2011年-ECB欧洲央行_Mapping_the_State_of_Financial_Stability_8页_323kb

Summary of "Mapping the State of Financial Stability"

Core Content

The document introduces the Self-Organising Financial Stability Map (SOFSM), a novel methodology for mapping the state of financial stability and identifying systemic risks. It is based on Self-Organising Maps (SOMs), a type of unsupervised neural network that reduces high-dimensional data to a two-dimensional visual representation. The SOFSM is designed to detect macro-financial vulnerabilities, visualise systemic risk sources, and predict financial crises, offering an alternative to traditional early warning systems that rely on binary classifications or statistical models.

Main Points

1. Importance of Financial Stability Mapping

• Understanding systemic risk and financial vulnerabilities is essential for policymakers to prevent crises or improve the shock-absorption capacity of the financial system.
• Traditional early warning systems often use univariate or multivariate models, such as logit/probit models, which may fail to capture the complexity of financial vulnerabilities.
• The SOFSM provides a more comprehensive and visual approach to identifying the stages of the financial stability cycle (pre-crisis, crisis, post-crisis, tranquil).

2. Methodology of the SOFSM

• Self-Organising Map (SOM) is used for data and dimensionality reduction.
• The SOM algorithm maps input data onto a two-dimensional grid, preserving the neighbourhood relationships in the data.
• The map is divided into clusters based on Ward's hierarchical clustering using class variables that represent the financial stability cycle stages.
• Feature planes are used to visualise the distribution of individual indicators on the map, with cold and warm colours indicating low and high values respectively.
• Second-level clustering allows for more detailed analysis by grouping nodes into clusters, which represent broader financial states.

3. Systemic Events and Vulnerabilities

• Systemic financial crises are defined using a Financial Stress Index (FSI), which measures co-movement across market segments (money, equity, foreign exchange).
• The FSI is converted into a binary crisis variable, where a crisis is identified when the FSI exceeds the 90th percentile of its country-specific distribution.
• A set of 94 systemic events is identified across 28 economies from 1990 to 2010.
• Additional class variables are used to define pre-crisis, post-crisis, and tranquil periods.

4. Evaluation Framework

• The loss function is used to evaluate model performance, balancing the trade-off between false negatives (missing crises) and false positives (false alarms).
• The usefulness of the model is defined as the difference between the expected value of a guess and the actual loss, given a preference parameter $\mu$.
• The AUC (Area Under the Curve) is used to assess the predictive power of the model.
• The SOFSM is compared with a standard logit model using in-sample and out-of-sample data, and it performs at least as well in classification and prediction.

5. Training and Testing the SOFSM

• The SOFSM is trained using data from Q1 1990 to Q1 2005 and tested from Q2 2005 to Q2 2009.
• The model is semi-supervised, meaning it uses class variable information during training.
• The SOFSM has 137 nodes arranged in an $11 \times 13$ grid.
• On the training set, the SOFSM shows slightly better performance in terms of usefulness, recall positives, precision negatives, and AUC, though the logit model performs better in other metrics.
• On the test set, the SOFSM outperforms the logit model in overall accuracy due to a higher proportion of crisis episodes in the out-of-sample dataset.

Key Indicators and Variables

• Asset price developments and valuations
• Credit developments and leverage
• Government budget deficit and current account deficit
• Global indicators are derived by averaging transformed variables from major economies (US, euro area, Japan, UK)
• Feature planes help in visualising the relationship between individual indicators and the financial stability states

Temporal Analysis of Financial Stability

• The SOFSM successfully maps the financial stability states of the US and the euro area from Q1 2002 to Q2 2011.
• It identifies the pre-crisis state for both regions as early as Q1 2006.
• In Q1 2007, the US remains in the pre-crisis state, while the euro area enters the crisis state.
• By Q1 2008, both regions are classified as being in the crisis state.
• In Q1 2010, the euro area moves into the tranquil state, indicating that macro-prudential indicators may not capture vulnerabilities in smaller economies.
• In Q2 2011, the euro area is at the border of the pre-crisis cluster, suggesting a potential early warning signal, while the US is in the post-crisis and tranquil states.

Conclusion

• The SOFSM is a novel and effective tool for mapping financial stability and identifying systemic risk sources.
• It provides visual insights and early warning capabilities, complementing traditional statistical models.
• The SOFSM performs at least as well as the logit model in both in-sample and out-of-sample predictions.
• It highlights the importance of macro-financial vulnerabilities and the need for a broader set of indicators to capture the full spectrum of financial risks.