Predicting financial crashes by time series analysis
The project aims to identify hidden dynamics in financial systems and test potential early warning signals instructed by other subprojects. In contrast to biological systems, the mechanisms and the dynamics driving ‘contagion’ in financial markets are relatively little explored (Allen 2000; Kaminsky 2000; Claessens 2013). As regards the mechanisms that can drive co-movements in asset prices, the financial literature has stressed that these may simply reflect the economic interlinkages among financial entities (‘fundamental contagion’). Such inter-linkages can stem from trade or cross-border loan exposures and represent the flip-side of economic and financial integration. Asset price co-movements in excess of what such linkages would warrant are sometimes called ‘pure contagion’: they could be driven by investor panic and herding behaviour (Forbes 2002). As an intermediate case, investors may also become more sensitive to certain fundamentals. All these forms of contagion are observationally equivalent with an increase in asset price correlations. From a practical point of view the distinction between the underlying causes of financial contagion may be irrelevant. However, it may matter for the design of optimal policy responses.
With respect to the dynamics of financial contagion, recent episodes have shown that crises can spread rapidly across financial entities, countries and time zones. This is to be expected in an integrated global financial market, where new information should be priced immediately to prevent arbitrage opportunities arising. There is a large body of literature on predicting financial crisis, including books by (Schiff 2012, Dent 2009, McHugh 2013, Sornette 2004). The work in (Feigenbaum 1998; Sornette 2004) suggested that financial crisis events might fit log-periodic power laws that tend to infinity near the critical time. While in some cases this work can make predictions, it failed to capture important CTs (CXO 2005, Johansen 2003), perhaps due to the very simplistic nature of the model. This subproject aims to investigate more deeply into the mechanisms of CTs in financial markets to better understand which categories they belong to − or whether they cannot be categorised at all.
Research findings from other disciplines including medical sciences and biology have the potential to contribute to the understanding of financial crises, provided that they can utilise models for the spreading of a disease or cancer cells that have empirically testable implications for the behaviour or co-movement of the affected units (Montell 2008; Rieu 2000; Upadhyaya 2001; Glazier 1993). For example, the spreading of cancer cells in a human body may follow certain patterns (e.g., local linear spreading in one direction or circular spreading in all directions). Provided that these patterns generate typical co-movements among units, they could be used to identify ‘infections’ in financial markets. Their precise timing and localisation would allow the design of targeted policy interventions in the most critical market segments, while avoiding the use of blunt policy instruments (e.g. a cut in interest policy rates by the central bank) that can lead to moral hazard and negative side effects on financial stability. We will apply measures developed by the theoretical subprojects to analyse whether previous financial crises would have been detectable earlier using novel measures. Throughout the project we will work with financial institutions in Luxembourg and abroad to gain further data. For example, Roland Beck from the European Central Bank may provide additional data and insight into this subproject.