GLAD: Global–Local Approach; Disentanglement Learning for Financial Market Prediction

Table 4

The maximum drawdown (%) and volatility (%) for four datasets.

Stock	End-to-end learning			Representation learning
	Metric	Transformer [3]	Informer [5]	CoST [25]	GLAD_b	GLAD_c	GLAD_a	GLAD
S&P 500 index	Volatility	1.62	1.617	1.619	1.619	1.615	(1.61)	1.60
	Max drawdown	−28.50	−27.43	−28.26	−26.82	(−24.27)	(−24.27)	−23.07
Nikkei 225	Volatility	1.34	1.334	1.337	1.334	1.321	(1.322)	1.321
	Max drawdown	−19.08	−17.58	−18.57	−17.82	(−15.28)	−15.69	−14.92
Hang Seng HSI	Max drawdown	−28.87	−26.41	−27.26	−26.32	−16.59	(−17.68)	(−17.68)
CSI 300	Volatility	1.33	1.32	1.33	1.328	(1.315)	1.316	1.311
	Max drawdown	−17.14	−16.71	−17.14	−16.63	−15.66	−16.34	−15.66

The best results are highlighted in boldface, while the second-best results are enclosed within brackets.