This vignette^{1} discusses model checking in
`{fHMM}`

, that means the task of verifying whether the fitted
model describes the data well.

Since the observations are explained by different distributions (depending on the active state), model checking cannot be done by analyzing standard residuals. Instead, we consider “pseudo-residuals”. To transform all observations on a common scale, we proceed as follows: If \(X_t\) has the invertible distribution function \(F_{X_t}\), then

\[\begin{align*} Z_t=\Phi^{-1}(F_{X_t} (X_t)) \end{align*}\]

is standard normally distributed, where \(\Phi\) denotes the cumulative distribution function of the standard normal distribution. The observations, \((X_t)_t\), are modeled well if the so-called pseudo-residuals, \((Z_t)_t\), are approximately standard normally distributed, which can be visually assessed using quantile-quantile plots or further investigated using statistical tests such as the Jarque-Bera test (Zucchini, MacDonald, and Langrock 2016).

For HHMMs, we first decode the coarse-scale state process using the Viterbi algorithm. Subsequently, we assign each coarse-scale observation its distribution function under the fitted model and perform the transformation described above. Using the Viterbi-decoded coarse-scale states, we then treat the fine-scale observations analogously.

In `{fHMM}`

, pseudo-residuals can be computed via the
`compute_residuals()`

function, provided that the states have
been decoded beforehand.

We revisit the DAX example:

The following line computes the residuals and saves them into the
`model`

object:

The residuals can be visualized as follows:

For additional normality tests, the residuals can be extracted from
the `model`

object via the `residuals()`

method.
The following lines exemplary perform a Jarque–Bera
test (Jarque and Bera
1987):

Jarque, C. M., and A. K. Bera. 1987. “A Test for Normality of
Observations and Regression Residuals.” *International
Statistical Review / Revue Internationale de Statistique* 55:
163–72.

Zucchini, W., I. L. MacDonald, and R. Langrock. 2016. “Hidden
Markov Models for Time Series: An Introduction Using R, 2nd
Edition.” *Chapman and Hall/CRC*.

This vignette was build using R 4.4.0 with the

`{fHMM}`

1.4.1 package.↩︎