Peter Sarlin

Visual predictions of currency crises using self-organizing maps

Throughout the 1990s, four global waves of financial turmoil occurred. The beginning of the 21st century has also suffered from several crisis episodes, including the severe sub prime crisis. However, to date, the forecasting results are still disappointing. This paper examines whether new insights can be gained from the application of the Self-Organizing Map (SOM) – a non-parametric neural network-based visualization tool. We develop a SOM-based model for prediction of currency crises. We evaluate the predictive power of the model and compare it with that of a classical probit model. The results indicate that the SOM-based model is a feasible tool for predicting currency crises. Moreover, its visual capabilities facilitate the understanding of the factors and conditions that contribute to the emergence of currency crises in various parts of the world.

Self-organizing time map: An abstraction of temporal multivariate patterns

This paper adopts and adapts Kohonens standard self-organizing map (SOM) for exploratory temporal structure analysis. The self-organizing time map (SOTM) implements SOM-type learning to one-dimensional arrays for individual time units, preserves the orientation with short-term memory and arranges the arrays in an ascending order of time. The two-dimensional representation of the SOTM attempts thus twofold topology preservation, where the horizontal direction preserves time topology and the vertical direction data topology. This enables discovering the occurrence and exploring the properties of temporal structural changes in data. For representing qualities and properties of SOTMs, we adapt measures and visualizations from the standard SOM paradigm, as well as introduce a measure of temporal structural changes. The functioning of the SOTM, and its visualizations and quality and property measures, are illustrated on artificial toy data. The usefulness of the SOTM in a real-world setting is shown on poverty, welfare and development indicators.

Toward robust early-warning models: A horse race, ensembles and model uncertainty

This paper presents first steps toward robust models for crisis prediction. We conduct a horse race of conventional statistical methods and more recent machine learning methods as early-warning models. As individual models are in the literature most often built in isolation of other methods, the exercise is of high relevance for assessing the relative performance of a wide variety of methods. Further, we test various ensemble approaches to aggregating the information products of the built models, providing a more robust basis for measuring country-level vulnerabilities. Finally, we provide approaches to estimating model uncertainty in early-warning exercises, particularly model performance uncertainty and model output uncertainty. The approaches put forward in this paper are shown with Europe as a playground. Generally, our results show that the conventional statistical approaches are outperformed by more advanced machine learning methods, such as k-nearest neighbors and neural networks, and particularly by model aggregation approaches through ensemble learning.

Leading Indicators of Systemic Banking Crises: Finland in a Panel of EU Countries

This paper investigates leading indicators of systemic banking crises in a panel of 11 EU countries, with a particular focus on Finland. We use quarterly data from 1980Q1 to 2013Q2, in order to create a large number of macro-financial indicators, as well as their various transformations. We make use of univariate signal extraction and multivariate logit analysis to assess what factors lead the occurrence of a crisis and with what horizon the indicators lead a crisis. We find that loans-to-deposits and house price growth are the best leading indicators. Growth rates and trend deviations of loan stock variables also yield useful signals of impending crises. While the optimal lead horizon is three years, indicators generally perform well with lead times ranging from one to four years. We also tap into unique long time-series of the Finnish economy to perform historical explorations into macro-financial vulnerabilities.

Fuzzy clustering of the self-organizing map: some applications on financial time series

The Self-organizing map (SOM) has been widely used in financial applications, not least for time-series analysis. The SOM has not only been utilized as a stand-alone clustering technique, its output has also been used as input for second-stage clustering. However, one ambiguity with the SOM clustering is that the degree of membership in a particular cluster is not always easy to judge. To this end, we propose a fuzzy C-means clustering of the units of two previously presented SOM models for financial time-series analysis: financial benchmarking of companies and monitoring indicators of currency crises. It allows each time-series point to have a partial membership in all identified, but overlapping, clusters, where the cluster centers express the representative financial states for the companies and countries, while the fluctuations of the membership degrees represent their variations over time.

Ending over-lending: Assessing systemic risk with debt to cash flow

This paper operationalizes early theoretical contributions of Hyman Minsky and applies these in the context of economic sectors and nations. Following the view of boom-bust asset cycles, depicted by the endogenous build-up of risks and their abrupt unraveling, Minsky highlighted the relationship between debt obligations and cash flows. While leverage is oftentimes linked to the vulnerability of a nation, and hence systemic risk, one less explored measure of leverage is the debt-to-cash flow ratio (Debt/CF). Cash flows certainly have a well-known, academically verified connection to the ability of corporations to service and repay corporate debt. This paper investigates whether the relationship between the flow of a nations savings to its stock of total debt provides a means for understanding systemic risks. For a panel of 33 nations, we explore historic Debt/CF trends, as well as apply the same procedure to individual economic sectors. This assessment of systemic risk is arranged for presentation within a four-zone framework. In terms of an early-warning indicator, we show that the Debt/CF ratio e effectively stratifies systemic risks, and offers a useful platform toward macro-financial sustainability. Keywords: debt-to-cash flow, debt-to-gross saving, systemic risk, four-zone framework JEL codes: E210, F340, G010, H630

Visual tracking of the millennium development goals with a fuzzified self-organizing neural network

This paper uses the self-organizing map (SOM), a neural network-based projection and clustering technique, for monitoring the millennium development goals (MDGs). The eight MDGs represent commitments to reduce poverty and hunger, and to tackle ill-health, gender inequality, lack of education, lack of access to clean water and environmental degradation by 2015. This paper presents a SOM model for cross sectional and temporal visual benchmarking of countries and pairs the map with a geospatial dimension by mapping the clustering onto a geographic map. The temporal monitoring is facilitated by fuzzifying the second-level clustering with membership degrees. By creating an MDG index, and associating the SOM model with it, the model enables cross sectional and temporal analysis of the overall MDG progress of countries or regions. Further, the SOM model enables analysis of country-specific as well as regional performance according to a user-specified level of aggregation. The result of this paper is an MDG map for visual tracking and monitoring of the progress of MDG indicators.

Early-warning analysis for currency crises in emerging markets: A revisit with fuzzy clustering

Currency crises, also often called balance-of-payment crises, occur when massive capital outflows force a country to devalue or float its currency. The world-wide integration of capital markets since the 1980s and 1990s has increased the degree of capital mobility, which also determined a substantial turbulence in foreign exchange markets and frequent currency crises. In this paper, we explore advanced supporting instruments for predicting currency crises, based on an empirical study of the currency crisis episodes in 23 emerging markets around the world during the second half of last century. More specifically, we investigate the usefulness of prediction models built based on the fuzzy c-means method. First we build clustering models that partition data into a certain number of overlapping natural groups. Thereafter, we classify the data clusters into early-warning clusters and tranquil clusters. We compare the performance of our models with a conventional c-means clustering model and a benchmark probit model. The results show that the proposed models achieve a similar level of out-of-sample performance as the probit model and c-means model. The fuzzy approach also introduces additional explanatory advantages into the early-warning analysis process. Copyright

Data and dimension reduction for visual financial performance analysis

This article assesses the suitability of data and dimension reduction methods, and data–dimension reduction combinations, for visual financial performance analysis. Motivated by no comparable quantitative measure of all aspects of dimension reductions, this article attempts to capture the suitability of methods for the task through a qualitative comparison and illustrative experiments. While the discussion deals with differences of data–dimension reduction combinations in terms of their properties, the experiments illustrate their general applicability for financial performance analysis. The main conclusion is that topology-preserving data–dimension reduction combinations with predefined, regular grid shapes, such as the self-organizing map, are ideal tools for this task. We illustrate advantages of these types of methods with a visual financial performance analysis of large European banks.

On biologically inspired predictions of the global financial crisis

Early-warning models provide means for ex ante identification of elevated risks that may lead to a financial crisis. This paper taps into the early-warning literature by introducing biologically inspired models for predicting systemic financial crises. We create three models: a conventional statistical model, a back-propagation neural network (NN) and a neuro-genetic (NG) model that uses a genetic algorithm for choosing the optimal NN configuration. The models are calibrated and evaluated in terms of usefulness for policymakers that incorporates preferences between type I and type II errors. Generally, model evaluations show that biologically inspired models outperform the statistical model. NG models are, however, shown not only to provide largest usefulness for policymakers as an early-warning model, but also in form of decreased expertise and labor needed for, and uncertainty caused by, manual calibration of an NN. For better generalization of data-driven models, we also advocate adopting to the early-warning literature a training scheme that includes validation data.