Financial Innovation

Table 3 Results of best forecasting with the long short-term memory (LSTM) model

From: A hybrid econometrics and machine learning based modeling of realized volatility of natural gas

Number of layers			Neurons per layer			Number of lags			Window size
Q	Obs	Perc (%)	Q	Obs	Perc (%)	Q	Obs	Perc (%)	L	Obs	Perc (%)
1	66	26.2	20	23	9.1	28	51	20.2	700	109	43.3
2	111	44.0	40	36	14.3	30	138	54.8	1000	42	16.7
3	36	14.3	60	13	5.2	37	63	25.0	1400	23	9.1
4	39	15.5	100	27	10.7				2100	78	31.0
			150	153	60.7

Q is the number of layers, neurons per layer, and lags. L is the length of the windows to train. Obs is the days when the LSTM was the best model with this configuration

Back to article page