. 2020 Jul 29;7:101015. doi: 10.1016/j.mex.2020.101015

Table 6.

The p-values obtained from the two-sided Diebold-Mariano test between methods for the five time series discussed in this study.

Method		Data
A	B	1	2	3	4	5
SSA-LRF	SSA-LRF-Fuzzy^*	0.0251^b	0.1452^⁎⁎	0.0023^b	0.0007^b	0.1227^⁎⁎
	SSA-LRF-NN	0.0029^b	0.0026^b	0.0028^b	0.0005^b	0.0569^⁎⁎
	TLSAR	0.0215^⁎⁎	0.7328^⁎⁎	0.1986^⁎⁎	0.5652^⁎⁎	0.2881^⁎⁎
	TBATS	0.0115^⁎⁎	0.0185^b	0.1172^⁎⁎	0.0637^⁎⁎	0.8041^⁎⁎
	DSHW	0.0283^a	0.0052^b	0.0090^b	0.8733^⁎⁎	-
	ARIMA-NN	0.0033^a	0.0005^a	0.0001^a	0.3191^⁎⁎	0.4350^⁎⁎
	TLSNN	0.0085^b	0.0100^b	0.2859^⁎⁎	0.0002^b	0.2566^⁎⁎
SSA-LRF-Fuzzy^*	SSA-LRF-NN	0.0057^⁎⁎	0.0024^b	0.2086^⁎⁎	0.0588^⁎⁎	0.4088^⁎⁎
	TLSAR	0.3251^⁎⁎	0.3580^⁎⁎	0.0076^a	0.0096^a	0.0191^a
	TBATS	0.0715^⁎⁎	0.0271^b	0.1383^⁎⁎	0.1169^⁎⁎	0.0070^a
	DSHW	0.0003^a	0.0061^b	0.6698^⁎⁎	0.2342^⁎⁎	-
	ARIMA-NN	0.0006^a	0.0004^a	0.0000^a	0.0080^a	0.1643^⁎⁎
	TLSNN	0.0408^b	0.0167^a	0.0066^a	0.8940^a	0.0013^a
SSA-LRF-NN	TLSAR	0.0057^a	0.0088^a	0.0087^a	0.0039^a	0.0083^a
	TBATS	0.0057^a	0.0874^⁎⁎	0.0457^a	0.0041^a	0.0011^a
	DSHW	0.0000^a	0.5789^⁎⁎	0.1221^⁎⁎	0.1228^⁎⁎	-
	ARIMA-NN	0.0001^a	0.0003^a	0.0000^a	0.0057^a	0.1732^⁎⁎
	TLSNN	0.0506^⁎⁎	0.0166^a	0.0012^a	0.0910^⁎⁎	0.0004^a
TLSAR	TBATS	0.1546^⁎⁎	0.0329^⁎⁎	0.3108^⁎⁎	0.0723^⁎⁎	0.1770^⁎⁎
	DSHW	0.0000^a	0.0083^b	0.0257^b	0.8620^⁎⁎	-
	ARIMA-NN	0.0003^a	0.0005^a	0.0000^a	0.0422^a	0.7936^⁎⁎
	TLSNN	0.0590^⁎⁎	0.0202^⁎⁎	0.7965^⁎⁎	0.0163^b	0.8003^⁎⁎
TBATS	DSHW	0.0000^a	0.2699^⁎⁎	0.0989^⁎⁎	0.4494^⁎⁎	-
	ARIMA-NN	0.0002^a	0.0004^a	0.0000^a	0.0103^a	0.4640^⁎⁎
	TLSNN	0.3632^⁎⁎	0.9983^⁎⁎	0.1931^⁎⁎	0.1922^⁎⁎	0.0030^a
DSHW	ARIMA-NN	0.0203^a	0.0003^a	0.0000^a	0.4425^⁎⁎	-
DSHW	TLSNN	0.0000^b	0.1894^⁎⁎	0.0093^a	0.1335^⁎⁎	-
ARIMA-NN	TLSNN	0.0002^b	0.0003^b	0.0000^b	0.01862^b	0.7252^⁎⁎

Data 1: hourly Java-Bali load series in 2014.

Data 2: hourly Java-Bali load series in 2015.

Data 3: hourly Java-Bali load series in 2016.

Data 4: half hourly Bawen load series.

Data 5: weekly US ending stocks of the total gasoline.

means method A is more accurate than method B.

means method B is more accurate than method A.

^⁎

represents the chosen model determined based on the smallest values of MAPE and RMSE in the testing data.

^⁎⁎

means method A and method B has no different accuracy.