As we know, solution to $dX_t=\mu dt+\sigma dW_t$ is normal distributed and is light tailed; solution to $dX_t=\mu X_tdt+\sigma X_t dW_t$ is log-normal distributed and is heavy tailed. Is there any reference on discussion of criteria to identify the tail behavior from the driven SDE?
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
You want to consider:
$$dx_t=f(x_t)dt + \sigma(x_t)dW_t$$
In dimension 1, you can express analytically the stationary probability density (if it exists) as:
$$P(x)=\frac{1}{Z}\exp(-\int^x \frac{f(u)}{\sigma(u)^2} du)$$ where $Z$ is a normalizing constant.
Then you can express the various stationary moments using this formula and investigate the tail properties.
