Note: This page has been translated by MathWorks. Please click here

To view all translated materials including this page, select Japan from the country navigator on the bottom of this page.

To view all translated materials including this page, select Japan from the country navigator on the bottom of this page.

Traffic light test for value-at-risk (VaR) backtesting

`TestResults = tl(vbt)`

generates the traffic light (TL) test for value-at-risk (VaR)
backtesting.`TestResults`

= tl(`vbt`

)

The traffic light test is based on a binomial distribution. Suppose
`N`

is the number of observations, `p`

= 1 −
`VaRLevel`

is the probability of observing a failure if the
model is correct, and *x* is the number of failures.

The test computes the cumulative probability of observing up to
*x* failures, reported in the `'Probability'`

column,

$$Probability=Probability(X\le x|N,p)=F(x|N,p)$$

where $$F(x|N,p)$$ is the cumulative distribution of a binomial variable with
parameters *N* and *p*, with *p*
= 1 − *VaRLevel*. The three zones are defined based on this
cumulative probability:

Green: $$F(x|N,p)$$ ≤

`0.95`

Yellow:

`0.95`

< $$F(x|N,p)$$ ≤`0.9999`

Red:

`0.9999`

< $$F(x|N,p)$$

The probability of a Type-I error, reported in the `'TypeI'`

column, is $$TypeI=TypeI(x|N,p)=1-F(X\ge x|N,p)$$.

This probability corresponds to the probability of mistakenly rejecting the model
if the model were correct. *Probability* and
*TypeI* do not sum up to 1, they exceed 1 by exactly the
probability of having *x* failures.

The increase in scaling factor, reported in the `'Increase'`

column, is always `0`

for the `green`

zone and
always `1`

for the `red`

zone. For the
`yellow`

zone, it is an adjustment based on the relative
difference between the assumed VaR confidence level (*VaRLevel*)
and the observed confidence level (*x* / *N*),
where `N`

is the number of observations and*x* is
the number of failures. To find the increase under the assumption of a normal
distribution, compute the critical values *zAssumed* and
*zObserved*.

The increase to the baseline scaling factor is given by

$$Increase=Baseline\times \left(\frac{zAssumed}{zObserved}-1\right)$$

with the restriction that the increase cannot be negative or greater than
`1`

. The baseline scaling factor in the Basel rules is
3.

The `tl`

function computes the scaling factor following this
methodology, which is also described in the Basel document (see References). The
`tl`

function does not apply any ad-hoc adjustments.

[1] Basel Committee on Banking Supervision, *Supervisory Framework
for the Use of 'Backtesting' in Conjunction with the Internal Models Approach to
Market Risk Capital Requirements.* January, 1996, http://www.bis.org/publ/bcbs22.htm.

Was this topic helpful?