measuring market risk in eu new member states saša Žiković faculty of economics, university of...

Measuring Market Risk in Measuring Market Risk in EU New Member StatesEU New Member States

Saša Žiković

Faculty of Economics, University of Rijeka

13th Dubrovnik Economic Conference

Croatian National Bank

Dubrovnik, Croatia

RRisk measurement isk measurement in in EU new EU new member statesmember states

CEEC (Poland, Hungary, Czech Republic, Slovakia, Slovenia), Baltic states (Estonia, Lithuania, Latvia), Malta and Cyprus

Financial markets almost completely liberlised Consolidation trends in financial industry Domination of banks as financial intermediaries

(share in total f. assets > 80% - Slovakia > 90%) High concetration in banking sector Limited role of equity markets

Lack of research

Lagging behind EU-15 in: - financial legislation,

- market discipline, - insider trading, - disclosure of information, - knowledge of financial instruments, markets

and risks

Common features in the field of risk Common features in the field of risk managementmanagement

Similar past- state-owned banks- extension of credit - government guidelines /existing banking relationships - government, other banks, companies provided

support in distress- formal oversight, compliance with rules not risk mitigation - “moral hazard”- high costs of the system


Present & future- risk management departments in banks

- marking to market - fair value accounting (IAS 39) - quantification of risks - Value at Risk (VaR) models- Expected tail loss (ETL) models- Stress testing


Knowledge:- huge differences within national economies - foreign banks better versed

Implementation of Basel standards- foreign banks - acquired internal models, pressure from headquarters- domestic banks - standardised approach - middle/small banks without risk management

department


Implementation of Basel standards- understaffed risk management departments even

in larger banks- lack of knowledge and skilled employees- low % of banks use VaR to set trading limits- VaR not used for all risks; FX and equity risk - use of VaR for economic capital and capital requirements - medium term plans (desire, actual plan?)


Gross share of securities in assets of consolidated balance sheet of commercial banks in national economies, 2001- 2005

2005 14.5% 23.6% 28.3% 14.4% 28.0% 6.9% 2.9% 5.4%

Year POL SLVK CZE HUN SLO EST LAT LITH Avg

2001 16.8% 27.0% 27.8% 19.3% 28.1% 15.8% N/A 10.6% 20.8%

2002 16.3% 35.0% 26.0% 18.0% 34.0% 17.3% N/A 11.8% 22.6%

2003 16.4% 36.4% 26.4% 19.0% 34.2% 9.4% 4.9% 10.0% 19.6%

2004 16.2% 32.5% 26.6% 16.3% 28.9% 8.0% 3.9% 7.2% 17.4%

15.5%

Avg 16.0% 30.9% 27.0% 17.4% 30.6% 11.5% 3.9% 9.0% 19.2%

Gross share of securities in assets of consolidated balance sheet of commercial banks in national economies, 2001- 2005

2005 20,7% 43,9% 27,5%

Year Austria Germany France Average

2001 13,1% 18,5% 39,1% 23,6%

2002 12,2% 18,0% 37,7% 22,7%

2003 15,7% 18,5% 41,2% 25,1%

2004 17,0% 19,9% 42,2% 26,4%

17,9%

Average 15,2% 19,1% 40,8% 25,1%

Opposite trends between the EU new member states and EU old member states

Cleaning of banks’ balance sheets in EU new member states from:

- state issued securities - sale of interests in the companies (collateral for bad debts)

Statistical properties of capital markets Statistical properties of capital markets in in EU new member statesEU new member states

SBI20 BUX

WIG20 PX50


SKSM TALSE

RIGSE VILSE


CYSMGENL

MALTEX

Boom in the markets due to:

- catching up to EU standards

- strong inflow of foreign direct and

portfolio investments

- securities trading at a discount compared

to EU-15

CYSMGENL, WIG20 and MALTEX index diverge from strong positive trend

Statistical properties of capital markets in EU new member states

Basic statistics for stock indexes from EU new member states, 1.1.2000 - 31.12.2005

Statistical properties of capital markets in EU new member states

Normality tests for stock indexes from EU new member states, 1.1.2000 - 31.12.2005


All indexes have fat tails and asymmetry Returns are not normaly distributed Autocorrelation present in 7/10 indexes Heteroskedasticity present in all indexes Unsuitable for implementation of many VaR

models

Estimated ARMA-GARCH parameters for Estimated ARMA-GARCH parameters for stock indexes of EU new member states stock indexes of EU new member states

Basic GARCH (1,1) model was sufficient for all stock index except RIGSE (GJR-GARCH)

SBI20, VILSE, MALTEX and CYSMGENL - low persistence in volatility, very reactive

Stock indexes are far from integrated (contrary to EWMA volatility modelling - RiskMetrics model ) except CYSMGENL

Elementary assumptions of many VaR models are not satisfied

VaR models based on assumptions of normality and/or IID observations, will not perform satisfactory

Hybrid Historical simulationHybrid Historical simulation (HHS)(HHS)

Captures time varying volatility, asymmetry and leptokurtosis

Nonparametric bootstrapping standardized residuals

+ Parametric GARCH volatility forecasting

Bootstrapping - leptokurtosis and asymmetry GARCH - time varying volatility Modification of recursive bootstrap procedure

(Freedman, Peters, 1984) and volatility updating (Hull, White, 1998)


VaR at x% confidence level

G(.; t;N) empirical cumulative distribution function

Smooth density estimator (kernels)


Observation period

Growth with the passing of time = more conservative estimates

Arbitrary set the length = less conservative estimates

DataData and VaR modelsand VaR models

Returns collected from Bloomberg web site, period 01.01.2000 - 31.12.2005

1-day holding period VaR 95 and 99% confidence level Out-of-the-sample data sets = 500 latest

observations from each index

DataData and VaR modelsand VaR models

Tested VaR models

- Normal variance-covariance VaR,

- RiskMetrics system,

- Historical simulation 50, 100, 250 and 500 days,

- BRW Historical simulation λ = 0.97 and 0.99,

- GARCH RiskMetrics,

- Hybrid Historical simulation

ResultsResults

Number of VaR model failures, Kupiec and Christoffersen IND test, 95% confidence level

Kupiec test 8 4 1 1 0 0Christoffersen IND test 6 5 5 5 5 5

Kupiec test 1 1 0 0Christoffersen IND test 4 4 2 2

BRW λ=0,97

BRW λ=0,99

Model HS 50 HS 100 HS 250

ModelNormal

VCVRisk

Metrics

HS 500

GARCH RM

HHS

Results 95% clResults 95% cl

Majority of VaR models failed Kupiec test for at

least one stock index HHS, GARCH-RiskMetrics, BRW with λ = 0.97

and 0.99 passed the Kupiec test All VaR models failed Christoffersen IND test

Best performers: HHS and GARCH-RiskMetricsWorst performers: HS 50 and HS 100

ResultsResults

Number of VaR model failures, Kupiec and Christoffersen IND test, 99% confidence level

Kupiec test 10 10 3 1 9 1Christoffersen IND test 3 3 3 2 3 3

Kupiec test 7 4 3 0Christoffersen IND test 1 2 0 0

Model HS 50 HS 100 HS 250 HS 500BRW λ=0,97

BRW λ=0,99

HHSModelNormal

VCVRisk

MetricsGARCH

RM

Results 99% clResults 99% cl

Almost all VaR models perform very poorly Only HHS model passed the Kupiec test for all the indexes Majority of VaR models failed Christoffersen IND testHHS and GARCH-RiskMetrics model passed the Christoffersen IND test

Best performers: HHS, HS 500 and BRW λ = 0.99 Worst performers: HS 50, HS 100, BRW λ = 0.97 RiskMetrics model failed 4/10 indexes

ResultsResults

Historical simulation, Normal VCV and RiskMetrics system do not capture the dynamics of data generating processes of stock indexes from EU new member states

HHS model ranked as the best performer for 6/10 indexes Ranked 2. for remaining four indexes GARCH-RiskMetrics model ranked as the best performer for 3/10 indexes.

Overall ranking scores of VaR models by backtesting Overall ranking scores of VaR models by backtesting performance at 99% cperformance at 99% cll

QQualitative ualitative ccharacteristics of tested haracteristics of tested VaR modelsVaR models

ResultsResults

HHS model is the best performing tested VaR model across the stock indexes from EU new member states. GARCH-RiskMetrics and HS 500 model placed highWorst performers: HS with short observation periodsPopular VaR models (HS, Normal VCV and RiskMetrics system) placed low

CONCLUSIONCONCLUSION

All stock indexes from EU new member states are characterised by:

- fat tails and asymmetry

- autocorrelation and heteroskedasticity Returns are not IID Elementary assumption of many VaR models are not satisfied Simpler VaR models cannot be trusted, at best, provide only unconditional coverage


ARMA-GARCH models successfully capture the dynamics of stock indexes from EU new member statesVaR models that assume constant volatility or take a more simplistic view of volatility failExtensions of VaR models (BRW and RiskMetrics) show improvement over the basic models Modifying the RiskMetrics model with GARCH based volatility forecasting brings significant improvements to basic RiskMetrics model


HHS model is the best performing tested VaR model across the stock indexes from EU new member states.

VaR models that are commonly used in developed financial market are not suited for measuring market risk in EU new member states


Every VaR software package that a bank is thinking about implementing should be rigorously tested

National regulators have to take into consideration that simplistic, widely used VaR models are not well suited for developing financial markets.

Before allowance is given to banks on using internal VaR models national regulators should rigorously checks and analyse the backtesting performance as well as the theoretical framework of such models

Thank you for your attentionThank you for your attention

measuring market risk in eu new member states saša Žiković faculty of economics, university of...

Documents

field of risk management

equity risk

risk measurement

banks use var

measuring market risk

croatia slide

larger banks lack

owned banks extension