1. Home
  2. TSUMMARIZE directive

TSUMMARIZE directive

Displays characteristics of time series models.


PRINT = string tokens What to print (autocorrelations, expansion, impulse, piweight, psiweight); default *
GRAPH = string tokens What to display with graphs (autocorrelations, impulse, piweight, psiweight); default *
MAXLAG = scalar Maximum lag for results; default 30


TSM = TSMs Models to be displayed
AUTOCORRELATIONS = variates To save theoretical autocorrelations
IMPULSERESPONSE = variates To save impulse-response function
STEPFUNCTION = variates To save step function from impulse
PIWEIGHTS = variates To save pi-weights
PSIWEIGHTS = variates To save psi-weights
EXPANSION = TSMs To save expanded models
VARIANCE = scalars To save variance of each TSM


The TSUMMARIZE directive helps you investigate time-series models by displaying or saving various characteristics. These are the theoretical autocorrelation function of an ARIMA model, and the pi-weights and psi-weights; also the impulse-response function of a transfer-function model. TSUMMARIZE can derive the expanded form of a model, in which all seasonal terms are combined with the non-seasonal term.

For an ARIMA model in the TSM parameter, you can set only the AUTOCORRELATIONS, PSIWEIGHTS and PIWEIGHTS parameters. Also, you can set the IMPULSERESPONSE parameter only for a transfer-function model. You can set the EXPAND parameter for either type of model. The TSMs in any TSUMMARIZE statement must be completely defined; that is, you must have set the orders and parameters, and the lags if you are using them. The only exceptions are that Genstat takes the transformation parameter to be 1.0 if it is missing, and that the innovation variance of an ARIMA model need not be set.

The MAXLAG option specifies the maximum lag to which Genstat is to do calculations: this applies to autocorrelations, psi-weights, pi-weights and impulse responses. If MAXLAG is unset, the maximum lag is defined implicitly as the length of the first variate in the parameters. However, if the length of this variate is also undefined, the maximum lag cannot be defined and Genstat reports a fault.

You can set the PRINT and GRAPH options independently of the parameters: these store results, and display the various characteristics of models.

The AUTOCORRELATIONS parameter allows you to store the theoretical autocorrelation function of an ARIMA model. Such a model uniquely defines an autocorrelation function whose values r0rm are assigned by Genstat to the variate R, where m is the maximum lag. If the model has differencing parameters d=D=0, then the autocorrelation function is that of a series yt that follows this model.

If either d>0 or D>0, then the theoretical autocorrelations are calculated as if d=D=0, and so they correspond to those of the differenced yt series. This is because the autocorrelations of yt are undefined for non-stationary models.

The PSIWEIGHTS parameter allows you to store the theoretical psi-weights ψ0 … ψm of an ARIMA model. These are used internally by Genstat when error limits are calculated for forecasts obtained using the model. You will need them for example if you want to calculate the variance of the total of the forecast values up to some specified maximum lead time. They are defined for a non-seasonal model by

1 + ψ1B + ψ2B2 + … = θ(B) / { φ(B)∇d }

The PIWEIGHTS parameter allows you to store the theoretical pi-weights π0 … πm of an ARIMA model: these show explicitly how past values contribute to a forecast. The weights are defined by:

1 – π1B – π2B2 – … = { φ(B)∇d } / θ(B)

The IMPULSERESPONSE parameter allows you to store the theoretical impulse-response function, v0vm, of a transfer-function model. This function can help you interpret the model. The sequence is defined for a non-seasonal transfer-function model by:

ν0 + ν1B + ν2B2 + … = ω(B)Bb / { δ(B)∇d }

For an ARIMA model you can combine into one generalized autoregressive operator all the differencing operators, the non-seasonal autoregressive operators, and the seasonal autoregressive operators. The non-seasonal and seasonal moving-average operators may similarly be combined. This expanded model can be printed using the expansion setting of PRINT and saved using the EXPANSION parameter. It can be used to help you understand a series. But you might also want to re-estimate the parameters in the expanded model, to test whether the differencing operators or seasonal factors unnecessarily constrain the structure of the original model. If you have not previously defined one of the identifiers supplied by the EXPANSION parameter, Genstat will automatically define it to be a TSM, and its component variates will be set up to have the length defined by the corresponding model in the TSM parameter. The expansion does not change the transformation parameter of the model, nor the constant term, nor the innovation variance. If the model that you have supplied contains non-zero differencing orders, then the generalized model does not satisfy the stationarity constraint on the parameters; neither does the constant term have the same interpretation as it had in the supplied model. The expansion of transfer-function models exactly parallels that of ARIMA models.



See also



Commands for: Time series.


" Examples 2:7.7.3-4 "
" Display the autocorrelations of an AR[2] model."
TSM        AR[2]; ORDERS=!(2,0,0); PARAMETERS=!(1,15,2.5,0.5,-0.5)
TSUMMARIZE [MAXLAG=12; PRINT=autocorrelations] AR[2]
" Expand the seasonal ARIMA model used for modelling the number of
  airline passengers in Section 7.3.7."
VARIATE    [VALUES=0,1,1, 0,1,1,12] Ord
&          [VALUES=0,0,0.00143, 0.34, 0.54] Par
TSM        Airpass; ORDERS=Ord; PARAMETERS=Par
PRINT      Airpass
TSUMMARIZE [PRINT=expansion] Airpass
Updated on June 17, 2019

Was this article helpful?