Modelling Financial Crises

Table of contents:

 

Chapter I: INTRODUCTION………………………………………………………………………………………………………. 3

1)     Background of the problem………………………………………………………………………………………………. 3

2)     Aims and objectives………………………………………………………………………………………………………… 3

3)     Research questions………………………………………………………………………………………………………… 3

4)     Structure of dissertation…………………………………………………………………………………………………… 4

Chapter II: LITERATURE REVIEW…………………………………………………………………………………………….. 4

1)     Introduction…………………………………………………………………………………………………………………….. 4

2)     Financial crisis of 2008-2009 and its effects………………………………………………………………………. 4

3)     Prior attempts to forecast financial crises………………………………………………………………………….. 5

4)     Conclusion…………………………………………………………………………………………………………………….. 8

Chapter III: METHODOLOGY……………………………………………………………………………………………………. 8

1)     Sources of Data……………………………………………………………………………………………………………… 8

2)     Methodology of Analysis………………………………………………………………………………………………….. 9

REFERENCES LIST………………………………………………………………………………………………………………. 11

 

 

 

Chapter I: INTRODUCTION

Background of the problem

Global financial crisis of 2008-2009 was one of the most serious since after the crisis of 1930s known as ‘great depression’. An example of the devastating effect of the crisis is its effect on Iceland. During 2008 – 2009 the currency of Iceland dropped more than 50% and more than 90% of the domestic financial system was seen to be collapsing (Matthiasson, 2008). The country underwent a deep recession with an output that declined by about 12% from its pre-crisis peak in 2007 to its post-crisis trough in 2009 (Claessens et al., 2013). According to Claessens and Kose (2013), financial crises equally heavily hit rich and poor countries, and thus they are ‘equal opportunities menace’ (Reinhart and Rogoff, 2009).

During the crisis period, conduct of valid policy is additionally hampered by the fact that economic forecasts are less inaccurate. According to Lewis and Pain (2014), projections of OECD regarding GDP and inflation were systematically biased during the crises – grow rates were overestimated and inflation was underestimated.

Therefore it is important for the policymakers and the investment community to be able to predict financial crises, as this would help to avoid considerable losses for investors and to take necessary policy measures for policy-makers in order to either prevent the crisis or cushion its negative effect on the economy.

Aims and objectives

The main aim is to determine the appropriate way to model the prediction of financial crisis, based on a number of leading indicators. This aim entails a number of objectives to pursue along the way: (1) determine the appropriate way to measure a financial crisis, (2) identify financial crisis instances for the selected countries, (3) determine the suitable indicators that can potentially be used as leading indicators to predict crises, (4) estimate empirically the validity of the selected indicators to predict financial crises,  (5) using the available data estimate the likelihood of crises in the selected countries in the sample, (6) Provide policy recommendations and implications.

Research questions

This research will also be targeted to answer the following research questions, with a statistically significant level. These questions are linked to the above aim and objectives. The methods such as analysis of prior literature, as well as quantitative analysis methods are used to answer these questions: (1) What is the appropriate way to measure and identify a financial crisis? (2) What factors are helpful to predict the start of a financial crisis? (3) What countries are likely to incur a financial crisis in the near future?

It is expected that financial crises can be determined based on the analysis of changes in banking system and asset prices bubbles, and crises is identified when such changed exceed certain threshold. Also, it is expected that variables that can evidence abnormal activities on the financial and real estate markets, as well as troubled state of the monetary authority could be effectively used as leading indicators.

Structure of dissertation

This dissertation has the structure as described further. Chapter II discusses prior literature about the effect of financial crisis on international economy, as well as analysis of the prior empirical studies of forecasting financial crises. Chapter III describes methodology for the forecasting financial crises and the data sources. Chapter IV provides the obtained empirical results and discussion of those. And finally Chapter V provides conclusion with the review of key results obtained, policy implications and recommendations for the future research.

 

Chapter II: LITERATURE REVIEW

Introduction

This chapter provides the review regarding the financial crises. The first part of this review considers the effect of financial crises on the economy. The second part, which is larger, provides the survey of the precious analyses of the prediction of financial crises. At the end the key results are summarised and the contribution of this study is considered towards the existing stock of literature.

 

Financial crises and their effects

Reinhart and Rogoff (2009) investigated the history of financial crises, mainly being government defaults, starting from the 14^th century default in Britain and ending with the US subprime crisis of 2007-2008 that preceded the global financial crisis. The researchers reject the false syndrome ‘this time is different’, and claim that all financial crises have a lot of similarities. The researchers argue that the sovereign domestic and external defaults as well as inflation crises are tightly related. It is so because when government defaults on its obligations, it can hardly be relied upon to sustain the value of the domestic currency.

Claessens and Kose (2013) discuss the cases and effects of the financial crises. The researchers indicate that financial crises trigger economic recession and cause significant decreases in economic output, and also in such variables as consumption, investment, and industrial production. Financial indicators follow the drop in real sector, and fall is observed in asset prices and availability of credit. Claessens and Kose (2013) analysed a number of previous empirical studies and concluded that average decrease in credit was 7%, and decrease in housing prices and stock market prices were 12% and 15%, respectively.

Claessens et al. (2013) differentiate between short-run and medium-term effects of financial crises. The short-term effects of financial crisis include recession, which is usually more severe and recovery takes longer time when recession is triggered by financial crisis. Medium-term effect pertains to the depressed economic output that cannot return to its pre-crisis levels for a substantial period of time. Also the researchers pointed out that for a high proportion of countries revival from the recession following the financial crisis occurs without significant credit growth.

 

Prior attempts to forecast financial crises

This section is aiming to survey the prior studies regarding the forecasting and predicting the financial crises, and analyse identify those methodologies that provided high-performing results.  Analysis is performed for some methodological studies as well as for the studies that performed actual empirical analysis.

A. Measurement of crisis

Chui (2002) indicates that there is no a single objective definition of crisis. Claessens and Kose (2013) identify four types of financial crises – currency crises and ‘sudden stops’ (jointly referred to as ‘first group’), debt crises and banking crises (‘second group’). Names of all some types are rather self-evident; ‘sudden stop’ refers to the situation when a significant drop in international capital inflows. Thus, according to the researchers financial crises can hardly be described based on only one indicator. According to Claessens and Kose (2013), currency crises and ‘sudden stops’ can be defined based on quantitative approach, but identification of ‘second group’ crises requires subjective judgement. Overlap of different types often occurs in practice. Most widely used are so called dichotomous or binary variables to denote crisis. Such variables take value of one for crisis periods and value of zero for the no-crisis periods.

Reinhart and Rogoff (2009) define currency crisis when exchange rates depreciate by more than 15 percent in a year, or increase in inflation over 20 percent per year. Sudden stop can be measured when capital inflows decrease by more than one standard deviation below its mean value (Claessens and Kose, 2013).

Speaking about the way to measure financial crises, Chui (2002) describes so called ‘crisis pressure index’ (CPI) is cited to be used to proxy the crisis and its strength. The index is obtained as the weighted sum of the percentage decrease in foreign exchange reserves and the depreciation of exchange rate as a percentage. The approach of signalling analysis defines crisis when CPI crosses certain threshold. An example of such threshold is the level of three standard deviations around the mean value of the index.

Laeven and Valencia (2012) provided an updated database of banking crises. The database includes all crises (including banking crises, currency crises, and sovereign debt crises) in different countries that occurred in the period from 1970 till 2011. The researchers identified a banking crisis when the following conditions were identified (both): firstly, significant distress of banks is observed; secondly, significant policy interventions into the banking system identified. The distress in banking system is identified if bank runs occur, bank liquidations are performed and losses by the banking system are earned. And policy intervention is identified when at least three out of the five listed criteria are met, which identify considerable government support channelled to the banking sector.

 

B. Studies before the global financial crisis of 2008 – 2009

Chui (2002) provides classification of available models for prediction of balance of payment and economic crises. Signalling and discrete choice models are the ones widely used in the research, along with structural approach, and contagion method. Signalling model and discrete choice model are both based on the use of pressure index as the dependent variable. Pressure index is constructed as the weighted average of the currency depreciation and changes in reserves and crises are determined when certain threshold is violated. Signalling approach considers the occurrence and non-occurrence of crises within the 24 months’ window following the triggered indicator. Chosen levels for different thresholds for factor variables are determined as the minimizers of the noise-to-signal ratio. Variable is considered useful when it results in the NSR that is less than 1.0. Discrete choice model is methodology of determining the probability of crisis based on binary regression analysis. Its advantage is that this model can conveniently consider the effect of several variables (factors) simultaneously, whereas signalling approach considers the effect of only one variable at a time.

Also, Chui (2002) explained the trade-off that takes place between the use of a higher frequency data and lower frequency data. The higher frequency data is useful for obtaining more earlier and timely predictive indicators. And lower frequency data is more widely and readily available and enables cross-country studies. The leading indicators considered by Chui (2002) include current account to GDP ratio, Reserves to debt ratio, current account overvaluation, GDP growth, equity index, and domestic credit growth.

Reagle and Salvatore (2000) used probit regression methodology to predict crises in Asian countries in 1997-1998. The researchers used six indicators: current account balance to GDP ratio, total country foreign debt to GDP ratio, ratio of short-term debt to GDP, the ratio of current account less FDI to GDP, the ratio of foreign debt service to GDP, and the available foreign reserves in the months of imports. The dependent variable used was binary variable that equalled to one in case of crisis and zero otherwise.

The use of stock prices as a leading indicator for prediction of financial crisis of 1997 – 1998 was used by Broome and Morley (2004). The researchers used monthly data for the stock market returns in a home country, as well as returns on the stock market in Hong-Kong, China, Japan and the United States, in order to predict the currency crisis in the economy.  The researchers defined currency crisis as a depreciation by more than 2% in a given month. Probit binary regression model was used to investigate the effect of the defined variables on the probability of a crisis.

Frankel and Saravelos (2010) summarises quite a number of empirical studies that were performed prior to year 2008 and analysed frequencies of the use of different leading indicators in those studies. The top 10 leading indicators were as follows: reserves, real exchange rate, credit, current account, money supply, exports or imports, inflation, equity returns and real interest rate.

 

C. Studies after the global financial crisis of 2008 – 2009

The revival of studies to forecast crises occurred after the global financial crisis of 2008-2009. These studies used the available methods as well as suggested complications that were expected to increase the predictability of crises.

Frankel and Saravelos (2010) investigated whether method of leading indicators would be helpful to predict the crisis of 2008 – 2009.  The researchers identified that the level of reserves and real exchange rates are significant leading indicators.

Babecky et al. (2013) investigated the real of leading indicators for prediction of financial crises. The researchers considered 36 countries from the EU and OECD, and analysed them for over the period from 1970 till 2010. The study concluded that the most relevant leading indicators for the prediction of financial crises are such variables as domestic housing prices, stock market prices, and credit growth. The researchers used combined measure of costs, which included loss of output and employment and fiscal deficit, to proxy financial crises. The researchers used panel vector auto-regression method combined with Bayesian averaging.

Rose and Spiegel (2011) investigated the determinants of the crisis intensity during the global financial crisis. The study determined that countries with higher per capita GDP and higher deregulation of financial markets incurred harder hurt during the crisis. Also, countries that had current account surpluses had lower negative impact during the crisis.

Rose and Spiegel (2012) suggests Multiple Indicator Multiple Cause (MIMIC) model in order to identify the leading indicators of the global financial crisis in years 2008-2009. The researchers focused on intra-country effects and ignored the cross-country ‘contagion’ effect. Despite that the researchers analysed more than 60 different possible factors that were unable to identify any statistically significant early warning variable to predict crises.

El-Shagi et al. (2013) suggests that the use of early warning signals showed its usefulness for identification of crises, but was not broadly adopted.   The researchers propose bootstrap specification of the model for different type of crisis. The conclusion suggests that the use of warning signals is useful method and provides out-of-sample forecasts of similar quality as in-sample forecasts.

In recent years more elaborated methods to predict financial crises abounded. Wolters (2015) used dynamic stochastic general equilibrium (DSGE) model to forecast the financial crises. The results showed that the use of several DSGE models increased the forecast precision. And also, DSGE model tended to overestimate economic volatility.

 

Conclusion

There are several approaches to identification of crises, some of them are based on numerical approach and others – on subjective judgement.

The two basic methods to analyse the predictive power of variables on the crisis are the signalling approach and discrete choice model. The former considers one factor at a time, and the latter considers several factors in a binary regression model. The most popular leading indicators are the current account balance, national debt, stock market dynamics, GDP growth, and foreign reserves.

The contribution of this study is based on the use of the most recent data for developed and developing countries in Europe and Asia, and also on making the analysis regarding the likelihood of crisis in the near future for the sampled countries.

 

 

Chapter III: METHODOLOGY

The current chapter provides the description of sources of data and required variables. Also, the methodology to predict financial crises, which is based on the discrete choice approach, is described in detail in this chapter.

Sources of Data

Two parts of data are required for the analysis. The first part of data is needed to estimate the crisis incidences for the countries in the sample. The second part of data is required to assess the suitability of different variables to be used as leading indicators for the prediction of financial crises.

All variables were obtained from reputable international cross-country databases – World Development Indicators by the World Bank, and International Financial Statistics by IMF.

The data is obtained for the developed and developing countries in Europe and in Asia. The time span of analysis includes the last 25 years from year 1991 till 2015.

 

Methodology of Analysis

Approach is to study the leading indicators for the two types of financial crises – currency crises and ‘full stops’. Estimate the effect of key leading indicators for the probability of both types of the crises, and compare whether predictive power of indicators differ between them.

A. Dependent variables

Dependent variable is the binary variable that takes value one for crisis and zero otherwise. The definition is based on Reinhart and Rogoff (2009) and Claessens and Kose (2013). The crisis is defined when either event occurs: (1) exchange rate of national currency to USD depreciates by more than 25% in a year, (2) CPI Inflation exceeds 20% in a year, (3) capital inflows decrease by more than one standard deviation over the adjacent 5 year period.

Dependent variables are computed for each country in the sample over the period from 1991 till 2015.

1. Independent variables

The set of nine potential leading indicators include the following variables (following Reagle and Salvatore, 2000; Frankel and Saravelos, 2010): (1) Current account to GDP ratio – ‘CAR’, (2) total foreign debt to GDP ratio – ‘TDR’, (3) short-term debt to GDP ratio – ‘STDR’, (4) ratio of foreign debt service to GDP – ‘SER’, (5) Amount of foreign reserves in the weeks of imports – ‘RES’, (6) Return on stock market index – ‘RET’, (7) Domestic credit to GDP ratio – ‘CRED’, (8) real GDP growth – ‘GDP’, (9) real interest rates – ‘RATE’.

C. Analysis methods

Analysis is based on the two methods – signalling method approach and the discrete choice model.

For signalling method, each of the abovementioned nine independent variables is analysed in order to determine its threshold for each country in the sample. Following Chui (2002), noise-to-signal ratio is computed as follows:

(1)

.Where, A – signal was issued and crisis was observed, B – signal was issued but no crisis was observed, C – no signal was issued but crisis was observed, and D – no signal issued and no crisis observed.  Threshold for every variable is determined by using ‘Goal seek’ function of Excel. Analysis is performed for selected countries.

The discrete choice model is estimated by suing the abovementioned dependent variable on the left hand side and the independent variables on the right hand side.

(2)

Where, C_it is the binary indicator of crisis for a country i in year t. And all other variables are used according to the definitions of independent variables provided earlier.

The relevant estimation methods are logit and probit models, which is suitable for cases with binary dependent variable, and used in a number of prior studies (e.g. Reagle and Salvatore, 2000).

 

D. Software

Software tools used for the analysis in this study include MS Excel spreadsheet and E-Views. Excel was used for preliminary compilation of the data, to determine the incidences of crisis and for the analysis using signalling method approach. And E-Views software was used for implementation of discrete choice model via binary variable regression analysis.

 

 

 

REFERENCES LIST

Babecky, J., Havranek, T., Mateju, J., Rusnak, M., Smidkova, K., and Vasicek, B. (2013) ‘Leading indicators of crisis incidence: Evidence from developed countries’, Journal of International Money and Finance, Vol. 35, 1 -19

Brooks, C. (2008) ‘Introductory Econometrics for Finance’, 2nd edition, Cambridge university press

Broome, S., and Morley, B. (2004) ‘Stock prices as a leading indicator of the East Asian ?nancial crisis’, Journal of Asian Economics, Vol. 15, 189 – 197

Chui M.(2002) ‘Leading Indicators of Balance of payments crises: A partial review’ , Working paper No. 171, Bank of England

Claessens, S., and Kose, M. (2013) ‘Financial Crises: Explanations, Types, and Implications’, IMF Working Paper WP/13/28

Claessens, S., Kose, M. A., Laeven, L., & Valencia, F. (2013) ‘Understanding Financial Crises: Causes, Consequences, and Policy Responses’,

Laeven, L.. & Valencia F. (2012) ‘Systemic Banking Crises Database: An Update’, IMF Working Papers 12/163, International Monetary Fund

Lewis, C., and Pain, N. (2014) ‘Lessons from OECD forecasts during and after the financial crisis’, OECD Journal: Economic Studies, Vol. 2014, 9 – 39

Frankel, J., and Saravelos, G. (2010) ‘Are Leading Indicators Of Financial Crises Useful For Assessing Country Vulnerability? Evidence from The 2008-09 Global Crisis’, NBER Working Paper 16047

Matthiasson, T. (2008) ‘Spinning out of control, Iceland in crisis”, Nordic Journal of political economy, 34(3), 1-19

Reagle, D., and Salvatore, D. (2000) ‘Forecasting Financial Crises in Emerging Market Economies’, Open Economies Review, Vol. 11, 247 – 259

Reinhart, C., and Rogoff, K. (2009) ‘This Time is Different: A Panoramic View of Eight Centuries of Financial Crises’, NBER Working Paper 13882

Rose, A., and Spiegel, M. (2011) ‘Cross-country causes and consequences of the crisis: An update’, European Economic Review, Vol. 55, 309 – 324

Rose, A., and Spiegel, M. (2012) ‘Cross-country causes and consequences of the 2008 crisis: Early warning’, Japan and The World Economy, Vol. 24, 1 – 16

Wolters, M. H. (2015) ‘Evaluating point and density forecasts of DSGE models’, Journal of Applied Econometrics, 30(1), 74-96