Jump Diffusion Models in Finance: Extending Diffusion Processes to Incorporate Jump
Components in Asset Price Dynamics
Introduction
Diffusion processes have long been used in finance to model the stochastic behavior of asset
prices over time. However, empirical evidence has shown that the basic continuous diffusion
model is unable to fully capture important stylized features of financial time series such as fat
tails and excess kurtosis. To address this shortcoming, researchers have extended the standard
diffusion framework to incorporate jump components. These jump diffusion models provide a
more realistic characterization of asset price dynamics by allowing for discontinuities in price
paths.
In this assignment, I will explore jump diffusion models and their use in finance. Specifically, I
will discuss the motivations for extending standard diffusion processes to include jumps. I will
then introduce some key jump diffusion models including the Merton model and the mixed
exponential jump diffusion model. Finally, I will discuss some empirical applications and
evidence regarding these models. The overall goal is to provide an overview of how
incorporating jump components has improved our ability to model important properties of asset
returns and better match real-world financial data.
Motivation for Jump Diffusion Models
There are a few important empirical observations that motivated the development of jump
diffusion models as an alternative to standard diffusion processes for modeling asset prices:
Excess kurtosis in return distributions: Diffusion processes like Brownian motion yield asset
return distributions that are normally distributed with finite variance. However, empirical
analyses consistently find that return distributions exhibit "fat tails" and excess kurtosis relative
to the normal distribution. This suggests discontinuities or "jumps" may be present.
Volatility clustering: Volatility tends to be clustered over time rather than independent and
identically distributed as assumed by standard diffusion models. Jumps could help account for
sudden increases in volatility.
Feedback effects: Major market events like crashes sometimes coincide with announcements or
macroeconomic shifts that could reasonably be expected to suddenly impact asset valuations.
Jumps allow modeling of such discontinuities.
Implied volatility smiles: Option pricing models based on diffusion processes cannot fully
replicate the implied volatility "smile" patterns seen in market data, where out of the money
options have higher volatilities than would be predicted. Jumps help explain these patterns.
Taken together, these empirical regularities indicate standard diffusion models are too simplistic
and that incorporating random jumps could produce models with superior descriptive power.
Jumps allow for discontinuities that generate fatter tails, clustered volatility, and market events
not easily modeled through smooth continuous paths alone.
Merton Model
One of the earliest and most influential jump diffusion models is the Merton (1976) model.
Unlike standard diffusion processes, the Merton model allows for sudden discontinuous "jumps"
in the process in addition to the continuous diffusion component. Specifically, the model extends
geometric Brownian motion as follows:
dS(t)/S(t) = μdt + σdz(t) + dq(t)
Where:
- S(t) is the asset price at time t
- μ is the drift parameter
- σ is the volatility parameter
- dz(t) is the increment of a Wiener process, representing the continuous diffusion component
- dq(t) represents the discontinuous jump component
Merton assumes the jump component follows a simple compound Poisson process, where
jumps occur randomly according to a Poisson process with intensity λ. The jump size η is
exponentially distributed with parameter ν.
Therefore, between jumps the process behaves like geometric Brownian motion. But at jump
times, the asset price undergoes instantaneous random multiplicative jumps up or down in price
according to the distribution of the jump size η.
A key insight of Merton's model is that it allowed him to develop a closed-form solution for the
value of European options on the asset using a risk-neutral valuation framework, despite the
additional complexity introduced by jumps. This represented an important conceptual and
practical advance, helping to establish jump diffusion models as a potentially useful mainstream
alternative to standard diffusion-based models.
Empirically, Merton-type models have been found to generate more realistic heavy-tailed return
distributions compared to diffusion-only models. They also better fit volatility smiles seen in
option markets. However, the exponential jump size assumption implies a symmetric distribution
which has been criticized as unrealistic and not robustly supported empirically. This motivated
extensions of the basic Merton framework.
Mixed Exponential Jump Diffusion Model
One such extension is the mixed exponential jump diffusion (MEJD) model proposed by Bates
(1996). In the MEJD model, jump sizes are assumed to follow a mixture of two exponential
distributions rather than a single exponential. Specifically:
dS(t)/S(t) = μdt + σdz(t) + dq(t)
Where:
- The jump intensity λ follows a Poisson process as in Merton
- With probability p, jump size η ~ Exp(ν1)
- With probability 1-p, jump size η ~ Exp(ν2)
Allowing for two exponential distributions with different parameters ν1 and ν2 makes the overall
jump size distribution asymmetric, addressing a key limitation of Merton's model. Empirically it is
well established that large downward jumps tend to be more frequent than large upward jumps.
Bates showed the MEJD model also permits a tractable option pricing solution under the risk-
neutral valuation framework. Risk-neutral parameters can be calibrated to market option prices
to infer investor risk perceptions, akin to how implied volatilities are estimated from option prices
under diffusion models.
Empirical tests have found substantial evidence that the MEJD model provides a much better fit
to important patterns in financial time series compared to one-distribution jump models or pure
diffusions. For example, it more accurately replicates the leptokurtic and skewed distribution of
stock returns, as well as the implied volatility smiles and skews observed for equity index
options. The ability to capture asymmetric jump behavior with different tail weights has been an
important empirical success and justification for the mixed distribution approach over Merton's
simpler specification.
Empirical Applications and Evidence
A variety of empirical studies have estimated jump diffusion and more advanced specifications
using financial market and asset return data:
- Bates (2000) estimated the MEJD model using S&P 500 index and individual stock return
data, finding statistically significant evidence of jumps and superior fit relative to standard
diffusion models.
- Eraker et al. (2003) used an MCMC technique to estimate stochastic volatility jump diffusion
models, again finding strong empirical support and dominance over alternatives for S&P 500
and currency returns.
- Broadie et al. (2007) estimated stochastic volatility MEJD models and applied them to option
pricing on 30 stocks in the Dow Jones Industrial Average, finding useful improvements over
pure diffusion models.
- Christoffersen et al. (2010) applied MEJD models augmented with stochastic volatility to hedge
fund return data, providing evidence hedge funds exhibit significant jump components.
- Kawakami (2015) estimated a fractional MEJD model using Japanese stock index data,
demonstrating its ability to capture long memory properties also evident in returns.
These studies consistently report that jump diffusion specifications including MEJD models
provide a substantially improved ability to fit important stylized facts of returns like non-normal
heavy tails, volatility clustering, and kurtosis. They also confirm jumps help replicate key
patterns in option prices like volatility smiles. Overall, there is very strong empirical evidence
that including discontinuities through random jumps enhances the realism and practical
application of stochastic process models in finance.
Conclusion
In conclusion, jump diffusion models provide a significant extension of the standard diffusion
framework used for modeling asset prices and returns. Empirical regularities in financial time
series motivate introducing discontinuous jumps to better characterize properties like heavy
tails, volatility clustering, and option smiles seen in practice. Models like the Merton jump
diffusion and mixed exponential jump diffusion specifications developed by Merton and Bates
are parsimonious yet flexible enough to generate realistic dynamics and tractable pricing
formulas. Numerous empirical studies applying these models to real return data establish their
superior ability to fit and characterize important stylized facts compared to ordinary diffusion
processes alone. While refinements continue to be made, jump diffusion modeling represents
an important conceptual advance and practical tool for researchers seeking models with greater
descriptive accuracy for asset price behavior.
Diffusion processes have long been used in finance to model the stochastic behavior of asset
prices over time. However, empirical evidence has shown that the basic continuous diffusion
model is unable to fully capture important stylized features of financial time series such as fat
tails and excess kurtosis. To address this shortcoming, researchers have extended the standard
diffusion framework to incorporate jump components. These jump diffusion models provide a
more realistic characterization of asset price dynamics by allowing for discontinuities in price
paths.
In this assignment, I will explore jump diffusion models and their use in finance. Specifically, I
will discuss the motivations for extending standard diffusion processes to include jumps. I will
then introduce some key jump diffusion models including the Merton model and the mixed
exponential jump diffusion model. Finally, I will discuss some empirical applications and
evidence regarding these models. The overall goal is to provide an overview of how
incorporating jump components has improved our ability to model important properties of asset
returns and better match real-world financial data.
Motivation for Jump Diffusion Models
There are a few important empirical observations that motivated the development of jump
diffusion models as an alternative to standard diffusion processes for modeling asset prices:
Excess kurtosis in return distributions: Diffusion processes like Brownian motion yield asset
return distributions that are normally distributed with finite variance. However, empirical
analyses consistently find that return distributions exhibit "fat tails" and excess kurtosis relative
to the normal distribution. This suggests discontinuities or "jumps" may be present.
Volatility clustering: Volatility tends to be clustered over time rather than independent and
identically distributed as assumed by standard diffusion models. Jumps could help account for
sudden increases in volatility.
Feedback effects: Major market events like crashes sometimes coincide with announcements or
macroeconomic shifts that could reasonably be expected to suddenly impact asset valuations.
Jumps allow modeling of such discontinuities.
Implied volatility smiles: Option pricing models based on diffusion processes cannot fully
replicate the implied volatility "smile" patterns seen in market data, where out of the money
options have higher volatilities than would be predicted. Jumps help explain these patterns.
Taken together, these empirical regularities indicate standard diffusion models are too simplistic
and that incorporating random jumps could produce models with superior descriptive power.
Jumps allow for discontinuities that generate fatter tails, clustered volatility, and market events
not easily modeled through smooth continuous paths alone.
Merton Model
One of the earliest and most influential jump diffusion models is the Merton (1976) model.
Unlike standard diffusion processes, the Merton model allows for sudden discontinuous "jumps"
in the process in addition to the continuous diffusion component. Specifically, the model extends
geometric Brownian motion as follows:
dS(t)/S(t) = μdt + σdz(t) + dq(t)
Where:
- S(t) is the asset price at time t
- μ is the drift parameter
- σ is the volatility parameter
- dz(t) is the increment of a Wiener process, representing the continuous diffusion component
- dq(t) represents the discontinuous jump component
Merton assumes the jump component follows a simple compound Poisson process, where
jumps occur randomly according to a Poisson process with intensity λ. The jump size η is
exponentially distributed with parameter ν.
Therefore, between jumps the process behaves like geometric Brownian motion. But at jump
times, the asset price undergoes instantaneous random multiplicative jumps up or down in price
according to the distribution of the jump size η.
A key insight of Merton's model is that it allowed him to develop a closed-form solution for the
value of European options on the asset using a risk-neutral valuation framework, despite the
additional complexity introduced by jumps. This represented an important conceptual and
practical advance, helping to establish jump diffusion models as a potentially useful mainstream
alternative to standard diffusion-based models.
Empirically, Merton-type models have been found to generate more realistic heavy-tailed return
distributions compared to diffusion-only models. They also better fit volatility smiles seen in
option markets. However, the exponential jump size assumption implies a symmetric distribution
which has been criticized as unrealistic and not robustly supported empirically. This motivated
extensions of the basic Merton framework.
Mixed Exponential Jump Diffusion Model
One such extension is the mixed exponential jump diffusion (MEJD) model proposed by Bates
(1996). In the MEJD model, jump sizes are assumed to follow a mixture of two exponential
distributions rather than a single exponential. Specifically:
dS(t)/S(t) = μdt + σdz(t) + dq(t)
Where:
- The jump intensity λ follows a Poisson process as in Merton
- With probability p, jump size η ~ Exp(ν1)
- With probability 1-p, jump size η ~ Exp(ν2)
Allowing for two exponential distributions with different parameters ν1 and ν2 makes the overall
jump size distribution asymmetric, addressing a key limitation of Merton's model. Empirically it is
well established that large downward jumps tend to be more frequent than large upward jumps.
Bates showed the MEJD model also permits a tractable option pricing solution under the risk-
neutral valuation framework. Risk-neutral parameters can be calibrated to market option prices
to infer investor risk perceptions, akin to how implied volatilities are estimated from option prices
under diffusion models.
Empirical tests have found substantial evidence that the MEJD model provides a much better fit
to important patterns in financial time series compared to one-distribution jump models or pure
diffusions. For example, it more accurately replicates the leptokurtic and skewed distribution of
stock returns, as well as the implied volatility smiles and skews observed for equity index
options. The ability to capture asymmetric jump behavior with different tail weights has been an
important empirical success and justification for the mixed distribution approach over Merton's
simpler specification.
Empirical Applications and Evidence
A variety of empirical studies have estimated jump diffusion and more advanced specifications
using financial market and asset return data:
- Bates (2000) estimated the MEJD model using S&P 500 index and individual stock return
data, finding statistically significant evidence of jumps and superior fit relative to standard
diffusion models.
- Eraker et al. (2003) used an MCMC technique to estimate stochastic volatility jump diffusion
models, again finding strong empirical support and dominance over alternatives for S&P 500
and currency returns.
- Broadie et al. (2007) estimated stochastic volatility MEJD models and applied them to option
pricing on 30 stocks in the Dow Jones Industrial Average, finding useful improvements over
pure diffusion models.
- Christoffersen et al. (2010) applied MEJD models augmented with stochastic volatility to hedge
fund return data, providing evidence hedge funds exhibit significant jump components.
- Kawakami (2015) estimated a fractional MEJD model using Japanese stock index data,
demonstrating its ability to capture long memory properties also evident in returns.
These studies consistently report that jump diffusion specifications including MEJD models
provide a substantially improved ability to fit important stylized facts of returns like non-normal
heavy tails, volatility clustering, and kurtosis. They also confirm jumps help replicate key
patterns in option prices like volatility smiles. Overall, there is very strong empirical evidence
that including discontinuities through random jumps enhances the realism and practical
application of stochastic process models in finance.
Conclusion
In conclusion, jump diffusion models provide a significant extension of the standard diffusion
framework used for modeling asset prices and returns. Empirical regularities in financial time
series motivate introducing discontinuous jumps to better characterize properties like heavy
tails, volatility clustering, and option smiles seen in practice. Models like the Merton jump
diffusion and mixed exponential jump diffusion specifications developed by Merton and Bates
are parsimonious yet flexible enough to generate realistic dynamics and tractable pricing
formulas. Numerous empirical studies applying these models to real return data establish their
superior ability to fit and characterize important stylized facts compared to ordinary diffusion
processes alone. While refinements continue to be made, jump diffusion modeling represents
an important conceptual advance and practical tool for researchers seeking models with greater
descriptive accuracy for asset price behavior.
Diffusion processes have long been used in finance to model the stochastic behavior of asset
prices over time. However, empirical evidence has shown that the basic continuous diffusion
model is unable to fully capture important stylized features of financial time series such as fat
tails and excess kurtosis. To address this shortcoming, researchers have extended the standard
diffusion framework to incorporate jump components. These jump diffusion models provide a
more realistic characterization of asset price dynamics by allowing for discontinuities in price
paths.
In this assignment, I will explore jump diffusion models and their use in finance. Specifically, I
will discuss the motivations for extending standard diffusion processes to include jumps. I will
then introduce some key jump diffusion models including the Merton model and the mixed
exponential jump diffusion model. Finally, I will discuss some empirical applications and
evidence regarding these models. The overall goal is to provide an overview of how
incorporating jump components has improved our ability to model important properties of asset
returns and better match real-world financial data.
Motivation for Jump Diffusion Models
There are a few important empirical observations that motivated the development of jump
diffusion models as an alternative to standard diffusion processes for modeling asset prices:
Excess kurtosis in return distributions: Diffusion processes like Brownian motion yield asset
return distributions that are normally distributed with finite variance. However, empirical
analyses consistently find that return distributions exhibit "fat tails" and excess kurtosis relative
to the normal distribution. This suggests discontinuities or "jumps" may be present.
Volatility clustering: Volatility tends to be clustered over time rather than independent and
identically distributed as assumed by standard diffusion models. Jumps could help account for
sudden increases in volatility.
Feedback effects: Major market events like crashes sometimes coincide with announcements or
macroeconomic shifts that could reasonably be expected to suddenly impact asset valuations.
Jumps allow modeling of such discontinuities.
Implied volatility smiles: Option pricing models based on diffusion processes cannot fully
replicate the implied volatility "smile" patterns seen in market data, where out of the money
options have higher volatilities than would be predicted. Jumps help explain these patterns.
Taken together, these empirical regularities indicate standard diffusion models are too simplistic
and that incorporating random jumps could produce models with superior descriptive power.
Jumps allow for discontinuities that generate fatter tails, clustered volatility, and market events
not easily modeled through smooth continuous paths alone.
Merton Model
One of the earliest and most influential jump diffusion models is the Merton (1976) model.
Unlike standard diffusion processes, the Merton model allows for sudden discontinuous "jumps"
in the process in addition to the continuous diffusion component. Specifically, the model extends
geometric Brownian motion as follows:
dS(t)/S(t) = μdt + σdz(t) + dq(t)
Where:
- S(t) is the asset price at time t
- μ is the drift parameter
- σ is the volatility parameter
- dz(t) is the increment of a Wiener process, representing the continuous diffusion component
- dq(t) represents the discontinuous jump component
Merton assumes the jump component follows a simple compound Poisson process, where
jumps occur randomly according to a Poisson process with intensity λ. The jump size η is
exponentially distributed with parameter ν.
Therefore, between jumps the process behaves like geometric Brownian motion. But at jump
times, the asset price undergoes instantaneous random multiplicative jumps up or down in price
according to the distribution of the jump size η.
A key insight of Merton's model is that it allowed him to develop a closed-form solution for the
value of European options on the asset using a risk-neutral valuation framework, despite the
additional complexity introduced by jumps. This represented an important conceptual and
practical advance, helping to establish jump diffusion models as a potentially useful mainstream
alternative to standard diffusion-based models.
Empirically, Merton-type models have been found to generate more realistic heavy-tailed return
distributions compared to diffusion-only models. They also better fit volatility smiles seen in
option markets. However, the exponential jump size assumption implies a symmetric distribution
which has been criticized as unrealistic and not robustly supported empirically. This motivated
extensions of the basic Merton framework.
Mixed Exponential Jump Diffusion Model
One such extension is the mixed exponential jump diffusion (MEJD) model proposed by Bates
(1996). In the MEJD model, jump sizes are assumed to follow a mixture of two exponential
distributions rather than a single exponential. Specifically:
dS(t)/S(t) = μdt + σdz(t) + dq(t)
Where:
- The jump intensity λ follows a Poisson process as in Merton
- With probability p, jump size η ~ Exp(ν1)
- With probability 1-p, jump size η ~ Exp(ν2)
Allowing for two exponential distributions with different parameters ν1 and ν2 makes the overall
jump size distribution asymmetric, addressing a key limitation of Merton's model. Empirically it is
well established that large downward jumps tend to be more frequent than large upward jumps.
Bates showed the MEJD model also permits a tractable option pricing solution under the risk-
neutral valuation framework. Risk-neutral parameters can be calibrated to market option prices
to infer investor risk perceptions, akin to how implied volatilities are estimated from option prices
under diffusion models.
Empirical tests have found substantial evidence that the MEJD model provides a much better fit
to important patterns in financial time series compared to one-distribution jump models or pure
diffusions. For example, it more accurately replicates the leptokurtic and skewed distribution of
stock returns, as well as the implied volatility smiles and skews observed for equity index
options. The ability to capture asymmetric jump behavior with different tail weights has been an
important empirical success and justification for the mixed distribution approach over Merton's
simpler specification.
Empirical Applications and Evidence
A variety of empirical studies have estimated jump diffusion and more advanced specifications
using financial market and asset return data:
- Bates (2000) estimated the MEJD model using S&P 500 index and individual stock return
data, finding statistically significant evidence of jumps and superior fit relative to standard
diffusion models.
- Eraker et al. (2003) used an MCMC technique to estimate stochastic volatility jump diffusion
models, again finding strong empirical support and dominance over alternatives for S&P 500
and currency returns.
- Broadie et al. (2007) estimated stochastic volatility MEJD models and applied them to option
pricing on 30 stocks in the Dow Jones Industrial Average, finding useful improvements over
pure diffusion models.
- Christoffersen et al. (2010) applied MEJD models augmented with stochastic volatility to hedge
fund return data, providing evidence hedge funds exhibit significant jump components.
- Kawakami (2015) estimated a fractional MEJD model using Japanese stock index data,
demonstrating its ability to capture long memory properties also evident in returns.
These studies consistently report that jump diffusion specifications including MEJD models
provide a substantially improved ability to fit important stylized facts of returns like non-normal
heavy tails, volatility clustering, and kurtosis. They also confirm jumps help replicate key
patterns in option prices like volatility smiles. Overall, there is very strong empirical evidence
that including discontinuities through random jumps enhances the realism and practical
application of stochastic process models in finance.
Conclusion
In conclusion, jump diffusion models provide a significant extension of the standard diffusion
framework used for modeling asset prices and returns. Empirical regularities in financial time
series motivate introducing discontinuous jumps to better characterize properties like heavy
tails, volatility clustering, and option smiles seen in practice. Models like the Merton jump
diffusion and mixed exponential jump diffusion specifications developed by Merton and Bates
are parsimonious yet flexible enough to generate realistic dynamics and tractable pricing
formulas. Numerous empirical studies applying these models to real return data establish their
superior ability to fit and characterize important stylized facts compared to ordinary diffusion
processes alone. While refinements continue to be made, jump diffusion modeling represents
an important conceptual advance and practical tool for researchers seeking models with greater
descriptive accuracy for asset price behavior.
Diffusion processes have long been used in finance to model the stochastic behavior of asset
prices over time. However, empirical evidence has shown that the basic continuous diffusion
model is unable to fully capture important stylized features of financial time series such as fat
tails and excess kurtosis. To address this shortcoming, researchers have extended the standard
diffusion framework to incorporate jump components. These jump diffusion models provide a
more realistic characterization of asset price dynamics by allowing for discontinuities in price
paths.
In this assignment, I will explore jump diffusion models and their use in finance. Specifically, I
will discuss the motivations for extending standard diffusion processes to include jumps. I will
then introduce some key jump diffusion models including the Merton model and the mixed
exponential jump diffusion model. Finally, I will discuss some empirical applications and
evidence regarding these models. The overall goal is to provide an overview of how
incorporating jump components has improved our ability to model important properties of asset
returns and better match real-world financial data.
Motivation for Jump Diffusion Models
There are a few important empirical observations that motivated the development of jump
diffusion models as an alternative to standard diffusion processes for modeling asset prices:
Excess kurtosis in return distributions: Diffusion processes like Brownian motion yield asset
return distributions that are normally distributed with finite variance. However, empirical
analyses consistently find that return distributions exhibit "fat tails" and excess kurtosis relative
to the normal distribution. This suggests discontinuities or "jumps" may be present.
Volatility clustering: Volatility tends to be clustered over time rather than independent and
identically distributed as assumed by standard diffusion models. Jumps could help account for
sudden increases in volatility.
Feedback effects: Major market events like crashes sometimes coincide with announcements or
macroeconomic shifts that could reasonably be expected to suddenly impact asset valuations.
Jumps allow modeling of such discontinuities.
Implied volatility smiles: Option pricing models based on diffusion processes cannot fully
replicate the implied volatility "smile" patterns seen in market data, where out of the money
options have higher volatilities than would be predicted. Jumps help explain these patterns.
Taken together, these empirical regularities indicate standard diffusion models are too simplistic
and that incorporating random jumps could produce models with superior descriptive power.
Jumps allow for discontinuities that generate fatter tails, clustered volatility, and market events
not easily modeled through smooth continuous paths alone.
Merton Model
One of the earliest and most influential jump diffusion models is the Merton (1976) model.
Unlike standard diffusion processes, the Merton model allows for sudden discontinuous "jumps"
in the process in addition to the continuous diffusion component. Specifically, the model extends
geometric Brownian motion as follows:
dS(t)/S(t) = μdt + σdz(t) + dq(t)
Where:
- S(t) is the asset price at time t
- μ is the drift parameter
- σ is the volatility parameter
- dz(t) is the increment of a Wiener process, representing the continuous diffusion component
- dq(t) represents the discontinuous jump component
Merton assumes the jump component follows a simple compound Poisson process, where
jumps occur randomly according to a Poisson process with intensity λ. The jump size η is
exponentially distributed with parameter ν.
Therefore, between jumps the process behaves like geometric Brownian motion. But at jump
times, the asset price undergoes instantaneous random multiplicative jumps up or down in price
according to the distribution of the jump size η.
A key insight of Merton's model is that it allowed him to develop a closed-form solution for the
value of European options on the asset using a risk-neutral valuation framework, despite the
additional complexity introduced by jumps. This represented an important conceptual and
practical advance, helping to establish jump diffusion models as a potentially useful mainstream
alternative to standard diffusion-based models.
Empirically, Merton-type models have been found to generate more realistic heavy-tailed return
distributions compared to diffusion-only models. They also better fit volatility smiles seen in
option markets. However, the exponential jump size assumption implies a symmetric distribution
which has been criticized as unrealistic and not robustly supported empirically. This motivated
extensions of the basic Merton framework.
Mixed Exponential Jump Diffusion Model
One such extension is the mixed exponential jump diffusion (MEJD) model proposed by Bates
(1996). In the MEJD model, jump sizes are assumed to follow a mixture of two exponential
distributions rather than a single exponential. Specifically:
dS(t)/S(t) = μdt + σdz(t) + dq(t)
Where:
- The jump intensity λ follows a Poisson process as in Merton
- With probability p, jump size η ~ Exp(ν1)
- With probability 1-p, jump size η ~ Exp(ν2)
Allowing for two exponential distributions with different parameters ν1 and ν2 makes the overall
jump size distribution asymmetric, addressing a key limitation of Merton's model. Empirically it is
well established that large downward jumps tend to be more frequent than large upward jumps.
Bates showed the MEJD model also permits a tractable option pricing solution under the risk-
neutral valuation framework. Risk-neutral parameters can be calibrated to market option prices
to infer investor risk perceptions, akin to how implied volatilities are estimated from option prices
under diffusion models.
Empirical tests have found substantial evidence that the MEJD model provides a much better fit
to important patterns in financial time series compared to one-distribution jump models or pure
diffusions. For example, it more accurately replicates the leptokurtic and skewed distribution of
stock returns, as well as the implied volatility smiles and skews observed for equity index
options. The ability to capture asymmetric jump behavior with different tail weights has been an
important empirical success and justification for the mixed distribution approach over Merton's
simpler specification.
Empirical Applications and Evidence
A variety of empirical studies have estimated jump diffusion and more advanced specifications
using financial market and asset return data:
- Bates (2000) estimated the MEJD model using S&P 500 index and individual stock return
data, finding statistically significant evidence of jumps and superior fit relative to standard
diffusion models.
- Eraker et al. (2003) used an MCMC technique to estimate stochastic volatility jump diffusion
models, again finding strong empirical support and dominance over alternatives for S&P 500
and currency returns.
- Broadie et al. (2007) estimated stochastic volatility MEJD models and applied them to option
pricing on 30 stocks in the Dow Jones Industrial Average, finding useful improvements over
pure diffusion models.
- Christoffersen et al. (2010) applied MEJD models augmented with stochastic volatility to hedge
fund return data, providing evidence hedge funds exhibit significant jump components.
- Kawakami (2015) estimated a fractional MEJD model using Japanese stock index data,
demonstrating its ability to capture long memory properties also evident in returns.
These studies consistently report that jump diffusion specifications including MEJD models
provide a substantially improved ability to fit important stylized facts of returns like non-normal
heavy tails, volatility clustering, and kurtosis. They also confirm jumps help replicate key
patterns in option prices like volatility smiles. Overall, there is very strong empirical evidence
that including discontinuities through random jumps enhances the realism and practical
application of stochastic process models in finance.
Conclusion
In conclusion, jump diffusion models provide a significant extension of the standard diffusion
framework used for modeling asset prices and returns. Empirical regularities in financial time
series motivate introducing discontinuous jumps to better characterize properties like heavy
tails, volatility clustering, and option smiles seen in practice. Models like the Merton jump
diffusion and mixed exponential jump diffusion specifications developed by Merton and Bates
are parsimonious yet flexible enough to generate realistic dynamics and tractable pricing
formulas. Numerous empirical studies applying these models to real return data establish their
superior ability to fit and characterize important stylized facts compared to ordinary diffusion
processes alone. While refinements continue to be made, jump diffusion modeling represents
an important conceptual advance and practical tool for researchers seeking models with greater
descriptive accuracy for asset price behavior.
Diffusion processes have long been used in finance to model the stochastic behavior of asset
prices over time. However, empirical evidence has shown that the basic continuous diffusion
model is unable to fully capture important stylized features of financial time series such as fat
tails and excess kurtosis. To address this shortcoming, researchers have extended the standard
diffusion framework to incorporate jump components. These jump diffusion models provide a
more realistic characterization of asset price dynamics by allowing for discontinuities in price
paths.
In this assignment, I will explore jump diffusion models and their use in finance. Specifically, I
will discuss the motivations for extending standard diffusion processes to include jumps. I will
then introduce some key jump diffusion models including the Merton model and the mixed
exponential jump diffusion model. Finally, I will discuss some empirical applications and
evidence regarding these models. The overall goal is to provide an overview of how
incorporating jump components has improved our ability to model important properties of asset
returns and better match real-world financial data.
Motivation for Jump Diffusion Models
There are a few important empirical observations that motivated the development of jump
diffusion models as an alternative to standard diffusion processes for modeling asset prices:
Excess kurtosis in return distributions: Diffusion processes like Brownian motion yield asset
return distributions that are normally distributed with finite variance. However, empirical
analyses consistently find that return distributions exhibit "fat tails" and excess kurtosis relative
to the normal distribution. This suggests discontinuities or "jumps" may be present.
Volatility clustering: Volatility tends to be clustered over time rather than independent and
identically distributed as assumed by standard diffusion models. Jumps could help account for
sudden increases in volatility.
Feedback effects: Major market events like crashes sometimes coincide with announcements or
macroeconomic shifts that could reasonably be expected to suddenly impact asset valuations.
Jumps allow modeling of such discontinuities.
Implied volatility smiles: Option pricing models based on diffusion processes cannot fully
replicate the implied volatility "smile" patterns seen in market data, where out of the money
options have higher volatilities than would be predicted. Jumps help explain these patterns.
Taken together, these empirical regularities indicate standard diffusion models are too simplistic
and that incorporating random jumps could produce models with superior descriptive power.
Jumps allow for discontinuities that generate fatter tails, clustered volatility, and market events
not easily modeled through smooth continuous paths alone.
Merton Model
One of the earliest and most influential jump diffusion models is the Merton (1976) model.
Unlike standard diffusion processes, the Merton model allows for sudden discontinuous "jumps"
in the process in addition to the continuous diffusion component. Specifically, the model extends
geometric Brownian motion as follows:
dS(t)/S(t) = μdt + σdz(t) + dq(t)
Where:
- S(t) is the asset price at time t
- μ is the drift parameter
- σ is the volatility parameter
- dz(t) is the increment of a Wiener process, representing the continuous diffusion component
- dq(t) represents the discontinuous jump component
Merton assumes the jump component follows a simple compound Poisson process, where
jumps occur randomly according to a Poisson process with intensity λ. The jump size η is
exponentially distributed with parameter ν.
Therefore, between jumps the process behaves like geometric Brownian motion. But at jump
times, the asset price undergoes instantaneous random multiplicative jumps up or down in price
according to the distribution of the jump size η.
A key insight of Merton's model is that it allowed him to develop a closed-form solution for the
value of European options on the asset using a risk-neutral valuation framework, despite the
additional complexity introduced by jumps. This represented an important conceptual and
practical advance, helping to establish jump diffusion models as a potentially useful mainstream
alternative to standard diffusion-based models.
Empirically, Merton-type models have been found to generate more realistic heavy-tailed return
distributions compared to diffusion-only models. They also better fit volatility smiles seen in
option markets. However, the exponential jump size assumption implies a symmetric distribution
which has been criticized as unrealistic and not robustly supported empirically. This motivated
extensions of the basic Merton framework.
Mixed Exponential Jump Diffusion Model
One such extension is the mixed exponential jump diffusion (MEJD) model proposed by Bates
(1996). In the MEJD model, jump sizes are assumed to follow a mixture of two exponential
distributions rather than a single exponential. Specifically:
dS(t)/S(t) = μdt + σdz(t) + dq(t)
Where:
- The jump intensity λ follows a Poisson process as in Merton
- With probability p, jump size η ~ Exp(ν1)
- With probability 1-p, jump size η ~ Exp(ν2)
Allowing for two exponential distributions with different parameters ν1 and ν2 makes the overall
jump size distribution asymmetric, addressing a key limitation of Merton's model. Empirically it is
well established that large downward jumps tend to be more frequent than large upward jumps.
Bates showed the MEJD model also permits a tractable option pricing solution under the risk-
neutral valuation framework. Risk-neutral parameters can be calibrated to market option prices
to infer investor risk perceptions, akin to how implied volatilities are estimated from option prices
under diffusion models.
Empirical tests have found substantial evidence that the MEJD model provides a much better fit
to important patterns in financial time series compared to one-distribution jump models or pure
diffusions. For example, it more accurately replicates the leptokurtic and skewed distribution of
stock returns, as well as the implied volatility smiles and skews observed for equity index
options. The ability to capture asymmetric jump behavior with different tail weights has been an
important empirical success and justification for the mixed distribution approach over Merton's
simpler specification.
Empirical Applications and Evidence
A variety of empirical studies have estimated jump diffusion and more advanced specifications
using financial market and asset return data:
- Bates (2000) estimated the MEJD model using S&P 500 index and individual stock return
data, finding statistically significant evidence of jumps and superior fit relative to standard
diffusion models.
- Eraker et al. (2003) used an MCMC technique to estimate stochastic volatility jump diffusion
models, again finding strong empirical support and dominance over alternatives for S&P 500
and currency returns.
- Broadie et al. (2007) estimated stochastic volatility MEJD models and applied them to option
pricing on 30 stocks in the Dow Jones Industrial Average, finding useful improvements over
pure diffusion models.
- Christoffersen et al. (2010) applied MEJD models augmented with stochastic volatility to hedge
fund return data, providing evidence hedge funds exhibit significant jump components.
- Kawakami (2015) estimated a fractional MEJD model using Japanese stock index data,
demonstrating its ability to capture long memory properties also evident in returns.
These studies consistently report that jump diffusion specifications including MEJD models
provide a substantially improved ability to fit important stylized facts of returns like non-normal
heavy tails, volatility clustering, and kurtosis. They also confirm jumps help replicate key
patterns in option prices like volatility smiles. Overall, there is very strong empirical evidence
that including discontinuities through random jumps enhances the realism and practical
application of stochastic process models in finance.
Conclusion
In conclusion, jump diffusion models provide a significant extension of the standard diffusion
framework used for modeling asset prices and returns. Empirical regularities in financial time
series motivate introducing discontinuous jumps to better characterize properties like heavy
tails, volatility clustering, and option smiles seen in practice. Models like the Merton jump
diffusion and mixed exponential jump diffusion specifications developed by Merton and Bates
are parsimonious yet flexible enough to generate realistic dynamics and tractable pricing
formulas. Numerous empirical studies applying these models to real return data establish their
superior ability to fit and characterize important stylized facts compared to ordinary diffusion
processes alone. While refinements continue to be made, jump diffusion modeling represents
an important conceptual advance and practical tool for researchers seeking models with greater
descriptive accuracy for asset price behavior.
Diffusion processes have long been used in finance to model the stochastic behavior of asset
prices over time. However, empirical evidence has shown that the basic continuous diffusion
model is unable to fully capture important stylized features of financial time series such as fat
tails and excess kurtosis. To address this shortcoming, researchers have extended the standard
diffusion framework to incorporate jump components. These jump diffusion models provide a
more realistic characterization of asset price dynamics by allowing for discontinuities in price
paths.
In this assignment, I will explore jump diffusion models and their use in finance. Specifically, I
will discuss the motivations for extending standard diffusion processes to include jumps. I will
then introduce some key jump diffusion models including the Merton model and the mixed
exponential jump diffusion model. Finally, I will discuss some empirical applications and
evidence regarding these models. The overall goal is to provide an overview of how
incorporating jump components has improved our ability to model important properties of asset
returns and better match real-world financial data.
Motivation for Jump Diffusion Models
There are a few important empirical observations that motivated the development of jump
diffusion models as an alternative to standard diffusion processes for modeling asset prices:
Excess kurtosis in return distributions: Diffusion processes like Brownian motion yield asset
return distributions that are normally distributed with finite variance. However, empirical
analyses consistently find that return distributions exhibit "fat tails" and excess kurtosis relative
to the normal distribution. This suggests discontinuities or "jumps" may be present.
Volatility clustering: Volatility tends to be clustered over time rather than independent and
identically distributed as assumed by standard diffusion models. Jumps could help account for
sudden increases in volatility.
Feedback effects: Major market events like crashes sometimes coincide with announcements or
macroeconomic shifts that could reasonably be expected to suddenly impact asset valuations.
Jumps allow modeling of such discontinuities.
Implied volatility smiles: Option pricing models based on diffusion processes cannot fully
replicate the implied volatility "smile" patterns seen in market data, where out of the money
options have higher volatilities than would be predicted. Jumps help explain these patterns.
Taken together, these empirical regularities indicate standard diffusion models are too simplistic
and that incorporating random jumps could produce models with superior descriptive power.
Jumps allow for discontinuities that generate fatter tails, clustered volatility, and market events
not easily modeled through smooth continuous paths alone.
Merton Model
One of the earliest and most influential jump diffusion models is the Merton (1976) model.
Unlike standard diffusion processes, the Merton model allows for sudden discontinuous "jumps"
in the process in addition to the continuous diffusion component. Specifically, the model extends
geometric Brownian motion as follows:
dS(t)/S(t) = μdt + σdz(t) + dq(t)
Where:
- S(t) is the asset price at time t
- μ is the drift parameter
- σ is the volatility parameter
- dz(t) is the increment of a Wiener process, representing the continuous diffusion component
- dq(t) represents the discontinuous jump component
Merton assumes the jump component follows a simple compound Poisson process, where
jumps occur randomly according to a Poisson process with intensity λ. The jump size η is
exponentially distributed with parameter ν.
Therefore, between jumps the process behaves like geometric Brownian motion. But at jump
times, the asset price undergoes instantaneous random multiplicative jumps up or down in price
according to the distribution of the jump size η.
A key insight of Merton's model is that it allowed him to develop a closed-form solution for the
value of European options on the asset using a risk-neutral valuation framework, despite the
additional complexity introduced by jumps. This represented an important conceptual and
practical advance, helping to establish jump diffusion models as a potentially useful mainstream
alternative to standard diffusion-based models.
Empirically, Merton-type models have been found to generate more realistic heavy-tailed return
distributions compared to diffusion-only models. They also better fit volatility smiles seen in
option markets. However, the exponential jump size assumption implies a symmetric distribution
which has been criticized as unrealistic and not robustly supported empirically. This motivated
extensions of the basic Merton framework.
Mixed Exponential Jump Diffusion Model
One such extension is the mixed exponential jump diffusion (MEJD) model proposed by Bates
(1996). In the MEJD model, jump sizes are assumed to follow a mixture of two exponential
distributions rather than a single exponential. Specifically:
dS(t)/S(t) = μdt + σdz(t) + dq(t)
Where:
- The jump intensity λ follows a Poisson process as in Merton
- With probability p, jump size η ~ Exp(ν1)
- With probability 1-p, jump size η ~ Exp(ν2)
Allowing for two exponential distributions with different parameters ν1 and ν2 makes the overall
jump size distribution asymmetric, addressing a key limitation of Merton's model. Empirically it is
well established that large downward jumps tend to be more frequent than large upward jumps.
Bates showed the MEJD model also permits a tractable option pricing solution under the risk-
neutral valuation framework. Risk-neutral parameters can be calibrated to market option prices
to infer investor risk perceptions, akin to how implied volatilities are estimated from option prices
under diffusion models.
Empirical tests have found substantial evidence that the MEJD model provides a much better fit
to important patterns in financial time series compared to one-distribution jump models or pure
diffusions. For example, it more accurately replicates the leptokurtic and skewed distribution of
stock returns, as well as the implied volatility smiles and skews observed for equity index
options. The ability to capture asymmetric jump behavior with different tail weights has been an
important empirical success and justification for the mixed distribution approach over Merton's
simpler specification.
Empirical Applications and Evidence
A variety of empirical studies have estimated jump diffusion and more advanced specifications
using financial market and asset return data:
- Bates (2000) estimated the MEJD model using S&P 500 index and individual stock return
data, finding statistically significant evidence of jumps and superior fit relative to standard
diffusion models.
- Eraker et al. (2003) used an MCMC technique to estimate stochastic volatility jump diffusion
models, again finding strong empirical support and dominance over alternatives for S&P 500
and currency returns.
- Broadie et al. (2007) estimated stochastic volatility MEJD models and applied them to option
pricing on 30 stocks in the Dow Jones Industrial Average, finding useful improvements over
pure diffusion models.
- Christoffersen et al. (2010) applied MEJD models augmented with stochastic volatility to hedge
fund return data, providing evidence hedge funds exhibit significant jump components.
- Kawakami (2015) estimated a fractional MEJD model using Japanese stock index data,
demonstrating its ability to capture long memory properties also evident in returns.
These studies consistently report that jump diffusion specifications including MEJD models
provide a substantially improved ability to fit important stylized facts of returns like non-normal
heavy tails, volatility clustering, and kurtosis. They also confirm jumps help replicate key
patterns in option prices like volatility smiles. Overall, there is very strong empirical evidence
that including discontinuities through random jumps enhances the realism and practical
application of stochastic process models in finance.
Conclusion
In conclusion, jump diffusion models provide a significant extension of the standard diffusion
framework used for modeling asset prices and returns. Empirical regularities in financial time
series motivate introducing discontinuous jumps to better characterize properties like heavy
tails, volatility clustering, and option smiles seen in practice. Models like the Merton jump
diffusion and mixed exponential jump diffusion specifications developed by Merton and Bates
are parsimonious yet flexible enough to generate realistic dynamics and tractable pricing
formulas. Numerous empirical studies applying these models to real return data establish their
superior ability to fit and characterize important stylized facts compared to ordinary diffusion
processes alone. While refinements continue to be made, jump diffusion modeling represents
an important conceptual advance and practical tool for researchers seeking models with greater
descriptive accuracy for asset price behavior.
Diffusion processes have long been used in finance to model the stochastic behavior of asset
prices over time. However, empirical evidence has shown that the basic continuous diffusion
model is unable to fully capture important stylized features of financial time series such as fat
tails and excess kurtosis. To address this shortcoming, researchers have extended the standard
diffusion framework to incorporate jump components. These jump diffusion models provide a
more realistic characterization of asset price dynamics by allowing for discontinuities in price
paths.
In this assignment, I will explore jump diffusion models and their use in finance. Specifically, I
will discuss the motivations for extending standard diffusion processes to include jumps. I will
then introduce some key jump diffusion models including the Merton model and the mixed
exponential jump diffusion model. Finally, I will discuss some empirical applications and
evidence regarding these models. The overall goal is to provide an overview of how
incorporating jump components has improved our ability to model important properties of asset
returns and better match real-world financial data.
Motivation for Jump Diffusion Models
There are a few important empirical observations that motivated the development of jump
diffusion models as an alternative to standard diffusion processes for modeling asset prices:
Excess kurtosis in return distributions: Diffusion processes like Brownian motion yield asset
return distributions that are normally distributed with finite variance. However, empirical
analyses consistently find that return distributions exhibit "fat tails" and excess kurtosis relative
to the normal distribution. This suggests discontinuities or "jumps" may be present.
Volatility clustering: Volatility tends to be clustered over time rather than independent and
identically distributed as assumed by standard diffusion models. Jumps could help account for
sudden increases in volatility.
Feedback effects: Major market events like crashes sometimes coincide with announcements or
macroeconomic shifts that could reasonably be expected to suddenly impact asset valuations.
Jumps allow modeling of such discontinuities.
Implied volatility smiles: Option pricing models based on diffusion processes cannot fully
replicate the implied volatility "smile" patterns seen in market data, where out of the money
options have higher volatilities than would be predicted. Jumps help explain these patterns.
Taken together, these empirical regularities indicate standard diffusion models are too simplistic
and that incorporating random jumps could produce models with superior descriptive power.
Jumps allow for discontinuities that generate fatter tails, clustered volatility, and market events
not easily modeled through smooth continuous paths alone.
Merton Model
One of the earliest and most influential jump diffusion models is the Merton (1976) model.
Unlike standard diffusion processes, the Merton model allows for sudden discontinuous "jumps"
in the process in addition to the continuous diffusion component. Specifically, the model extends
geometric Brownian motion as follows:
dS(t)/S(t) = μdt + σdz(t) + dq(t)
Where:
- S(t) is the asset price at time t
- μ is the drift parameter
- σ is the volatility parameter
- dz(t) is the increment of a Wiener process, representing the continuous diffusion component
- dq(t) represents the discontinuous jump component
Merton assumes the jump component follows a simple compound Poisson process, where
jumps occur randomly according to a Poisson process with intensity λ. The jump size η is
exponentially distributed with parameter ν.
Therefore, between jumps the process behaves like geometric Brownian motion. But at jump
times, the asset price undergoes instantaneous random multiplicative jumps up or down in price
according to the distribution of the jump size η.
A key insight of Merton's model is that it allowed him to develop a closed-form solution for the
value of European options on the asset using a risk-neutral valuation framework, despite the
additional complexity introduced by jumps. This represented an important conceptual and
practical advance, helping to establish jump diffusion models as a potentially useful mainstream
alternative to standard diffusion-based models.
Empirically, Merton-type models have been found to generate more realistic heavy-tailed return
distributions compared to diffusion-only models. They also better fit volatility smiles seen in
option markets. However, the exponential jump size assumption implies a symmetric distribution
which has been criticized as unrealistic and not robustly supported empirically. This motivated
extensions of the basic Merton framework.
Mixed Exponential Jump Diffusion Model
One such extension is the mixed exponential jump diffusion (MEJD) model proposed by Bates
(1996). In the MEJD model, jump sizes are assumed to follow a mixture of two exponential
distributions rather than a single exponential. Specifically:
dS(t)/S(t) = μdt + σdz(t) + dq(t)
Where:
- The jump intensity λ follows a Poisson process as in Merton
- With probability p, jump size η ~ Exp(ν1)
- With probability 1-p, jump size η ~ Exp(ν2)
Allowing for two exponential distributions with different parameters ν1 and ν2 makes the overall
jump size distribution asymmetric, addressing a key limitation of Merton's model. Empirically it is
well established that large downward jumps tend to be more frequent than large upward jumps.
Bates showed the MEJD model also permits a tractable option pricing solution under the risk-
neutral valuation framework. Risk-neutral parameters can be calibrated to market option prices
to infer investor risk perceptions, akin to how implied volatilities are estimated from option prices
under diffusion models.
Empirical tests have found substantial evidence that the MEJD model provides a much better fit
to important patterns in financial time series compared to one-distribution jump models or pure
diffusions. For example, it more accurately replicates the leptokurtic and skewed distribution of
stock returns, as well as the implied volatility smiles and skews observed for equity index
options. The ability to capture asymmetric jump behavior with different tail weights has been an
important empirical success and justification for the mixed distribution approach over Merton's
simpler specification.
Empirical Applications and Evidence
A variety of empirical studies have estimated jump diffusion and more advanced specifications
using financial market and asset return data:
- Bates (2000) estimated the MEJD model using S&P 500 index and individual stock return
data, finding statistically significant evidence of jumps and superior fit relative to standard
diffusion models.
- Eraker et al. (2003) used an MCMC technique to estimate stochastic volatility jump diffusion
models, again finding strong empirical support and dominance over alternatives for S&P 500
and currency returns.
- Broadie et al. (2007) estimated stochastic volatility MEJD models and applied them to option
pricing on 30 stocks in the Dow Jones Industrial Average, finding useful improvements over
pure diffusion models.
- Christoffersen et al. (2010) applied MEJD models augmented with stochastic volatility to hedge
fund return data, providing evidence hedge funds exhibit significant jump components.
- Kawakami (2015) estimated a fractional MEJD model using Japanese stock index data,
demonstrating its ability to capture long memory properties also evident in returns.
These studies consistently report that jump diffusion specifications including MEJD models
provide a substantially improved ability to fit important stylized facts of returns like non-normal
heavy tails, volatility clustering, and kurtosis. They also confirm jumps help replicate key
patterns in option prices like volatility smiles. Overall, there is very strong empirical evidence
that including discontinuities through random jumps enhances the realism and practical
application of stochastic process models in finance.
Conclusion
In conclusion, jump diffusion models provide a significant extension of the standard diffusion
framework used for modeling asset prices and returns. Empirical regularities in financial time
series motivate introducing discontinuous jumps to better characterize properties like heavy
tails, volatility clustering, and option smiles seen in practice. Models like the Merton jump
diffusion and mixed exponential jump diffusion specifications developed by Merton and Bates
are parsimonious yet flexible enough to generate realistic dynamics and tractable pricing
formulas. Numerous empirical studies applying these models to real return data establish their
superior ability to fit and characterize important stylized facts compared to ordinary diffusion
processes alone. While refinements continue to be made, jump diffusion modeling represents
an important conceptual advance and practical tool for researchers seeking models with greater
descriptive accuracy for asset price behavior.
Diffusion processes have long been used in finance to model the stochastic behavior of asset
prices over time. However, empirical evidence has shown that the basic continuous diffusion
model is unable to fully capture important stylized features of financial time series such as fat
tails and excess kurtosis. To address this shortcoming, researchers have extended the standard
diffusion framework to incorporate jump components. These jump diffusion models provide a
more realistic characterization of asset price dynamics by allowing for discontinuities in price
paths.
In this assignment, I will explore jump diffusion models and their use in finance. Specifically, I
will discuss the motivations for extending standard diffusion processes to include jumps. I will
then introduce some key jump diffusion models including the Merton model and the mixed
exponential jump diffusion model. Finally, I will discuss some empirical applications and
evidence regarding these models. The overall goal is to provide an overview of how
incorporating jump components has improved our ability to model important properties of asset
returns and better match real-world financial data.
Motivation for Jump Diffusion Models
There are a few important empirical observations that motivated the development of jump
diffusion models as an alternative to standard diffusion processes for modeling asset prices:
Excess kurtosis in return distributions: Diffusion processes like Brownian motion yield asset
return distributions that are normally distributed with finite variance. However, empirical
analyses consistently find that return distributions exhibit "fat tails" and excess kurtosis relative
to the normal distribution. This suggests discontinuities or "jumps" may be present.
Volatility clustering: Volatility tends to be clustered over time rather than independent and
identically distributed as assumed by standard diffusion models. Jumps could help account for
sudden increases in volatility.
Feedback effects: Major market events like crashes sometimes coincide with announcements or
macroeconomic shifts that could reasonably be expected to suddenly impact asset valuations.
Jumps allow modeling of such discontinuities.
Implied volatility smiles: Option pricing models based on diffusion processes cannot fully
replicate the implied volatility "smile" patterns seen in market data, where out of the money
options have higher volatilities than would be predicted. Jumps help explain these patterns.
Taken together, these empirical regularities indicate standard diffusion models are too simplistic
and that incorporating random jumps could produce models with superior descriptive power.
Jumps allow for discontinuities that generate fatter tails, clustered volatility, and market events
not easily modeled through smooth continuous paths alone.
Merton Model
One of the earliest and most influential jump diffusion models is the Merton (1976) model.
Unlike standard diffusion processes, the Merton model allows for sudden discontinuous "jumps"
in the process in addition to the continuous diffusion component. Specifically, the model extends
geometric Brownian motion as follows:
dS(t)/S(t) = μdt + σdz(t) + dq(t)
Where:
- S(t) is the asset price at time t
- μ is the drift parameter
- σ is the volatility parameter
- dz(t) is the increment of a Wiener process, representing the continuous diffusion component
- dq(t) represents the discontinuous jump component
Merton assumes the jump component follows a simple compound Poisson process, where
jumps occur randomly according to a Poisson process with intensity λ. The jump size η is
exponentially distributed with parameter ν.
Therefore, between jumps the process behaves like geometric Brownian motion. But at jump
times, the asset price undergoes instantaneous random multiplicative jumps up or down in price
according to the distribution of the jump size η.
A key insight of Merton's model is that it allowed him to develop a closed-form solution for the
value of European options on the asset using a risk-neutral valuation framework, despite the
additional complexity introduced by jumps. This represented an important conceptual and
practical advance, helping to establish jump diffusion models as a potentially useful mainstream
alternative to standard diffusion-based models.
Empirically, Merton-type models have been found to generate more realistic heavy-tailed return
distributions compared to diffusion-only models. They also better fit volatility smiles seen in
option markets. However, the exponential jump size assumption implies a symmetric distribution
which has been criticized as unrealistic and not robustly supported empirically. This motivated
extensions of the basic Merton framework.
Mixed Exponential Jump Diffusion Model
One such extension is the mixed exponential jump diffusion (MEJD) model proposed by Bates
(1996). In the MEJD model, jump sizes are assumed to follow a mixture of two exponential
distributions rather than a single exponential. Specifically:
dS(t)/S(t) = μdt + σdz(t) + dq(t)
Where:
- The jump intensity λ follows a Poisson process as in Merton
- With probability p, jump size η ~ Exp(ν1)
- With probability 1-p, jump size η ~ Exp(ν2)
Allowing for two exponential distributions with different parameters ν1 and ν2 makes the overall
jump size distribution asymmetric, addressing a key limitation of Merton's model. Empirically it is
well established that large downward jumps tend to be more frequent than large upward jumps.
Bates showed the MEJD model also permits a tractable option pricing solution under the risk-
neutral valuation framework. Risk-neutral parameters can be calibrated to market option prices
to infer investor risk perceptions, akin to how implied volatilities are estimated from option prices
under diffusion models.
Empirical tests have found substantial evidence that the MEJD model provides a much better fit
to important patterns in financial time series compared to one-distribution jump models or pure
diffusions. For example, it more accurately replicates the leptokurtic and skewed distribution of
stock returns, as well as the implied volatility smiles and skews observed for equity index
options. The ability to capture asymmetric jump behavior with different tail weights has been an
important empirical success and justification for the mixed distribution approach over Merton's
simpler specification.
Empirical Applications and Evidence
A variety of empirical studies have estimated jump diffusion and more advanced specifications
using financial market and asset return data:
- Bates (2000) estimated the MEJD model using S&P 500 index and individual stock return
data, finding statistically significant evidence of jumps and superior fit relative to standard
diffusion models.
- Eraker et al. (2003) used an MCMC technique to estimate stochastic volatility jump diffusion
models, again finding strong empirical support and dominance over alternatives for S&P 500
and currency returns.
- Broadie et al. (2007) estimated stochastic volatility MEJD models and applied them to option
pricing on 30 stocks in the Dow Jones Industrial Average, finding useful improvements over
pure diffusion models.
- Christoffersen et al. (2010) applied MEJD models augmented with stochastic volatility to hedge
fund return data, providing evidence hedge funds exhibit significant jump components.
- Kawakami (2015) estimated a fractional MEJD model using Japanese stock index data,
demonstrating its ability to capture long memory properties also evident in returns.
These studies consistently report that jump diffusion specifications including MEJD models
provide a substantially improved ability to fit important stylized facts of returns like non-normal
heavy tails, volatility clustering, and kurtosis. They also confirm jumps help replicate key
patterns in option prices like volatility smiles. Overall, there is very strong empirical evidence
that including discontinuities through random jumps enhances the realism and practical
application of stochastic process models in finance.
Conclusion
In conclusion, jump diffusion models provide a significant extension of the standard diffusion
framework used for modeling asset prices and returns. Empirical regularities in financial time
series motivate introducing discontinuous jumps to better characterize properties like heavy
tails, volatility clustering, and option smiles seen in practice. Models like the Merton jump
diffusion and mixed exponential jump diffusion specifications developed by Merton and Bates
are parsimonious yet flexible enough to generate realistic dynamics and tractable pricing
formulas. Numerous empirical studies applying these models to real return data establish their
superior ability to fit and characterize important stylized facts compared to ordinary diffusion
processes alone. While refinements continue to be made, jump diffusion modeling represents
an important conceptual advance and practical tool for researchers seeking models with greater
descriptive accuracy for asset price behavior.
