The filtering test shows inconsistent results for both the twice integrated Brownian Motion (i2bm) and the twice integrated OU (i2ou) SDEs. Possibly a bug in the end point proposals, since there are examples of worse filtering mean estimation performance than the bootstrap, even for the first component (that is observed with low noise).
The filtering test shows inconsistent results for both the twice integrated Brownian Motion (i2bm) and the twice integrated OU (i2ou) SDEs. Possibly a bug in the end point proposals, since there are examples of worse filtering mean estimation performance than the bootstrap, even for the first component (that is observed with low noise).