the multifractal properties of the network indicators and return series are investigated. Empirical results reveal that in the MSCI world market, the return process and associated network parameters all exhibit a high degree of multifractality that account for a level of inefficiency in the studied markets. In Chapter 5, the network approach is used to improve the accuracy of volatility forecasts from the standard GARCH model and its modified versions (E-GARCH and GJR-GARCH). Two network factors appended to the standard GARCH model and the results were tested using 23 return series from the MSCI world index. In line with our expectation was the superior performance of the Network-GARCH model over all selected GARCH class models. The results achieved were notably conclusive and consistent across all 23 counties in both in-sample and out-sample forecast. At the same time, in cross-benchmark comparisons, it was found that the GARCH (1,1) provides the worst overall performance by all evaluation criteria. There was no reason to believe that the E-GARCH model was not as good as the GJR-GARCH model in predicting conditional volatility.