Statscraft 0.15.0, release v0.15.0
ACF
Autocorrelations of time series, with standard errors, Box-Ljung tests and charts.
Syntax
ACF [VARIABLES=] series [/DIFF=n] [/SDIFF=n] [/PERIOD=n] [/LN | /NOLOG] [/SEASONAL]
[/MXAUTO=n] [/SERROR=IND | MA] [/PACF] [/APPLY].
Before the correlations each series may be logged (LN; a series with a value not above 0
is skipped), differenced DIFF times (DIFF alone: once) and then differenced over the period
SDIFF times. The period comes from PERIOD (at least 2), else TSET PERIOD, else the DATE
variables; without one SDIFF and SEASONAL are left out with a warning. MXAUTO (else TSET
MXAUTO, else 16) is the highest lag, at most the number of values less one; SEASONAL shows
only the lags that are multiples of the period. A missing value (user-missing too, unless
TSET MISSING=INCLUDE) drops the products it is part of; the mean and the counts use the
valid values. Each series gets a table: autocorrelation r_k, its standard error, and the
Box-Ljung statistic of lags 1 to k with k df and its significance. IND (the default, or
TSET ACFSE) takes the standard error of white noise, sqrt((n - k) / (n (n + 2))); MA takes
Bartlett's sqrt((1 + 2 (r_1^2 + ... + r_(k-1)^2)) / n). PACF adds the partial
autocorrelations, with standard error 1 / sqrt(n). A chart per series shows the
coefficients with limits at two standard errors. APPLY without a name repeats the previous
ACF, with the subcommands given now replacing its own; named models are not supported.
DIFFERENCE is another name for DIFF. Case weights are not used.
Example
ACF VARIABLES=daily_orders /DIFF /MXAUTO=21 /PACF.
statscraft help ACF prints the same text in the terminal. All commands