Skip to main content
Journal for Labour Market Research

Table 2 Out-of-sample forecast accuracy—MAPE and DM test

From: Forecasting the unemployment rate using the degree of agreement in consumer unemployment expectations

 

MAPE

DM

ARIMA

ARIMAX_1

ARIMAX_2

ARIMA vs. ARIMAX_1

ARIMA vs. ARIMAX_2

Austria

 2017

0.66

4.94

5.01

− 4.16

− 4.45

 2012

3.13

3.09

3.09

1.84

1.81

Germany

 2017

9.08

2.69

8.30

6.49

3.84

 2012

6.23

2.37

9.97

3.46

− 4.08

France

 2017

6.22

5.49

5.03

11.24

16.61

 2012

1.78

1.10

1.68

3.68

3.97

Italy

 2017

6.28

1.49

6.43

3.79

− 1.80

 2012

1.07

1.55

1.22

− 2.01

− 1.29

Greece

 2017

6.36

1.58

4.93

4.23

5.30

 2012

6.72

8.47

5.99

− 2.76

4.28

Portugal

 2017

8.58

3.94

3.93

6.18

5.06

 2012

5.45

4.03

2.32

5.50

4.28

Netherlands

 2017

7.13

6.15

6.69

6.15

5.67

 2012

1.85

3.44

2.68

− 3.52

− 3.55

United Kingdom

 2017

2.13

1.79

7.19

0.74

− 5.15

 2012

7.45

6.74

7.12

3.44

3.46

  1. MAPE stands for the mean percentage absolute error. The ARIMAX models incorporate the proxies of unemployment based on the measures of agreement (ARIMAX_1) and disagreement (ARIMAX_2). Diebold–Mariano test statistic with Newey–West estimator. Null hypothesis: the difference between the two competing series is non-significant. A negative sign of the statistic implies that the second model has bigger forecasting errors. Critical value at the 5% level: 2.028
Back to article page