The introduction of machine learning algorithms for the analysis of indicators of financial stability
Баймет Файзулла, магистрант
Магистратуры Прикладных Финансов при АО «НАРХОЗ»
В данной статье рассмотрены проблемы применения различных методик, моделей и комбинаций с индикаторами раннего реагирования кризисной ситуации. В связи с ростом неопределенности и нестабильности рыночной среды, необходимо недопущение перерастания кризисного явления в кризисную ситуацию, которая имеет большие масштабы потерь, имеющие высокую скорость реализации. Необходимо постоянное наблюдение, прогнозирование кризисов и определение на раннем этапе роста вероятности кризиса.
Ключевые слова: макропруденциальное регулирование, индикаторы финансовой устойчивости, вероятность кризиса, машинное обучение, Random Forest.
Abstract. This article discusses the problems of applying various methods, models and combinations with indicators of early warning indicators to a crisis situation. In connection with the growth of uncertainty and instability of the market environment, it is necessary to prevent the crisis from developing into a crisis situation, which has large scale losses with a high speed of realization. Constant monitoring, forecasting crises and determining the probability of a crisis at an early stage of growth are needed.
Keywords: macroprudential regulation, indicators of financial stability, probability of a crisis, machine learning, Random Forest.
1. Financial soundness indicators
Financial soundness indicators are indicators of the current financial condition and sustainability of the entire sector of a country’s financial institutions, as well as the corporate sector and the household sector, which are clients of financial institutions. These indicators represent an innovative and new area of macroeconomic data.
Financial soundness indicators are indicators of the current financial condition and sustainability of a country’s financial institutions and their counterparties from the corporate sector and the household sector. They include both aggregated data for individual institutions and indicators that give an idea of the markets in which financial institutions operate. FSI are calculated and distributed for use in macroprudential analysis. Such as analysis is an assessment and control of the strengths and vulnerabilities of financial systems in order to increase financial stability and, in particular, to reduce the likelihood of a financial system crash.
Introduction to Financial Soundness Indicators (FSI):
• FSI serve as indicators of the current state of the country’s financial sector and institutional units (financial institutions, corporations, households).
• FSI is used as a summary of the financial sustainability of institutions and markets in the economy.
• FSI is a subgroup of a much wider group of macro-prudential indicators (such as credit-to-GDP, current account balance-to-GDP, gross fixed capital formation to GDP, inflation rate, real effective exchange rate, short term interest rate) that should be considered in the analysis of financial stability.
• FSI analysis complements other types of sustainability assessment, such as early warning indicators and macroeconomic vulnerability assessment.
• Most FSIs aggregate oversight data collected across banks and are necessary to assess the risks to the financial system as a whole. (necessary, but insufficient)
• Aggregation allows you to identify (some) risks that are not traceable at the micro level, especially those related to common open positions.
2. How does the Forests model look like?
Number of decision trees. Random Forests technology. It is inspired by the work of Breiman, Friedman, Olshen, and Stone on the CART © decision tree and Breiman’s Bagger (bootstrap aggregation). In the Forests we can analyze the number of target attributes, depending on the subject of the study. In the case of a credit card or a credit card account success or failure of the number of similar seen in the business intelligence (Bl) and customer relationship management (CRM) literature.
It is a commonplace to find credit card scoring analyzers beginning with or fewer predictors. While there may be fewer than 10 predictors. It is important to understand the list of predictors, identifying the eligible predictors. It is because of the fact that it’s possible.
Once the target and the eligible predictors are identified, the CART-like decision tree begins. There are several important ways in which this tree differs from a standard CART tree, however. First, we need you to get the tree. Instead we use a bootstrap sample (described in more detail below). Bootstrap sample it includes 2/3 of the original training data. Next, we’ll have been able to complete the process. This has been selected at random. In this case, it’s not a completely random splitter. Finally, the decision tree is maximized and then left unpruned.
It is not a problem. However, it’s possible that it’s not a problem. In fact, it is possible to make decisions such as neural networks, solo decision trees, logistic regression, or support vector machines (SVM).
What are the main advantages of Random Forests (RF) as a modeling tool?
Automatic predictor selection from any number of candidates (potentially thousands):
• The analyst does not need to do any variable selection or data reduction.
• RF will ultimately identify the best predictors automatically.
Ability to handle data without preprocessing:
• Data do not need to rescaled, transformed, or modified in any way.
• Resistance to outliers in any column (predictors or target)
• Automatic handling of missing values
• RF models are often considerably more accurate than a single tree.
• The accuracy achieved is often competitive with the best alternative methods.
Resistance to Overtraining:
• Growing a large number of RF trees does not create a risk of overfitting.
• Each tree is a completely independent random experiment. s Built-in self-testing using” Out of Bag”
• Self-testing is based on an extension of cross validation that is repeated several hundred times.
• Self-tests provide highly reliable assessments of the reliability of the RF model.
• Trees are grown at high speed because few variables are in use at any one time.
RF Cluster identification:
• RF can be used to generate tree-based clusters that are metric free.
• Variables defining clusters are chosen automatically and can vary across clusters.
• RF offers novel graphical displays that can yield new insights into data.
It is proposed to introduce and evaluate the Random Forest model, as one of the machine learning models, to identify the factors affecting the deterioration of the financial situation of the bank and the banking conglomerate, as well as to identify recommendations for the application of early warning measures and methods for determining the factors that affect the deterioration of the financial situation of the bank and the banking conglomerate. Comparison of the Random Forest model with existing assessment methods in the Republic of Kazakhstan.
When analyzing financial stability, you can use a group of macroprudential indicators, such as credit to GDP, current account balance to GDP, gross fixed capital formation to GDP, inflation rate, real effective exchange rate, short-term interest rate, real effective exchange rate, oil price , real estate, stock market prices.
List of relevant literature:
1. Financial soundness Indicators. Drafting Guide. INTERNATIONAL MONETARY FUND, 2007.
2. Breiman, L, Friedman, J., Olshen. R., & Stone, C. (1984). “Classification and Regression Trees”, Wadsworth.
3. Efron, B., & Tibshirani, R. (1993). “An Introduction to the Bootstrap”. Chapman and Hall.
4. LEO BREIMAN Statistics Department, University of California, Berkeley, CA 94720 Random Forests
5. LEO BREIMAN (Statistics Department, University of California. Berkele), CA 94720 Bagging Predictors
6. Michie, D., Spiegelhalter, D.J. & Taylor, C.C. (19~4). Machine Learning, Neural and Statistical Classification.
7. Salford Systems. Salford Predictive Modeler. Introducing Salford Predictive Modeler. Introduction to Random Forest (guides).