StochasticVolatilitySVARResult.identification_test#

StochasticVolatilitySVARResult.identification_test(*, n_blocks=10, window=9)[source]#

Test the identifying condition pair by pair on the posterior-median shocks.

Identification here rests on the shocks’ variance paths not being proportional. That is not something the posterior can certify: with proportional paths the likelihood is flat over rotations, yet the volatility prior still hands back a sharp posterior at the rotation whose shocks look most like clustered volatility. So the condition is tested on the data instead. Under proportionality the ratio of two shocks’ variances is constant over time whatever the common path does; the statistic is the precision-weighted dispersion of block-wise log variance ratios, chi-squared with n_blocks - 1 degrees of freedom under the null. A large p-value for any pair means the data cannot tell those two shocks’ rotation apart, and the summary says so.

Parameters:

  • n_blocks (int) – Contiguous sample blocks, at least 3.

  • window (int) – Odd length of the moving average that proxies the pair’s common variance when weighting blocks.

Return type:

tuple[NDArray[double], NDArray[double]]

Returns:

(statistics, p_values), each (k, k) symmetric.

Raises:

SpecificationError – If the settings are malformed for the sample.