Comparison of the models
Concentration data
Its ACF
Its PACF. AR(2)?
AR(1)? m = 17.06, p-value a 1 = , p-value Portmanteau test 47.44, p-value 0.003
AR(2)? m = 17.06, p-value a 1 = , p-value a 2 = , p-value Portmanteau test 26.97, p-value
AR(3)? m = 17.06, p-value a 1 = , p-value a 2 = , p-value a 3 = , p-value Portmanteau test 26.14, p-value
Increments of the data
ACF of the increments
PACF of the increments
MA(1)? b 1 = 0.701, p-value Portmanteau test 29.14, p-value
MA(2)? b 1 = , p-value b 2 = , p-value Portmanteau test 24.77, p-value
AR(4)? a 1 = , p-value a 2 = , p-value a 3 = , p-value a 4 = , p-value Portmanteau test 25.5, p-value
AR(5)? a 1 = , p-value a 2 = , p-value a 3 = , p-value a 4 = , p-value a 5 = , p-value Portmanteau test 25.4, p-value
AR(6)? a 1 = , p-value a 2 = , p-value a 3 = , p-value a 4 = , p-value a 5 = , p-value a 6 = , p-value Portmanteau test 14.81, p-value
AR(7)? a 1 = , p-value a 2 = , p-value a 3 = , p-value a 4 = , p-value a 5 = , p-value a 6 = , p-value a 7 = , p-value Portmanteau test 14.83, p-value
Coal Production data
Its ACF
Its PACF. AR(2)?
AR(1)? m = 3.772, p-value a 1 = , p-value Portmanteau test 21.44, p-value
AR(2)? m = 3.802, p-value a 1 = , p-value a 2 = , p-value Portmanteau test 12.03, p-value 0.97
AR(3)? m = 3.809, p-value a 1 = , p-value a 2 = , p-value a 3 = , p-value Portmanteau test 10.61, p-value
Profit Margin data
Its ACF
Its PACF. AR(1)?
AR(1)? m = 4.699, p-value a 1 = 0.876, p-value Portmanteau test 22.34, p-value Still, ρ(4) is out of range
AR(2)? m = 4.716, p-value a 1 = 1.026, p-value a 2 = , p-value Portmanteau test 21.98, p-value
ARMA(1,1)? m = 4.714, p-value a 1 = , p-value b 1 = , p-value Portmanteau test 19.06, p-value
Parts Availability data
Its increments
ACF for the increments. MA(1)?
PACF for the increments. AR(2)?
AR(1)? a 1 = , p-value Portmanteau test 22.4, p-value ρ(2) is way out of range though
AR(2)? a 1 = , p-value a 2 = , p-value Portmanteau test 14.5, p-value
AR(3)? a 1 = , p-value a 2 = , p-value a 3 = , p-value Portmanteau test 12.24, p-value
MA(1)? b 1 = , p-value Portmanteau test 12.23, p-value
Treasury Bonds Yield data
Its increments
ACF for the increments. MA(1)??
PACF for the increments. AR(1)?
AR(1)? a 1 = , p-value Portmanteau test 29.02, p-value
AR(2)? a 1 = , p-value a 2 = , p-value Portmanteau test 28.96, p-value
MA(1)? b 1 = , p-value Portmanteau test 36.57, p-value
MA(2)? b 1 = , p-value b 2 = , p-value Portmanteau test 30.89, p-value
ARMA(1,1)? a 1 = , p-value b 1 = , p-value Portmanteau test 28.94, p-value
Crops Prices data
Take the Logarithm
Take the difference
Its ACF
MA(3) might work?
MA(3)? b 1 = , p-value b 2 = , p-value b 3 = , p-value Portmanteau test 27.11, p-value
Residual ACF for MA(3) model. Looks good, but the estimate for b(3) has a big p-value 0.157
MA(2)? b 1 = , p-value b 2 = , p-value Portmanteau test 54.11, p-value
Residual ACF for MA(2) model, Portmanteau test has p-value
Switch to AR models. Here is PACF of the data.
AR(8)? Or AR(4)? 8 th value is almost 4 standard deviations
AR(4)? a 1 = , p-value a 2 = , p-value a 3 = , p-value a 4 = , p-value Portmanteau test 39.62, p-value
Residual ACF for AR(4). Portmanteau test is a No (p-value 0.008)
AR(8)? a 1 = , p-value a 2 = , p-value a 3 = , p-value a 4 = , p-value a 5 = , p-value a 6 = , p-value a 7 = , p-value a 8 = , p-value Portmanteau test 21.27, p-value
Residual ACF for AR(8). Portmanteau test is a Yes (p-value 0.215). Finally?
Since 8 is sort of too many, let’s try mixed models
ARMA(2,1)? a 1 = , p-value a 2 = , p-value b 1 = , p-value Portmanteau test 28.44, p-value
Residuals for ARMA(2,1). Portmanteau test is a Yes (p-value 0.161)
Best AIC score = ARMA(8,1) a 1 = , p-value a 2 = , p-value a 3 = , p-value a 4 = , p-value a 5 = , p-value a 6 = , p-value a 7 = , p-value a 8 = , p-value b 1 = , p-value Portmanteau test 12.1, p-value
Residual ACF for ARMA(8,1). Note that the first 5 values are practically zeroes, one of the symptoms of over- parametrization