Mapping the state of financial stability

https://doi.org/10.1016/j.intfin.2013.05.002Get rights and content

Highlights

  • The paper introduces modern mapping techniques to the finance community.

  • It introduces a Self-Organizing Financial Stability Map (SOFSM).

  • It maps the state of financial stability and visualizes sources of systemic risks.

  • The SOFSM performs also well as an early-warning model.

  • The SOFSM is applied to the recent global financial crisis.

Abstract

The aim of this paper is to introduce modern mapping techniques to the finance community. Mapping techniques provide means for representing high-dimensional data on low-dimensional displays. This paper lays out a methodology called the Self-Organizing Financial Stability Map (SOFSM) based upon data and dimensionality reduction that can be used for mapping the state of financial stability and visualizing potential sources of systemic risks. Besides of its visualization capabilities, the SOFSM can be used as an early-warning model that can be calibrated according to policymakers’ preferences between missing systemic financial crises and issuing false alarms. An application of the SOFSM to the recent global financial crisis shows that it performs on par with a statistical benchmark model and correctly calls the crises that started in 2007 in the United States and the euro area.

Introduction

The recent global financial crisis has demonstrated the importance of understanding sources of domestic and global vulnerabilities that may lead to a systemic financial crisis.2 Early identification of financial stress would allow policymakers to introduce policy actions to decrease or prevent further build up of vulnerabilities or otherwise enhance the shock absorption capacity of the financial system. Finding the individual sources of vulnerability and risk is of central importance since that allows targeted actions for repairing specific cracks in the financial system.

Much of the empirical literature deals with early-warning models that rely on conventional statistical methods, such as the univariate signals approach or multivariate logit/probit models.3 However, financial crises are complex events driven by non-linearly related and non-normally distributed economic and financial factors.4 These non-linearities derive, for example, from the fact that crises become more likely as the number of fragilities increase. Potentially due to restrictive assumptions, e.g. on linearity or error distributions, conventional statistical techniques may fail in modelling these events. Novel early-warning models attempt to model these complex relationships by applying non-linear techniques (Demyanyk and Hasan, 2010). For example, Peltonen (2006), Marghescu et al. (2010) and Fioramanti (2008) show that non-linear and non-parametric techniques outperform a probit model in predicting currency and debt crises. However, while the utilization of non-linear techniques may increase a posteriori prediction accuracies to a minor extent, Peltonen (2006) and Berg et al. (2005) demonstrate that the results of a priori predictions of financial crises remain disappointing. Given the changing nature of the occurrences of these extreme events, stand-alone numerical analyzes are unlikely to comprehensively describe them. As a complement, this motivates the development of tools with clear visual capabilities and intuitive interpretability, enabling an amplification of understanding through human perception.

One obvious reason why the interpretability of the monitoring systems has not been adequately addressed is related to the complexity of the problem. A large number of indicators are often required to accurately assess the sources of financial instability. As with statistical tables, standard two- and three-dimensional visualizations have, of course, their limitations for high dimensions, not to mention the challenge of including a temporal or cross-sectional dimension or assessing multiple countries over time. Although composite indices of leading indicators and predicted probabilities of early-warning models enable comparison across countries and over time, these methods fall short in describing he numerous sources of distress.5 The recent work by IMF staff on the Global Financial Stability Map (GFSM) (Dattels et al., 2010) has sought to disentangle the sources of risks by a mapping of six composite indices.6 Even here, however, the GFSM spider chart visualization of six indices falls short in disentangling the individual sources of risk. Moreover, familiar limitations of spider charts are, for example, that the area does not scale one-to-one with increases in variables and that the area itself depends on the order of dimensions, leading to complex comparisons of risks not only over time, but also across countries. In addition, the use of adjustment based on market and domain intelligence, especially during crisis episodes, and the absence of a systematic evaluation gives neither a transparent data-driven measure of financial stress nor an objective anticipation of the GFSM's future precision. Indeed, the GFSM comes with the following caveat: “given the degree of ambiguity and arbitrariness of this exercise the results should be viewed merely illustrative”.7

Methods for exploratory data analysis such as data and dimensionality reduction techniques may help in overcoming these shortcomings by illustrating data structures in easily understandable forms. Data reduction (or clustering) provides summaries of data by compressing information, while dimensionality reduction (or projection) provides low-dimensional representations of similarity relations in data. The Self-Organizing Map (SOM) (Kohonen, 1982, Kohonen, 2001) holds promise for the task by combining the aims of data and dimensionality reduction. It is capable of providing an easily interpretable non-linear description of the multidimensional data distribution on a two-dimensional plane without losing sight of individual indicators. The two-dimensional output of the SOM makes it particularly useful for visualizations, or summarizations, of large amounts of information. By 2005, over 7700 works had featured the SOM (Pöllä et al., 2009). While extensively applied to topics in engineering and medicine, the literature is short of thorough testing of the SOM for financial stability surveillance. In the emerging market context, Arciniegas and Arciniegas Rueda (2009), Sarlin and Marghescu (2011), Sarlin (2011) and Resta (2009) have applied the SOM to indicators of currency crises, debt crises and general economic and financial performance, respectively. The SOM has not, to the best of our knowledge, been earlier applied to monitoring systemic risk or assessing the global dimensions of financial stability, including global macro-financial proxies as well as individual advanced and emerging market economies.

The main aim of this paper is to introduce and promote the awareness of mapping techniques in the field of finance and the policymaking community in general and financial stability surveillance in particular. The use of mapping techniques in financial stability surveillance is illustrated by laying out a methodology based upon data and dimensionality reduction for mapping the state of financial stability, and visualizing potential sources of systemic risks.

The methodology uses five elements for constructing a Self-Organizing Financial Stability Map (SOFSM): (1) data and dimensionality reduction based upon the SOM, (2) identification of systemic financial crises (3) macro-financial indicators of vulnerabilities and risks, (4) an evaluation framework for assessing model performance, and (5) a model training framework. As an enhancement to the GFSM proposed by the IMF, the SOFSM not only functions as a display for mapping financial stability, but also performs well as an early-warning model in predicting out-of-sample systemic financial crises and allows disentangling the individual sources of vulnerability. The SOFSM is implemented as a complement to an early-warning model for predicting systemic financial crises (Lo Duca and Peltonen, 2013) by using a similar set of vulnerability indicators and definition of systemic financial crisis. Hence, while being evaluated as an early-warning model, the SOFSM is a complementary tool that should be treated as a starting point rather than an ending point for financial stability surveillance.

Inspired by Minsky's (1982) and Kindleberger's (1996) vindicated financial fragility view of a credit or asset cycle, the SOFSM introduces the notion of the financial stability cycle. Thus, the SOFSM can be used to monitor macro-financial vulnerabilities by locating a country in the financial stability cycle: being it either in the pre-crisis, crisis, post-crisis or tranquil state. We illustrate how the SOFSM can be used for visualizing samples of the panel dataset, i.e. cross-sectional data over time on a country level, as well as different levels of aggregation, such as emerging and advanced economies. In addition, when assessing a topologically ordered SOFSM, we use the concept of a financial stability neighborhood for assessing potential for contagion through similarities in macro-financial conditions. That is, a crisis in one position on the map indicates propagation of financial distress to adjacent locations. This type of representation may help in identifying the changing nature of crises. Likewise, we illustrate how the SOFSM can be used for visualizing scenario analysis results, by showing how positive and negative shocks on domestic and global levels affect the location of the euro area in the financial stability cycle.

The paper has also some more technical contributions compared to the earlier SOM literature. Out of the above applications, only the model for assessing exchange-rate pressure by Sarlin and Marghescu (2011) is evaluated thoroughly and systematically in terms predictive performance. In addition, we not only perform thorough tests of predictive capabilities and robustness, but also account for policymakers’ preferences between missing systemic financial crises (type I error) and issuing false alarms (type II error), as well as the imbalanced frequency of crisis and tranquil times, by calibrating the model with a novel evaluation framework (Sarlin, 2013b). Robustness of the SOFSM is tested by varying the SOM parameters, threshold values of the models, policymaker's preferences between type I and II errors, and the forecast horizons. The SOFSM also makes use of a semi-supervised SOM, rather than the standard unsupervised SOM, used in all above applications to financial indicators, in order to better model the financial stability cycle.8 Moreover, as a key contribution is the use of the SOFSM as visualization display, we emphasize the correct use of color scales. All used colors are calibrated such that perceived distances in color correctly represent distances in data.

The paper is structured as follows. Section 2 introduces the five elements of the SOFSM, while Section 3 presents its training and evaluation as well as robustness checks. Section 4 illustrates how the SOFSM can be used for detecting signs of vulnerabilities and potential for contagion and for mapping the state of financial stability over time and across countries, as well as for different levels of aggregation. The section also exemplifies how the SOFSM can be used for visualizing scenario analysis results. Section 5 concludes.

Section snippets

Methodology

This section describes the five elements that are necessary for constructing the Self-Organizing Financial Stability Map (SOFSM): (1) data and dimensionality reduction based upon the Self-Organizing Map (SOM), (2) identification of systemic financial crises, (3) macro-financial indicators of vulnerabilities and risks, (4) a model evaluation framework for assessing performance, and (5) a model training framework.

Self-Organizing Financial Stability Map (SOFSM)

This section presents the creation of the Self-Organizing Financial Stability Map (SOFSM). We first use the training and evaluation frameworks for constructing the SOFSM and then perform thorough robustness checks.

Mapping the state of financial stability

In this section, we use the SOFSM for detecting signs of vulnerabilities and potential for contagion, and for mapping the state of financial stability using macro-financial conditions. We map samples of the panel dataset by showing cross-sectional and time-series data on the two-dimensional SOFSM. We also compute aggregates for groups of countries for exploring states of financial stability globally, in advanced economies and in emerging economies. In addition, we illustrate how a scenario

Conclusions

The aim of this paper is to introduce modern mapping techniques to the finance community. Mapping techniques provide means for representing high-dimensional data on low-dimensional displays. In this paper, the use of mapping techniques in financial stability surveillance is illustrated by laying out a methodology called the Self-Organizing Financial Stability Map (SOFSM) for mapping the state of financial stability, and visualizing potential sources of systemic risks.

The SOFSM is a

References (52)

  • P. Sarlin

    On policymakers’ loss functions and the evaluation of early warning systems

    Economics Letters

    (2013)
  • I.E. Arciniegas Rueda et al.

    SOM-based data analysis of speculative attacks’ real effects

    Intelligent Data Analysis

    (2009)
  • R. Balakrishnan et al.

    The Transmission of Financial Stress from Advanced to Emerging Economies

    (2009)
  • A. Berg et al.

    Assessing early warning systems: how have they worked in practice?

    IMF Staff Papers

    (2005)
  • C. Borio et al.

    Asset Prices, Financial and Monetary Stability: Exploring the Nexus

    (2002)
  • C. Borio et al.

    Securing Sustainable Price Stability: Should Credit Come Back from the Wilderness?

    (2004)
  • T.F. Cox et al.

    Multidimensional Scaling

    (2001)
  • P. Dattels et al.

    Can you Map Global Financial Stability?

    (2010)
  • G. Deboeck

    Software tools for self-organizing map

  • G. Deboeck

    Best practices in data mining using self-organizing maps

  • A. Demirgüç-Kunt et al.

    Monitoring banking sector fragility. A multivariate logit

    World Bank Economic Review

    (2000)
  • R. Dornbusch et al.

    Contagion: how it spreads and how it can be stopped

    World Bank Research Observer

    (2000)
  • T. Eklund et al.

    Evaluating a SOM-based financial benchmarking tool

    Journal of Emerging Technologies in Accounting

    (2008)
  • J.C. Forte et al.

    Advantages and drawbacks of the Batch Kohonen algorithm

  • C.S. Hakkio et al.

    Financial Stress: What is it, How can it be measured and Why does it matter?

    Federal Reserve Bank of Kansas City Economic Review

    (2009)
  • M. Harrower et al.

    Colorbrewer.org: an online tool for selecting colour schemes for maps

    Cartographic Journal

    (2003)
  • Cited by (0)

    The authors want to thank anonymous referees, Barbro Back, Tomas Eklund, Kristian Koerselman, Marco Lo Duca, Fredrik Lucander, and seminar participants at the Bank of Finland seminar on 11 November 2011 in Helsinki, the 14th Annual DNB Research Conference ‘Complex systems: Towards a better understanding of financial stability and crises’ on 3–4 November 2011 in Amsterdam, the ESCB Macro-Prudential Research (MaRs) Network workshop on 14–15 April 2011 in Frankfurt am Main, the International Joint Conference on Artificial Intelligence (IJCAI’11) workshop on Chance Discovery on 16–22 July 2011 in Barcelona, the ECB Financial Stability seminar on 16 September 2011 in Frankfurt am Main, the First Conference of the ESCB MaRs Network on 5–6 October 2011 in Frankfurt am Main, the Bank of Finland Institute for Economies in Transition (BOFIT) and the Data Mining and Knowledge Management Laboratory at Åbo Akademi University for useful comments and discussions. All remaining errors are of our own. The views presented in the paper are those of the authors and do not necessarily represent the views of the European Central Bank or the Eurosystem.

    1

    Directorate General Financial Stability, Financial Stability Surveillance Division, European Central Bank, Kaiserstrasse 29, 60311 Frankfurt am Main, Germany.

    View full text