SARIMAX simulate

[David Waterworth]

The R forecast package has a simulate() which allows monte-carlo simulation of a SARIMA model, i.e

sim <- ts(matrix(0, nrow=30, ncol=10000), start=end(residuals)[1]+1)

for(i in seq(10000))

  sim[,i] <- simulate(fit, nsim=30)

simulates 10000 paths, nsteps=30

Does the statsmodels SARIMAX have the same functionality?

Dave

[josef...@gmail.com]

I don't know what simulation methods are in the statespace models, 

but there is an old function for simulating ARMA processes.

https://www.statsmodels.org/stable/generated/statsmodels.tsa.arima_process.arma_generate_sample.html 

Josef 

 

Dave

--
You received this message because you are subscribed to the Google Groups "pystatsmodels" group.
To unsubscribe from this group and stop receiving emails from it, send an email to pystatsmodel...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/pystatsmodels/40210055-1a40-463c-907d-59e68d00aea1%40googlegroups.com.

[Chad Fulton]

On Fri, Oct 18, 2019 at 1:37 PM <josef...@gmail.com> wrote:

On Fri, Oct 18, 2019 at 4:17 AM David Waterworth <david.wa...@cimenviro.com> wrote:

The R forecast package has a simulate() which allows monte-carlo simulation of a SARIMA model, i.e

sim <- ts(matrix(0, nrow=30, ncol=10000), start=end(residuals)[1]+1)

for(i in seq(10000))

  sim[,i] <- simulate(fit, nsim=30)

simulates 10000 paths, nsteps=30

Does the statsmodels SARIMAX have the same functionality?

I don't know what simulation methods are in the statespace models, 

but there is an old function for simulating ARMA processes.

https://www.statsmodels.org/stable/generated/statsmodels.tsa.arima_process.arma_generate_sample.html 

Josef 

 

Unfortunately this is an area which has not received enough development attention for the state space models. There does exist a `simulate` method, but it currently is not very user friendly. Its default is to simulate a *new* dataset following the current state space model, which is not the same thing as simulating *additional* datapoints after the existing dataset, and unfortunately the current interface does not make it entirely straightforward to do the latter.

However, you can do it, e.g. as follows:

# Create and fit the model

mod = sm.tsa.SARIMAX(endog, <your spec here>)

res = mod.fit()

# Create a faux model to use for simulation

mod_sim = sm.tsa.SARIMAX([0] * 30, <your spec here>)

mod_sim.initialize_known(res.predicted_state[..., -1], res.predicted_state_cov[..., -1])

sim = np.zeros((30, 10000))
for i in range(10000):
    sim[:, i] = mod_sim.simulate(res.params, 30)

Best,

Chad

[David Waterworth]

Thanks Chad

I ended up using R's sim.ssarima from the smooth package as I just wanted some quick test data - but thatks for the tip

Regards

David