Abstract
Systemic failure problems are hot topics in econophysics,
financial engineering and computer engineering. For a long
time, this area lacks good data and it was difficult to
examine their modelings. Recently, empirical data analysis
of correlated failures in Wide-area network systems and the
researches on market quotes of credit derivatives have clarified
their statistical properties. The probability function
for the number of failures in storage systems has a long
tail, which means that the failures are highly correlated.
The implied loss functions of credit portfolios, which are
estimated based on the market quotes of iTraxx-CJ, CDX-IG
and iTraxx-Europe, also have the same nature. Credit markets
expect that the defaults of the assets in the portfolio do
not occur independently and their correlations are relatively
strong.
In this paper, we would like to review some results on a
method to study the correlated failure data. In addition, we
compare some theoretical models and understand their behaviors.
At first, we show the general method to construct correlated
binomial models. We introduce conditional correlations $\rho_ij$
and expectations $p_ij$, where the suffix $_ij$
means the condition that $i (j)$ of $N$ variables take $1 (resp. 0)$.
Based on them, we show how to construct
joint probability function and derive recursive relations,
which are necessary conditions to ensure the probability conservation
of the model. Next, by using the recursive relations we show
how to calibrate the correlation structures ($\rho_ij$ and $p_ij$)
based on the probability function. By the method, we compare
the correlation structures of several probabilistic models.
In addition, we also compare those of empirical probability
functions and discuss the physical mechanism to induce them.
References:\\
1) M.Hisakado, K.Kistukawa and S.Mori, J.Phys.A:Math.Gen.39(2006)
15365-15378. \\
2) S. Mori, K. Kitsukawa and M. Hisakado, Default Distribution and Credit
Market Implications (arXiv:physics0609093).\\
3) S.Mori, K.Kitsukawa, M.Hisakado,
Moody's Correlated Binomial Default Distributions for Inhomogeneous
Portfolios (arXiv:physics0603036).
Users
Please
log in to take part in the discussion (add own reviews or comments).