Value at Risk (VaR) with Fat Tails

Value at Risk (VaR) with Fat Tails: Extending VaR Models to Account for Fat-Tailed

Distributions and Extreme Losses

Introduction

Value at Risk (VaR) is one of the most widely used risk measurement tools employed by

financial institutions globally. By accounting for market risk factors, VaR provides an estimate of

potential portfolio losses over a specific time horizon for a given confidence level. However,

conventional VaR models making assumptions of normal distributions have significant

shortcomings in adequately addressing risks arising from fat-tailed distributions and extreme

losses. This paper examines the implications of fat tails for VaR modeling and discusses

normal theory predictions.

- Infinite variance compared to finite variance for normal distributions. Hence, standard deviation

is not a reliable risk measure.

- Tendency of extreme losses to occur in bunches or clusters rather than isolated events as

assumed by normal distributions.

- Longer memory of historical data - distant events still affect current risk estimates.

- Sensitivity to outlier observations, requiring larger sample sizes for stable estimations.

Examples of fat-tailed distributions commonly used in finance are the Student's t, stable

Paretian and generalized hyperbolic distributions. Financial time series exhibit properties of

return distributions changing over time, tending towards normal or fat tails depending on market

- Underestimation of tail risks and probability of extreme losses. VaR may miss stresses beyond

its estimation window.

- Procyclicality - VaR increases in turbulent periods but may fall just when risks are highest

- Sensitivity to parameter estimations prone to errors from non-stationarity in return distributions

and data limitations.

- Difficulty identifying and assessing sensitivities to individual risk factors in periods of market

stress.

- Inability to account for dependencies between risk factors that drive bursts of extreme losses.

- Assumption of independence of returns does not hold during crises when markets become

Russia/LTCM crisis and 2008 Global Financial Crisis which laid bare risks of fat tails. Regulators

Extending VaR to Address Fat Tails

Various methodologies have been developed to extend standard VaR for improved handling of

fat tails and extreme losses:

Cornish-Fisher Expansion

This adjusts VaR estimations by modifying the quantiles of the assumed normal distribution

using measures capturing the skewness and excess kurtosis present in actual return

distributions. This improves accuracy without changing the underlying normality assumption.

However, it does not address non-normal characteristics fundamentally.

Parametric Distributions

data and used directly for VaR calculations. Parameter stability over time needs verification.

orthogonal series without specifying a parent distribution. However, they are prone to overfitting

particularly in the tail regions with sparse data.

Conditional Autoregressive Value at Risk (CAViaR)

This captures time-varying volatilities and tail risks nonparametrically using autoregressive

distributions.

Extreme Value Theory

Applies extreme value distributions like Generalized Pareto to model losses beyond a high

threshold. Combined with standard VaR, it estimates tail losses to complement the core VaR

threats.

Complementing VaR with Measures of Tail Risk

While the above methods aid VaR modeling for fat tails, no single approach is a complete

solution. A robust risk measurement framework must therefore complement standard VaR with

additional tail risk measures to gauge downtail risks:

Conditional Tail Expectation (CTE)

Defined as the expected shortfall or Tail VaR, it measures potential losses beyond the VaR

level. As it captures sensitivities deeper in the tail, CTE provides a more risk-sensitive metric

than VaR alone.

Likelihood based measures

Probability of large losses (e.g. losses exceeding 3 standard deviations), expected shortfall in

Stress testing

Involves simulating impacts of historical as well as hypothetical extreme events to evaluate

evaluations help identify mitigants.

Leveraging Fat Tails for Risk Management

Modeling fat tails enhances risk quantification by better reflecting realities of financial markets. It

also presents opportunities when integrated with the overall risk management framework:

Risk appetite setting

Incorporating tail risk measures alongside standard VaR informs defining robust quantitative

hedging decisions to mitigate concentrations.

Stress/scenario planning

Comprehending tail dependencies aids crafting relevant stress scenarios encompassing

interconnected shock transmissions across sectors in times of turmoil.

Risk reporting

management prerogatives to scale back risk-taking proactively before losses materialize.

Capital planning

Integrating tail risk outputs supports assessing capital adequacy under extreme but plausible

stresses as mandated increasingly by regulators globally.

events.

Challenges and Mitigation

Rare extreme events have limited observations, complicating estimation while non-normality

complicates extrapolation. Using external data proxies and expert judgments helps overcome

this.

Model risk

Specification errors are more pronounced in tail regions prone to misspecification biases.

amplification of risk-taking incentives in good times.

Computational intensiveness

Some techniques involving simulations are onerous for frequent use. Balancing complexity,

practicality and precision is important.

Interpretability

communication skills to facilitate appropriate use.

Constant evolution

Continual refinements track changing market dynamics and emerging threats; historical focus

alone may lead to future unpreparedness.

governance guidelines and embedded risk culture helps maximize potentials of fat tails for

augmenting standard VaR with tailored techniques allows comprehending tail risks more

realistically. By complementing quantification with qualitative tools focused on outliers and

stresses, a comprehensive perspective emerges. Regular model validations and framework

flexibility check over-reliance on backward-looking techniques alone in a continually

capabilities to anticipate and withstand losses well beyond historical volatilities. Continued

oversight and decision making.

Value at Risk (VaR) is one of the most widely used risk measurement tools employed by

financial institutions globally. By accounting for market risk factors, VaR provides an estimate of

potential portfolio losses over a specific time horizon for a given confidence level. However,

conventional VaR models making assumptions of normal distributions have significant

losses. This paper examines the implications of fat tails for VaR modeling and discusses

Heavy tails imply relatively high probability of observing extreme values that normal distributions

- Higher likelihoods of outsized losses occurring even at low confidence levels compared to

normal theory predictions.

- Infinite variance compared to finite variance for normal distributions. Hence, standard deviation

is not a reliable risk measure.

assumed by normal distributions.

- Longer memory of historical data - distant events still affect current risk estimates.

- Sensitivity to outlier observations, requiring larger sample sizes for stable estimations.

Examples of fat-tailed distributions commonly used in finance are the Student's t, stable

return distributions changing over time, tending towards normal or fat tails depending on market

founded on implicit assumptions of normal or conditionally normal distributions. Key limitations

arise:

- Underestimation of tail risks and probability of extreme losses. VaR may miss stresses beyond

during crashes.

- Sensitivity to parameter estimations prone to errors from non-stationarity in return distributions

and data limitations.

- Difficulty identifying and assessing sensitivities to individual risk factors in periods of market

These limitations were apparent in failure of VaR models to anticipate losses during 1987, 1998

now emphasize the need to complement VaR with tail risk measures.

Extending VaR to Address Fat Tails

Various methodologies have been developed to extend standard VaR for improved handling of

fat tails and extreme losses:

This adjusts VaR estimations by modifying the quantiles of the assumed normal distribution

using measures capturing the skewness and excess kurtosis present in actual return

distributions. This improves accuracy without changing the underlying normality assumption.

However, it does not address non-normal characteristics fundamentally.

Parametric Distributions

Appropriate parametric distributions like t, generalized hyperbolic or skewed t are fitted to return

These estimate the empirical CDF nonparametrically through techniques like kernel dressing or

orthogonal series without specifying a parent distribution. However, they are prone to overfitting

This captures time-varying volatilities and tail risks nonparametrically using autoregressive

models for quantile forecasts. Different quantile specifications discern normal from fat-tailed

distributions.

Extreme Value Theory

Applies extreme value distributions like Generalized Pareto to model losses beyond a high

measure. This better reflects fat tail risks but transition point sensitivities remain.

augments historical observations with systematic stress scenarios evaluated through

the estimation period. However, it relies on identified scenarios that may still understate tail

threats.

Complementing VaR with Measures of Tail Risk

While the above methods aid VaR modeling for fat tails, no single approach is a complete

solution. A robust risk measurement framework must therefore complement standard VaR with

Conditional Tail Expectation (CTE)

Defined as the expected shortfall or Tail VaR, it measures potential losses beyond the VaR

level. As it captures sensitivities deeper in the tail, CTE provides a more risk-sensitive metric

than VaR alone.

Tail dependence indices like upper and lower tail dependence coefficients estimate inter-

relationships between assets during stressed periods. This aids stress testing and portfolio

Involves simulating impacts of historical as well as hypothetical extreme events to evaluate

resilience under tail risks outside the estimation window. Both sensitivity analysis and scenario

evaluations help identify mitigants.

Leveraging Fat Tails for Risk Management

Modeling fat tails enhances risk quantification by better reflecting realities of financial markets. It

and qualitative risk taking parameters.

Exposure management

Factors driving tail risks and their vulnerabilities influence active portfolio construction and

hedging decisions to mitigate concentrations.

Stress/scenario planning

interconnected shock transmissions across sectors in times of turmoil.

Risk reporting

Reporting complementary metrics alongside VaR sheds light on situations warranting

management prerogatives to scale back risk-taking proactively before losses materialize.

Understanding the illiquidity potential in downtails guides minimum buffer requirements well

beyond one-day horizons assumed in standard VaR.

Addressing such challenges through ongoing enhancements, corroborating measures,

governance guidelines and embedded risk culture helps maximize potentials of fat tails for

While fat tails pose significant challenges to VaR modeling rooted in normal assumptions,

augmenting standard VaR with tailored techniques allows comprehending tail risks more

realistically. By complementing quantification with qualitative tools focused on outliers and

stresses, a comprehensive perspective emerges. Regular model validations and framework

flexibility check over-reliance on backward-looking techniques alone in a continually

capabilities to anticipate and withstand losses well beyond historical volatilities. Continued

oversight and decision making.

potential portfolio losses over a specific time horizon for a given confidence level. However,

shortcomings in adequately addressing risks arising from fat-tailed distributions and extreme

losses. This paper examines the implications of fat tails for VaR modeling and discusses

methodological extensions that can improve risk quantification and management capabilities.

Understanding Fat Tails

Most financial return distributions exhibit fat or heavy tails compared to the normal distribution.

cannot capture well. Some key characteristics of fat-tailed distributions include:

- Higher likelihoods of outsized losses occurring even at low confidence levels compared to

normal theory predictions.

- Infinite variance compared to finite variance for normal distributions. Hence, standard deviation

is not a reliable risk measure.

- Tendency of extreme losses to occur in bunches or clusters rather than isolated events as

Examples of fat-tailed distributions commonly used in finance are the Student's t, stable

Paretian and generalized hyperbolic distributions. Financial time series exhibit properties of

conditions. This non-stationarity poses challenges for VaR models.

Limitations of Conventional VaR

Standard VaR frameworks based on volatility-covariance matrices and historical simulations are

founded on implicit assumptions of normal or conditionally normal distributions. Key limitations

arise:

its estimation window.

- Procyclicality - VaR increases in turbulent periods but may fall just when risks are highest

and data limitations.

- Difficulty identifying and assessing sensitivities to individual risk factors in periods of market

stress.

- Inability to account for dependencies between risk factors that drive bursts of extreme losses.

- Assumption of independence of returns does not hold during crises when markets become

highly correlated.

These limitations were apparent in failure of VaR models to anticipate losses during 1987, 1998

Russia/LTCM crisis and 2008 Global Financial Crisis which laid bare risks of fat tails. Regulators

now emphasize the need to complement VaR with tail risk measures.

Extending VaR to Address Fat Tails

Various methodologies have been developed to extend standard VaR for improved handling of

However, it does not address non-normal characteristics fundamentally.

Parametric Distributions

data and used directly for VaR calculations. Parameter stability over time needs verification.

Nonparametric Methods

These estimate the empirical CDF nonparametrically through techniques like kernel dressing or

orthogonal series without specifying a parent distribution. However, they are prone to overfitting

particularly in the tail regions with sparse data.

Conditional Autoregressive Value at Risk (CAViaR)

This captures time-varying volatilities and tail risks nonparametrically using autoregressive

Extreme Value Theory

Applies extreme value distributions like Generalized Pareto to model losses beyond a high

threshold. Combined with standard VaR, it estimates tail losses to complement the core VaR

measure. This better reflects fat tail risks but transition point sensitivities remain.

Scenario Analysis and Stressed VAR

augments historical observations with systematic stress scenarios evaluated through

the estimation period. However, it relies on identified scenarios that may still understate tail

threats.

Complementing VaR with Measures of Tail Risk

While the above methods aid VaR modeling for fat tails, no single approach is a complete

solution. A robust risk measurement framework must therefore complement standard VaR with

Likelihood based measures

Probability of large losses (e.g. losses exceeding 3 standard deviations), expected shortfall in

Tail dependence indices like upper and lower tail dependence coefficients estimate inter-

relationships between assets during stressed periods. This aids stress testing and portfolio

construction decisions.

Stress testing

Involves simulating impacts of historical as well as hypothetical extreme events to evaluate

evaluations help identify mitigants.

Modeling fat tails enhances risk quantification by better reflecting realities of financial markets. It

Risk appetite setting

Incorporating tail risk measures alongside standard VaR informs defining robust quantitative

and qualitative risk taking parameters.

Exposure management

Factors driving tail risks and their vulnerabilities influence active portfolio construction and

hedging decisions to mitigate concentrations.

Stress/scenario planning

Comprehending tail dependencies aids crafting relevant stress scenarios encompassing

interconnected shock transmissions across sectors in times of turmoil.

Risk reporting

Integrating tail risk outputs supports assessing capital adequacy under extreme but plausible

stresses as mandated increasingly by regulators globally.

Liquidity risk management

Understanding the illiquidity potential in downtails guides minimum buffer requirements well

beyond one-day horizons assumed in standard VaR.

The benefits of augmenting VaR with fat tail-centric risk insights are optimized when integrated

systematically across functions to strengthen organizational resilience to unforeseen market

events.

Challenges and Mitigation

Key challenges to effectively tackling fat tails through modeling include:

Rare extreme events have limited observations, complicating estimation while non-normality

this.

Model risk

Specification errors are more pronounced in tail regions prone to misspecification biases.

Robust validation and backtesting guard against false precision.

Measures may fluctuate strongly with market cycles necessitating framework flexibility to check

amplification of risk-taking incentives in good times.

Computational intensiveness

Some techniques involving simulations are onerous for frequent use. Balancing complexity,

Constant evolution

Continual refinements track changing market dynamics and emerging threats; historical focus

alone may lead to future unpreparedness.

Addressing such challenges through ongoing enhancements, corroborating measures,

practical risk management.

Conclusion

While fat tails pose significant challenges to VaR modeling rooted in normal assumptions,

augmenting standard VaR with tailored techniques allows comprehending tail risks more

stresses, a comprehensive perspective emerges. Regular model validations and framework

transforming risk landscape. Extending the conventional VaR paradigm thus strengthens

financial institutions globally. By accounting for market risk factors, VaR provides an estimate of

potential portfolio losses over a specific time horizon for a given confidence level. However,

conventional VaR models making assumptions of normal distributions have significant

shortcomings in adequately addressing risks arising from fat-tailed distributions and extreme

losses. This paper examines the implications of fat tails for VaR modeling and discusses

methodological extensions that can improve risk quantification and management capabilities.

Most financial return distributions exhibit fat or heavy tails compared to the normal distribution.

Heavy tails imply relatively high probability of observing extreme values that normal distributions

cannot capture well. Some key characteristics of fat-tailed distributions include:

- Higher likelihoods of outsized losses occurring even at low confidence levels compared to

normal theory predictions.

- Infinite variance compared to finite variance for normal distributions. Hence, standard deviation

- Longer memory of historical data - distant events still affect current risk estimates.

- Sensitivity to outlier observations, requiring larger sample sizes for stable estimations.

Paretian and generalized hyperbolic distributions. Financial time series exhibit properties of

conditions. This non-stationarity poses challenges for VaR models.

Limitations of Conventional VaR

Standard VaR frameworks based on volatility-covariance matrices and historical simulations are

founded on implicit assumptions of normal or conditionally normal distributions. Key limitations

its estimation window.

and data limitations.

- Difficulty identifying and assessing sensitivities to individual risk factors in periods of market

stress.

- Inability to account for dependencies between risk factors that drive bursts of extreme losses.

highly correlated.

These limitations were apparent in failure of VaR models to anticipate losses during 1987, 1998

Russia/LTCM crisis and 2008 Global Financial Crisis which laid bare risks of fat tails. Regulators

now emphasize the need to complement VaR with tail risk measures.

Extending VaR to Address Fat Tails

Various methodologies have been developed to extend standard VaR for improved handling of

using measures capturing the skewness and excess kurtosis present in actual return

distributions. This improves accuracy without changing the underlying normality assumption.

However, it does not address non-normal characteristics fundamentally.

Parametric Distributions

Appropriate parametric distributions like t, generalized hyperbolic or skewed t are fitted to return

data and used directly for VaR calculations. Parameter stability over time needs verification.

Mixture models combine distributions.

Nonparametric Methods

orthogonal series without specifying a parent distribution. However, they are prone to overfitting

models for quantile forecasts. Different quantile specifications discern normal from fat-tailed

Extreme Value Theory

Applies extreme value distributions like Generalized Pareto to model losses beyond a high

threshold. Combined with standard VaR, it estimates tail losses to complement the core VaR

measure. This better reflects fat tail risks but transition point sensitivities remain.

augments historical observations with systematic stress scenarios evaluated through

simulations. It provides a qualitative overlay to quantified measures and gauges risks beyond

the estimation period. However, it relies on identified scenarios that may still understate tail

threats.

Complementing VaR with Measures of Tail Risk

While the above methods aid VaR modeling for fat tails, no single approach is a complete

Defined as the expected shortfall or Tail VaR, it measures potential losses beyond the VaR

level. As it captures sensitivities deeper in the tail, CTE provides a more risk-sensitive metric

Probability of large losses (e.g. losses exceeding 3 standard deviations), expected shortfall in

the 5% tail etc. provide pointers to vulnerabilities in specific risk categories.

Tail dependence measures

relationships between assets during stressed periods. This aids stress testing and portfolio

resilience under tail risks outside the estimation window. Both sensitivity analysis and scenario

Leveraging Fat Tails for Risk Management

Modeling fat tails enhances risk quantification by better reflecting realities of financial markets. It

also presents opportunities when integrated with the overall risk management framework:

Risk appetite setting

Incorporating tail risk measures alongside standard VaR informs defining robust quantitative

and qualitative risk taking parameters.

Exposure management

Factors driving tail risks and their vulnerabilities influence active portfolio construction and

hedging decisions to mitigate concentrations.

Stress/scenario planning

Reporting complementary metrics alongside VaR sheds light on situations warranting

management prerogatives to scale back risk-taking proactively before losses materialize.

Capital planning

Integrating tail risk outputs supports assessing capital adequacy under extreme but plausible

stresses as mandated increasingly by regulators globally.

Liquidity risk management

Understanding the illiquidity potential in downtails guides minimum buffer requirements well

systematically across functions to strengthen organizational resilience to unforeseen market

Data limitations

Rare extreme events have limited observations, complicating estimation while non-normality

complicates extrapolation. Using external data proxies and expert judgments helps overcome

this.

Specification errors are more pronounced in tail regions prone to misspecification biases.

Robust validation and backtesting guard against false precision.

Procyclicality

Measures may fluctuate strongly with market cycles necessitating framework flexibility to check

Interpretability

- Tendency of extreme losses to occur in bunches or clusters rather than isolated events as

- Sensitivity to outlier observations, requiring larger sample sizes for stable estimations.

Examples of fat-tailed distributions commonly used in finance are the Student's t, stable

Paretian and generalized hyperbolic distributions. Financial time series exhibit properties of

return distributions changing over time, tending towards normal or fat tails depending on market

its estimation window.

- Procyclicality - VaR increases in turbulent periods but may fall just when risks are highest

during crashes.

- Sensitivity to parameter estimations prone to errors from non-stationarity in return distributions

and data limitations.

- Difficulty identifying and assessing sensitivities to individual risk factors in periods of market

stress.

- Inability to account for dependencies between risk factors that drive bursts of extreme losses.

- Assumption of independence of returns does not hold during crises when markets become

highly correlated.

These limitations were apparent in failure of VaR models to anticipate losses during 1987, 1998

Various methodologies have been developed to extend standard VaR for improved handling of

fat tails and extreme losses:

Cornish-Fisher Expansion

This adjusts VaR estimations by modifying the quantiles of the assumed normal distribution

distributions. This improves accuracy without changing the underlying normality assumption.

However, it does not address non-normal characteristics fundamentally.

Parametric Distributions

Appropriate parametric distributions like t, generalized hyperbolic or skewed t are fitted to return

Mixture models combine distributions.

Nonparametric Methods

particularly in the tail regions with sparse data.

Conditional Autoregressive Value at Risk (CAViaR)

This captures time-varying volatilities and tail risks nonparametrically using autoregressive

models for quantile forecasts. Different quantile specifications discern normal from fat-tailed

distributions.

Extreme Value Theory

Applies extreme value distributions like Generalized Pareto to model losses beyond a high

threshold. Combined with standard VaR, it estimates tail losses to complement the core VaR

measure. This better reflects fat tail risks but transition point sensitivities remain.

Scenario Analysis and Stressed VAR

augments historical observations with systematic stress scenarios evaluated through

the estimation period. However, it relies on identified scenarios that may still understate tail

solution. A robust risk measurement framework must therefore complement standard VaR with

additional tail risk measures to gauge downtail risks:

Conditional Tail Expectation (CTE)

Defined as the expected shortfall or Tail VaR, it measures potential losses beyond the VaR

than VaR alone.

Likelihood based measures

Probability of large losses (e.g. losses exceeding 3 standard deviations), expected shortfall in

the 5% tail etc. provide pointers to vulnerabilities in specific risk categories.

Tail dependence measures

Tail dependence indices like upper and lower tail dependence coefficients estimate inter-

Stress testing

Involves simulating impacts of historical as well as hypothetical extreme events to evaluate

resilience under tail risks outside the estimation window. Both sensitivity analysis and scenario

evaluations help identify mitigants.

Leveraging Fat Tails for Risk Management

Modeling fat tails enhances risk quantification by better reflecting realities of financial markets. It

Risk appetite setting

Incorporating tail risk measures alongside standard VaR informs defining robust quantitative

and qualitative risk taking parameters.

Exposure management

Factors driving tail risks and their vulnerabilities influence active portfolio construction and

interconnected shock transmissions across sectors in times of turmoil.

Risk reporting

management prerogatives to scale back risk-taking proactively before losses materialize.

Capital planning

Integrating tail risk outputs supports assessing capital adequacy under extreme but plausible

stresses as mandated increasingly by regulators globally.

Liquidity risk management

Understanding the illiquidity potential in downtails guides minimum buffer requirements well

Key challenges to effectively tackling fat tails through modeling include:

Data limitations

Rare extreme events have limited observations, complicating estimation while non-normality

complicates extrapolation. Using external data proxies and expert judgments helps overcome

Model risk

Specification errors are more pronounced in tail regions prone to misspecification biases.

Robust validation and backtesting guard against false precision.

Procyclicality

Measures may fluctuate strongly with market cycles necessitating framework flexibility to check

practicality and precision is important.

Interpretability

communication skills to facilitate appropriate use.

Constant evolution

Continual refinements track changing market dynamics and emerging threats; historical focus

alone may lead to future unpreparedness.

Addressing such challenges through ongoing enhancements, corroborating measures,

practical risk management.

Conclusion

realistically. By complementing quantification with qualitative tools focused on outliers and

flexibility check over-reliance on backward-looking techniques alone in a continually

transforming risk landscape. Extending the conventional VaR paradigm thus strengthens

capabilities to anticipate and withstand losses well beyond historical volatilities. Continued

research keeps risk measurement synchronized with market complexities for optimized risk

oversight and decision making.

Value at Risk (VaR) is one of the most widely used risk measurement tools employed by

potential portfolio losses over a specific time horizon for a given confidence level. However,

conventional VaR models making assumptions of normal distributions have significant

shortcomings in adequately addressing risks arising from fat-tailed distributions and extreme

losses. This paper examines the implications of fat tails for VaR modeling and discusses

methodological extensions that can improve risk quantification and management capabilities.

Understanding Fat Tails

Most financial return distributions exhibit fat or heavy tails compared to the normal distribution.

- Infinite variance compared to finite variance for normal distributions. Hence, standard deviation

is not a reliable risk measure.

- Tendency of extreme losses to occur in bunches or clusters rather than isolated events as

- Longer memory of historical data - distant events still affect current risk estimates.

- Sensitivity to outlier observations, requiring larger sample sizes for stable estimations.

Examples of fat-tailed distributions commonly used in finance are the Student's t, stable

return distributions changing over time, tending towards normal or fat tails depending on market

founded on implicit assumptions of normal or conditionally normal distributions. Key limitations

- Underestimation of tail risks and probability of extreme losses. VaR may miss stresses beyond

its estimation window.

- Procyclicality - VaR increases in turbulent periods but may fall just when risks are highest

during crashes.

and data limitations.

- Difficulty identifying and assessing sensitivities to individual risk factors in periods of market

stress.

- Inability to account for dependencies between risk factors that drive bursts of extreme losses.

highly correlated.

Extending VaR to Address Fat Tails

Various methodologies have been developed to extend standard VaR for improved handling of

fat tails and extreme losses:

Cornish-Fisher Expansion

This adjusts VaR estimations by modifying the quantiles of the assumed normal distribution

using measures capturing the skewness and excess kurtosis present in actual return

distributions. This improves accuracy without changing the underlying normality assumption.

However, it does not address non-normal characteristics fundamentally.

Parametric Distributions

Appropriate parametric distributions like t, generalized hyperbolic or skewed t are fitted to return

Mixture models combine distributions.

These estimate the empirical CDF nonparametrically through techniques like kernel dressing or

particularly in the tail regions with sparse data.

Conditional Autoregressive Value at Risk (CAViaR)

This captures time-varying volatilities and tail risks nonparametrically using autoregressive

models for quantile forecasts. Different quantile specifications discern normal from fat-tailed

distributions.

Extreme Value Theory

Applies extreme value distributions like Generalized Pareto to model losses beyond a high

threshold. Combined with standard VaR, it estimates tail losses to complement the core VaR

measure. This better reflects fat tail risks but transition point sensitivities remain.

Scenario Analysis and Stressed VAR

augments historical observations with systematic stress scenarios evaluated through

While the above methods aid VaR modeling for fat tails, no single approach is a complete

solution. A robust risk measurement framework must therefore complement standard VaR with

Defined as the expected shortfall or Tail VaR, it measures potential losses beyond the VaR

level. As it captures sensitivities deeper in the tail, CTE provides a more risk-sensitive metric

than VaR alone.

Likelihood based measures

Probability of large losses (e.g. losses exceeding 3 standard deviations), expected shortfall in

relationships between assets during stressed periods. This aids stress testing and portfolio

Stress testing

Involves simulating impacts of historical as well as hypothetical extreme events to evaluate

resilience under tail risks outside the estimation window. Both sensitivity analysis and scenario

evaluations help identify mitigants.

Modeling fat tails enhances risk quantification by better reflecting realities of financial markets. It

also presents opportunities when integrated with the overall risk management framework:

Risk appetite setting

Incorporating tail risk measures alongside standard VaR informs defining robust quantitative

Stress/scenario planning

Comprehending tail dependencies aids crafting relevant stress scenarios encompassing

interconnected shock transmissions across sectors in times of turmoil.

Risk reporting

Reporting complementary metrics alongside VaR sheds light on situations warranting

management prerogatives to scale back risk-taking proactively before losses materialize.

Capital planning

Integrating tail risk outputs supports assessing capital adequacy under extreme but plausible

The benefits of augmenting VaR with fat tail-centric risk insights are optimized when integrated

events.

Challenges and Mitigation

Key challenges to effectively tackling fat tails through modeling include:

Data limitations

Rare extreme events have limited observations, complicating estimation while non-normality

complicates extrapolation. Using external data proxies and expert judgments helps overcome

this.

Model risk

Specification errors are more pronounced in tail regions prone to misspecification biases.

Computational intensiveness

Some techniques involving simulations are onerous for frequent use. Balancing complexity,

practicality and precision is important.

Interpretability

Tail risk outputs lack intuitive economic meaning compared to standard VaR requiring

communication skills to facilitate appropriate use.

Constant evolution

Continual refinements track changing market dynamics and emerging threats; historical focus

governance guidelines and embedded risk culture helps maximize potentials of fat tails for

augmenting standard VaR with tailored techniques allows comprehending tail risks more

realistically. By complementing quantification with qualitative tools focused on outliers and

stresses, a comprehensive perspective emerges. Regular model validations and framework

flexibility check over-reliance on backward-looking techniques alone in a continually

transforming risk landscape. Extending the conventional VaR paradigm thus strengthens

capabilities to anticipate and withstand losses well beyond historical volatilities. Continued

research keeps risk measurement synchronized with market complexities for optimized risk