Vous êtes sur la page 1sur 64

RISK CENTRIC

MACROECONOMIC
Risk Centric Macroeconomic

The case that a risk-markets dislocations


perspective of macroeconomics provides a
unified framework to think about the
mechanisms behind several of the main
economic imbalances, crises, and
structural fragilities observed in recent
decades in the global economy. This
perspective sheds light on the kind of
policies, especially unconventional ones,
that are likely to help the world economy
navigate this tumultuous environment.
Risk Centric Macroeconomic

The starting point of this risk-centric perspective


is the observation that economic activity
generates output and risks. Economic agents
must absorb both to ensure a smooth growth
process.
During normal expansions and contractions,
macroeconomists (and our models) mainly focus
on goods markets, studying whether the demand
for output is well aligned with potential output,
while risk markets considerations are relegated
to a secondary role, mostly relevant to the field
of finance.
Risk Centric Macroeconomic

These risk events take place in a global


economy with heterogenous and highly
interconnected financial markets. These
markets operate in different currencies and
are exposed to large swings in capital
flows. These flows provide many useful
services to the global economy but can be
fickle, perhaps because foreign crises
cause faster transitions from speculation.
LITERATURE REVIEW
Journal 1.
(Un)Conventional monetary policy and
bank risk-taking: A nonlinear
relationship
Introductio
n
THEORIES
Research Method
The Z-score is the most widely used accounting-based risk measure in the
banking literature to identify the balance-sheet vulnerability since it considers
a bank’s capital being insufficient to cover its losses in case of a default event.
These measures are merely asset quality indicators and are not related to any
default event, i.e., a bank may be aware that the quality of its assets is
deteriorating without being threatened by default.
• The market perception of bank risk is measured by the Distance to
Default, which is computed with standard Black and Scholes (1973) and
Merton (1974) option models and is applied to estimate individual
(Andries et al., 2018; Vassalou and Yuhang, 2004; Laeven, 2002)
• The DD is computed with the Merton (1974) option model and shows
how far a bank is from a default event, i.e., the lower its values, the
closer the bank is to insolvency. This measure has been applied to price
deposit insurance and to estimate individual and systemic bank risk
(Lehar, 2005
• DD reflects both the evolution of banks’ market value and their ability
to cover the market value of their total debt. As a result of the collapse
of stock markets across Europe, the market value of European banks
has fallen, threatening their ability to repay debt. As CEE banks were
scarcely affected by the crisis, the minimum level of the DD does not
reflect a threat to the stability of these banking systems.
Monetary policy tools

• To capture the conventional and, primarily, the unconventional character


of monetary policy, author rely on four different variables. We first
consider an observable >> interest rate variable, the policy interest, rate
(ir), the principal central banks’ instrument.
• the interest rate on the deposit facility (id), which may be an indicator of
unconventional monetary policy (UMP) since many European Central
Banks have set this policy rate below Zero.
• Authors also use the Central Banks’s Assets (in long) (CBA) to capture
quantitative easing measures. While the policy interest rate indicator
might be bounded by a zero lower limit, CBA captures the unconventional
monetary stance with an unusually high increase of the central bank’s
balance sheet.
• Monetary policy data are individual to countries, except for the interest
rate variables, which are identical for euro area members after their
accession. The CBA data are individually collected for all countries.
Conclusion (1)
 This paper investigates the effect of monetary policy - especially unconventional monetary
policy - on bank risk-taking behavior in Europe over the period 2000–2015.
 The Authors Using a dynamic panel model with a threshold effect, they estimate this effect
on two measures of bank risk: the Distance to Default, which reflects the market perception
of risk, and the asymmetric Z-score, which corresponds to an accounting-based measure of
the risk.
 This Paper find that loosening monetary policy (via low interest rates and increasing central
banks’ liquidity) has a harmful effect on banks’ risk, confirming the existence of the risk-
taking channel. Moreover, we show that this relationship is nonlinear, i.e., with the
sustainable implementation of unconventional monetary policies, the effects are stronger
below a certain threshold.

 There are four monetary policy variables: the policy interest rate, the interest on the
deposit facility, the shadow interest rate, which is a good proxy for unconventional
monetary policy (UMP), and the logarithm of central banks’ total assets. More precisely,
this paper aim to investigate whether monetary authorities’ instruments have a non-
linear effect on bank risk-taking behavior.
 The model provides interesting results when taking into account threshold effects. We
find a nonlinear effect for both measures. The decline in central bank rates, or the rise in
the monetary base, reflect riskier banks’ behavior. This result is in line with the literature
and suggests that the bank risk-taking channel was at play before the emer-gence of the
GFC. More important, our second contribution indicates
Conclusion (2)
 when interest rate indicators are low and below the threshold, the negative
relationship between bank risk measures and monetary policy is stronger.
Accounting for the central banks’ balance sheet policy indicates that
additional liquidity encourages banks to take riskier positions. Their behavior
becomes even riskier after the implementation of UMP. These findings are
confirmed when (1) using the loans-to-assets ratio as a measure of risk, (2)
using the traditional Z-score and (3) using the shadow short rate for each
country as a monetary policy variable.
 These results indicate that monetary policy does have an impact on banks’
risk-taking and that monetary authorities should be concerned about this.
They cannot set monetary conditions only on the basis of the final
objectives of monetary policy. They must also take into account their effect
on financial stability. A too accommodating monetary policy can contribute to
the accumulation of financial imbalances and difficulties in banking industry.
Prudential regulation, while reducing these risks, is not sufficient when
monetary policy is too accommodating.
Journal 2
(Monetary policy and systemic risk-taking in the euro area
banking sector)
Introduction
THEORIES
Economic Methodology Data Results
Results

This
This study
study uses
uses aa large
large
macro
macro financial
financial database
database This
This section
section presents
presents thethe
of
of 519
519 times
times series
series results
results of of CMP
CMP shocks
shocks
including
including systemic
systemic risk risk during
during 2002Q4-2017Q2,
2002Q4-2017Q2,
indicators
indicators of of thethe 30 30 and
and of of UMP
UMP and and
largest
largest euro
euro area
area “signaling
“signaling shocks”
shocks” during
during
banking
banking groups.
groups. These
These 2008Q4-2017Q2
2008Q4-2017Q2
banking
banking groups
groups come come ItIt seems
seems with
with Jimenez
Jimenez et et
from
from countries
countries that that al
al (2014)
(2014) suggestion
suggestion that that
behaved
behaved differently
differently lower
lower interest
interest rates
rates
during
during the the European
European reduce
reduce banks’
banks’ total
total credit
credit
sovereign
sovereign debt debt crisis
crisis risk
risksince
since the
the volume
volume of of
with
with respect
respect toto the
the bank
bank outstanding
outstanding loans loans isis
sovereign
sovereign nexus
nexus larger
larger than
than the
the volume
volume of of
Data
Data are
are form
form thethe ECB,
ECB, new
new loans
loans
Bloomberg,
Bloomberg, BIS, BIS, and and there
there isis evidence
evidence that that
Eurostat
Eurostat systemic
systemic riskrisk inin the
the form
form
of
of vulner-abilities
vulner-abilities
increases
increases in in the
the banking
banking
industry
industry via via contagion
contagion
and
and intercon-nectedness
intercon-nectedness
Conclusi
on
 The abundant empirical evidence available on the significance of the risk-taking channel
of monetary policy is not enough to be concerned about the possibility of a monetary
policy trade-off between price stability and financial stability. Systemic risk in the banking
sector, while operating through the traditional risk-taking channel of monetary policy, i.e.
via banks’ balance sheets, profitability and leverage, in order to be properly measured,
also requires the inclusion of contagion, interconnectedness, lending standards, time-
varying correlations and feedback effects between asset prices and banks’ balance
sheets. The several dimensions of systemic risk must be taken into account to be able to
draw robust policy implications.
 The results also highlight the importance of extending the study of systemic risk-taking to
non-bank financial intermediaries holding long-term liabilities and short-term assets, such
as pension funds, and to insurance companies and investment funds as well.
 The main policy implication of this research is that a persistently accommodative
monetary policy may drive a monetary authority with a price stability mandate to
consider a possible trade-off with financial stability. Several options are possible, although
discussing them goes beyond the scope of this paper. However, without changing existing
monetary policy frameworks to accommodate the possible trade-off, the results do
suggest, at a minimum, that the coordination between monetary and macro-prudential
policies requires serious consideration.
SYSTEMIC FINANCIAL RISK AND MACROECONOMIC 3
ACTIVITY IN CHINA
Researchers have constructed various risk indexes vis-à-vis
a wide range of systemic risks since the outbreak of the
global financial crisis in 2008. Nevertheless, most indexes
target a certain aspect of systemic risk, and are thus
incapable of measuring innately complex systemic risk in a
comprehensive way. In addition, all existing indexes of
systemic risk focus exclusively on the financial market.
Consequently, they overlook its connection with the real
economy .
In order to fill in the gap, we have synthesized a number of
financial risk measures to construct a comprehensive index
of the financial systemic risk of China by using the Principal
Components Quantile Regression (PCQR). Our results
show that this index is able to measure multi-dimension
financial risk and predict its impact on the real economy of
China more accurately than most other existing risk indexes
such as term spreads.
Method

1. The approach to forecasting the real economy


• Using the Principal Components Quantile Regression (PCQR)
method, we construct a systemic financial risk index that aggregates
information from 15 popular measures of systemic risk.
Method

Accordingly, any Ratio defined above that is < 1 indicates


efficient forecasting performance of risk index i.
Method
2. The construction of systemic risk index
Empirical Results
1. Measures of systemic risk
Empirical Results
2. Measurement of macroeconomic shocks
Conclusion

In this paper, we use the Principal Component


Quantile Regression (PCQR) to construct a
comprehensive systemic risk for China. Our PCQR
index is able to provide an accurate forecast of
macroeconomic shocks in China supported by the
empirical results.

A possible extension of this paper would be to


collect the data, provided its accessibility, that the
central bank of China uses for the construction of its
own risk index and employ an RMSE test to check if
governmental indexes perform any better than our
PCQR index does.
MACROECONOMIC RISK AND HEDGE FUND
RETURNS
4
This paper estimates hedge fund and mutual fund
exposure to newly proposed measures of
macroeconomic risk that are interpreted as measures of
economic uncertainty. We find that the resulting
uncertainty betas explain a significant proportion of the
cross-sectional dispersion in hedge fund returns.
However, the same is not true for mutual funds, for
which there is no significant relationship.
After controlling for a large set of fund characteristics
and risk factors, the positive relation between
uncertainty betas and future hedge fund returns remains
economically and statistically significant. Hence, we
argue that macroeconomic risk is a powerful
determinant of cross-sectional differences in hedge fund
returns.
Method

The macroeconomic variables we consider include


default spread, term spread, short-term interest rate
changes, aggregate dividend yield, equity market
index, inflation rate, unemployment rate, and the
growth rate of real Gross Domestic Product (GDP)
per capita.

Alternative measures of macroeconomic risk are


generated by esti- mating time-varying conditional
volatility of the afore- mentioned economic indicators
based on Vector Autoregressive–Generalized
Autoregressive Conditional Heteroskedasticity (VAR–
GARCH) model,
Result
Result
Result

Depending on the proxy for macroeconomic risk, hedge


funds in the highest uncertainty beta quintile generate
6% to 9% higher average annual returns compared to
funds in the lowest uncertainty beta quintile.
In addition, we investigate whether the predictive power
of uncertainty beta for future fund returns changes
across specific hedge fund categories. Empirical
analysis indicates that the economic and statistical
significance of the uncer- tainty betas gradually
improves as we move from the least directional to the
most directional strategies, implying a stronger relation
between uncertainty beta and future returns for funds
with sizeable time-series variation in uncertainty betas
Dynamics of the Impact of Currency Fluctuation On 5
Stock Market In India: Assessing The Pricing Of
Exchange Rate Risks
• This paper studies the dynamics of the impact of
currency fluctuation in india stock market by assessing
the pricing of exchange rate risk during the period
2005 – 2016 specifically before and after financial
crisis. Estimating a two-factor arbitrage pricing model,
using a random coefficient model, the paper present
evidence that stock returns react significantly to foreign
exchange rate fluctuation in the post-crisis period.
particularly, at 2012-2016 the exchange rate risk factor
is becoming a prominent determinant of stock returns,
indicating that indian investors are increasingly
expecting risk premiumon their investment for their
added exposureto exchange rate risk.
Empirical Model
data
1. Food & agro-based products (26)
• This study use secondary 2. Textiles (16)
data. The data was collected 3. Chemicals and chemical
using prowess which is a products (78)
database of the financial 4. Consumer goods (21)
performance of india 5. Construction material (24)
companies maintened by 6. Metals and metal products (24)
centre for monitoring indian 7. Machinery (31)
8. Transport equipment (26)
economy. The average of
9. Miscellaneous manufacturing (9)
the company stock prices for 10. Diversified (11)
each month was taken as the 11. Mining (6)
portfolio averages on which 12. Electricity (15)
monthly returns were 13. Services - other than financial
calculated. The portfolios are: (102)
14. Construction and Real Estate
(32)
15. Financial Services (79)
Methodology and Empirical Results

The market excess return series and exchange rate series have been
tested for stationarity using the ADF test. The correlation between the
excess market return series and the exchange rate series is calculated
to be 0.125 which is a reasonable explanation to justify the use of the
residuals of the regression between these two variables in order to
ensure orthogonality. The parameters, in equation (4) are estimated
using OLS.
The Hausman test between random and fixed effects suggested that a
random effects model would fit the data better. However, according to
Hsiao and Pesaran (2004), “Conventional models do not allow the
interaction of the individual specific and/or time-varying differences in the
included explanatory variables.” This was the rationale behind using
Random Coefficient models for estimation with the kind of panel data
we have. Aquino (2005) uses the Generalized Least Squares Seemingly
Unrelated Regression estimation technique to simultaneously estimate
the betas and pricing parameters.
Main findings remain invariant post such robustness test. Most of the
industry portfolios except for few industries such as ‘Electricity’,
‘Construction and Real Estate’ and ‘Financial Services’, we do not
observe non-linearity.
conclusion
• The impact of foreign exchange fluctuations on stock returns is
increasingly becoming a prominent issue to investors, financiers and
policymakers
• Empirical evidence suggests that investors are expectant of a risk
premium on their investment owing to the increased risk exposure
caused by exchange rate fluctuations, especially since the last few
years beginning from the crisis period. This implies inadequate
hedging by firms for the exchange rate risk and in the larger
macroeconomic sense, it displays market inefficiencies in the stock
market and/or foreign exchange market.
• The study has qualitatively inferred from the results obtained, that
Indian investors now require compensation for bearing exchange
rate risk. However, the quantitative aspects such as the difference
between the risk premiums demanded by investors in the various
time periods are beyond the scope of the present paper and are left
for future research.
Forecasting exchange rate value at risk using deep 6
belief network ensemble based approach
by kaijian He, lei ji, Geoffrey k.f.tso, bangzhu zhu, and yingchao zou

• This paper propose a new Value at Risk estimate based on


the Deep Belief Network ensemble model with Empirical
Mode Decomposition (EMD) technique. It attempts to capture
the multi-scale data features with the EMD-DBN ensemble
model and predict the risk movement more accurately.
Individual data components are extracted using EMD model
while individual forecasts can be calculated at different scales
using ARMA-GARCH model. The DBN model is introduced to
search for the optimal nonlinear ensemble weights to combine
the individual forecasts at different scales into the ensembled
exchange rate VaR forecasts. Empirical studies using major
exchange rates confirm that the proposed model
demonstrates the superior performance compared to the
benchmark models.
Methodology
Deep Learning (DL) has attracted significant research attentions from both
academics and industry since it was originally proposed in 2006 [17]. Deep
Belief Network (DBN) is a generative graphical model in the general deep
learning framework. Like traditional shallow neutral network, the neurons in DBN
have activation function and process the information. As the number of layers
increases rapidly from the shallow neural network to the deep neural network,
the number of the parameters involved in training deep neural network increases
exponentially. To reduce the over fitting problem, the pre-training using a stack of
a number of restricted Boltzmann machines(RBMs) is introduced in DBN. RBM
is used as the pre-training model to learn the hidden data feature in an
unsupervised learning process. A RBM is a neural network that consists of two
layers called visible (input) layer and hidden layer. In RBM, neurons in the
different levels of layers have mutual undirected connection while the neurons in
the same level of layers are independent [18]. In the structure of RBM , v and h
represent the state of the neurons of visible layer and hidden layer respectively.
When a number of RBMs are stacked in DBN, the hidden layer of the previous
RBM serves as the visible layer of the following RBM [19].
conclusion
• In this paper, we have proposed a DBN based nonlinear
ensemble model with EMD technique, for estimating value at
risk. DBN has been used to aggregate the ensemble multi
scale risk forecasts in the foreign exchange market. Positive
performance improvement in risk estimates have been
observed. Results in this paper have demonstrated that the
use of DBN model could identify more optimal ensemble
weights and better integrate the partial information from
extracted risk estimates. This approach leads to the
forecasting accuracy improvement of the multiscale VaR
estimate models. Work in this paper implies that new
innovative deep learning model can take into account the
domain knowledge in the risk estimate field to gain better
insights in the risk estimate field and achieve the improved
forecasting accuracy.
International trade, foreign direct investments, and firms’ systemic risk :
LITERATURE REVIEW Evidence from the Netherlands
7
Annelies Van Cauwenberge, Mark Vancauteren, Roel Braekers, Sigrid
Vandemaele(2019)
This paper measures the contribution of firms in the financial and non-financial
sectors to systemic risk. This paper quantify systemic risk as possible risk
spillovers from individual firms to the economy by taking into account time varying
linkages between the firm and the economy.
DATA AND MODEL ANALYSIS
 This paper use the conditional value-at-risk (ΔCoVaR), defined by Girardi and
Ergün (2013), as a parsimonious measure of systemic risk that makes it
possible to go further into the tail. Beside that, ΔCoVaR makes it possible to
combine the macro-prudential risk perspective (the systemic risk of the system)
and complement it with micro-prudential insights of individual firms (the systemic
risk contributions of the individual firms).
 Data of this research based on a novel dataset that combines data on
international trade and foreign direct investments with daily stock data for 67
Dutch listed companies from 2006–2015.
Systemic risk contributions of the financial and
non-financial sectors

 The administrative and support service activities sector is, on average, the
largest risk contributor, while the construction sector exhibits the highest volatility
in systemic risk contribution. Since consumers are less able to postpone the
consumption of goods from the wholesale and retail sector, this sector is (as
expected) the least systemic risk-contributing sector, on average.
 The results suggest that the distress of a firm in the administrative and support
service or construction sectors results in higher average losses and a higher
degree of volatility of the Dutch economy than the distress of a firm in the
wholesale and retail sector
Globalization as a determinant of
systemic risk

 Model (1) shows the fixed effects panel data results


with highly significant firm characteristics. Mainly
log(VaR) and, to a lesser extent, log(Size) explain
systemic risk contributions.
 Model (2) indicates that the three globalization
measures are highly significant determinants of
systemic risk. The results are in line with those of the
univariate analysis: trade-intensive firms contribute
less systemic risk; firms under foreign control and/or
with foreign subsidiaries have higher systemic risk
contributions than firms that are not under foreign
control and/or do not have foreign subsidiaries.
 model (3) sector dummies in all sector leaving
financial and insurance activities sector to compare
the systemic risk contribution of the non-financial
sector. The highest systemic risk contributions in the
sectors of administrative and support service,
construction, and transportation and storage.
Executive compensation among Australian mining and non-mining firms:
LITERATURE REVIEWRisk taking, long and short-term incentives
8
Subba Reddy Yarram, John Rice (2017)

This paper considers the levels and sensitivities of executive compensation in


mining and non-mining firms that were listed on the Australian All Ordinaries Index
for the period 2005 to 2013. These miners tend to have more volatile earnings,
operate with less certainty and higher risk in relation to capital investment

DATA AND MODEL ANALYSIS


 This study employs a sample of firms listed on the ASX and members of the
All Ordinaries Index for the period 2005 to 2013. All financial firms are
excluded given the highly regulated nature of the financial sector. The final
sample consists of 129 mining firms and 332 non-mining firms, with a total of
862 mining-firm-years and 2373 nonmining-firm-years for the study period of
2005 to 2013
 This paper estimate the following random effects panel data model with
cluster robust standard errors
DATA
correlations among variables

Larger frms have higher compensation. Similarly, proftability is


positively correlated to total compensation. Growth, performance
and higher levels of substantial shareholding on the other hand are
negatively related to total compensation. Firms that have
remuneration committees also show that these committees are
highly independent.
Results of Panel Random Effects Models

 Average industry pay levels have a


significant positive influence on pay
levels in individual firms, with this true for
both mining and non-mining firms.
 Economic variables have significant
influences on the pay levels of Australian
mining and non-mining firms. Firm size
has a significant positive influence on
pay levels.
 Similarly performance as measured by
Tobin's Q has a significant positive
influence on pay in both mining and non-
mining firms.
 Leverage have a significant negative
influence on the total compensation in
both mining and non-mining firms.
 Growth has a significant negative
influence on the pay level of both mining
and non-mining firms in the Australian
context.
How Informative are Variance Risk Premium and Implied Volatility for
Value-at-risk Prediction? International Evidence

Introduction
the rapid growth of financial markets over the recent decades
and the recurrence of financial crises along with the development
of new and more sophisticated financial instruments have
reinforced the need for accurate and efficientvolatility forecasting
tools. Furthermore, the latent character of the actual stock
market volatility makes the forecasting exercise more challenging
• The VIX, publicly disseminated by the Chicago Board OptionExchange
(CBOE) and commonly referred to as the “world’s barom-eter of market
volatility”, was the first successful attempt at creating and implementing a
volatility index. Its success has pavedthe way for other marketplaces to build
their own implied volatilityindexes. The VIX relevance arises not only from its
forward-lookingnature, since it embeds market sentiment and/or investors
expectations about future states, but also because it forms the backbone of a
host of volatility derivatives, particularly variance swaps.The expected
premium from selling a stock market variance in aswap contract reflects the
variance risk premium as the difference between the expected realized
variance and its risk-neutral counterpart
• The current paper attempts to investigating the information content of both implied volatility
and variance risk premium in terms of forecasting the VaR for international stock market
indexes. More specifically, this paper aims to analyse and appraise the practical usefulness
of volatility estimates from various volatility forecasting methodologies including Risk Metrics,
implied volatility and GARCH models with and without implied volatility under the VaR
framework.
• This paper contributes to the literature in a number of ways. The first contribution is to further
account for the variance risk premium as a potential predictor, and to fairly compare its
performance to both implied volatility and time series models.
• The second contribution consists in expanding upon the financial literature regarding the
market risk modelling by performing a direct comparison of the forecasting performance of
implied volatility and variance risk premium measures across two groups of countries with
different development levels of derivatives markets.
Methodology
•Variance risk premium measures

•VaR and volatility specifications

•VaR backtesting
Empirical Result

For the empirical analysis, data from seven developed


markets (France, Germany, Japan, Netherlands,
Switzerland, UK and US) and five emerging markets (Hong
Kong, India, Mexico, South Africa and South Korea) are
employed, covering the geographical regions of Africa,
Asia, America and Europe. For each market, the dataset is
daily and it consists of implied volatility index and closing
price ofthe underlying stock market index, obtained from
Thomson Reuters Eikon. The continuously compounded
daily returns are calculatedas the logarithmic difference of
daily closing prices. The sampleperiod starts on January 3,
2000 for developed countries, while it varies for emerging
markets, as reflected in Table 1, depending onthe
availability of implied volatility data. All datasets end on
August 28, 2015.
• provides summary statistics for daily
index returns,implied volatility and
the estimated variance risk premium
over the full sample period. The
return series display similar
statistical properties with respect to
skewness and kurtosis. Most of
them exhibit negative skewness.
The deviations of the median from
the mean and the values of the
excess kurtosis justify the use of the
skewed and leptokurtic distribution-
based VaR.
• As expected, the VRP is positive on
average for most of the investigated
markets.Fig. 1 depicts the VRP for the
S&P500 and HSI indexes over the full
sample period. Looking at Fig. 1, two
things stand out immediately. First, VRP
is almost always positive and displays
pronounced spikesin uncertain
episodes. For both market indexes, the
largest spikes broadly coincide and they
are observed in the third quarter of 2002
following the Enron debacle, during the
Lehman aftermath in 2008 and also
following the euro area debt crisis at the
end of 2011.
• As shown in Table 3, except for few casesof
emerging markets including India, Mexico and South
Korea, all the empirical violation rates for the RM
model are significantly higher than the theoretical
value. Such findings suggest that the RM model
underestimates the true VaR for most of the
developed markets. Regarding the CC test, the
highest p-values are attributed to the standard
GARCH and its augmented specification with the
relative VRP in three and five out of twelve cases,
respectively. Moreover, for these two models, the
hypothesis of correct conditional coverage cannot be
rejected at standard significance levels for most of
the investigated markets. Hence, for the long
position VaR, the best performing model is the G-
RVRP followed by the GARCH, while both the RM
and implied volatility models (i.e.,IVM and G-IV)
exhibit the worst performance. The results emerging
from the estimated right quantile highlight the
predictabilityof the VaR for short trading position,
where most of the models perform accurately well.
• In the relatively less extreme case of 5%-VaR, we can
see, in Table 4, that both the RM and the implied
volatility models are out performed by the GARCH for
either long or short positions. Although the RM does a
fairly good job in modelling the VaR for most of the
markets, its performance deteriorates significantly
compared to that incurred for the 1%-VaR. Again, the
inclusion of the variance risk premium improves the
forecasting accuracy of the GARCH. The superiority of
the variance risk premium in combination with the
GARCH seems to be more pronounced on developed
markets, whereas the non-augmented GARCH is the
most relevant specification in capturing risk in emerging
markets. Taken together, the results of the G-RVRP and
the G-VRP are very close in terms of modelling positive
and negative returns for different ˛-VaR levels, but the G-
RVRP performs slightly better due to the level
dependency of the VRP
• overall number of the 1%-VaR violations during the
crisis period is consistent with that during the whole
sample period. The empirical failure rate is equal to the
promised probability, and the null hypothesis of correct
UC for the various VaR models and large quantiles
cannot be rejected. The excellent performance of all the
VaR models during this period of acute economic and
financial strain highlights the usefulness of modelling
large negative and positive returns with skewed and fat-
tailed distributions. Yet, for the less extreme 5%-quantile
• show that the forecasting performance of the VaR models
deteriorates during the crisis period, predominantly in the case
of developed markets. The significant increase in VaR violations
demonstrates that the VaR estimates during the global financial
crisis are underestimated. This is mainly induced by an
overreaction to bad news as most of the VaR violations are
found in the left quantile. The overreaction of investors is not
surprising during this turbulent period and makes the estimation
of long posi-tion risk more challenging. Nevertheless, emerging
markets seem to be affected to a much lesser extent by the
crisis, as evidenced by the high statistical accuracy of most of
the investigated models for both long and short positions
• Table 8 reports the Average Minimum Capital Requirements
(AMCR) over the forecasting period, using both VaR and
sVaR forecasts generated from the different models. The G-
IV generates significantly lower AMCR values than the
competing models in developed markets, except for long
position risk in Switzerland, albeit they are very close to
those induced by the IVM, and we cannot reject the
hypothesis of mean equality in most cases. The results show
that the implied volatility models incur substantial capital
savings for both long and short positions. The best
performing models regarding the acktesting results in Section
4.2 (i.e., GARCH and G-RVRP) are significantly less
demanding than the G-IV, in terms of capital requirements,
only when it comes to emerging markets. This is the case for
Mexico and South Africa for both long and short positions,
and for South Korea when it comes to provision against long
position risk. Interestingly, even in these cases, the G-IV
cannot be significantly outperformed.
Conclusion
• On the methodological side, we apply the RM, the GARCH and the standalone implied volatility model under the
skewed-t distribution. We separately include the implied volatility, the VRP and its relative form as additional regressors
into the GARCH model. To assess the performance of the daily VaR estimates, we use newly developed Monte Carlo-
based backtests with a key emphasis on the property of conditional efficiency for the violation process featuring both
unconditional efficiency and absence of clustering. The results support earlier evidence that GARCH models including
asymmetric and heavy-tailed distributions are quite useful in quantifying and predicting market risk seeing that the
statistical sufficiency is achieved effortlessly for most of the investigated markets.
• More importantly, we find that the accuracy of VaR forecasts can be significantly enhanced by accounting for the
variance risk effect, especially for long trading positions and most markedly by including the relative VRP into the
GARCH model rather than its level. The performance ranking appears remarkably stable across challenging trading
environments and alternative measures of the variance risk premium.

Vous aimerez peut-être aussi