Forecasting and trading cryptocurrencies with machine learning under changing market conditions

Financial Innovation

Table 6 Forecasting ability of the models

Variables	Success rate (classification)	Success rate (regression)	MAE	RMSE	Theil’s U²
Validation sample
Linear (BTC)	49.69	57.72	4.25	5.79	71.49
Linear (ETH)	45.68	48.46	4.97	6.85	92.13
Linear (LTC)	49.38	45.37	5.73	8.14	98.13
RF (BTC)	57.10	56.17	4.30	5.77	93.80
RF (ETH)	50.00	55.86	5.06	6.85	96.26
RF (LTC)	51.23	47.84	5.99	8.34	94.93
SVM (BTC)	49.07	52.16	7.86	19.69	127.13
SVM (ETH)	53.40	53.40	8.26	15.65	59.44
SVM (LTC)	54.32	50.93	11.96	33.28	144.60
Test sample
Linear (BTC)	46.15	51.39	2.24	3.36	68.83
Linear (ETH)	53.85	54.46	3.65	5.20	80.60
Linear (LTC)	50.77	46.77	3.75	5.05	77.52
RF (BTC)	48.92	50.15	2.42	3.46	107.86
RF (ETH)	60.00	49.85	3.79	5.19	96.21
RF (LTC)	50.15	46.46	3.72	4.98	103.70
SVM (BTC)	51.08	50.15	2.98	4.25	625.61
SVM (ETH)	56.92	53.54	3.71	5.28	65.86
SVM (LTC)	55.69	59.69	3.59	4.98	43.87

This table shows some metrics aiming to assess the forecasting performance of all the proposed models: Linear models, Random Forests (RF) and Support Vector Machines (SVM). The Success Rate is the relative number of times that the model gives the right signal on the 1-day ahead return. This indicator is presented not only for the regression models, but also for their binary versions (Classification models). The other columns refer to the Mean Absolute Error (MAE), the Root Mean Square Error (RMSE) and the Theil’s U². This last metric represents the ratio of the Mean Squared Error (MSE) of the proposed model to the MSE of a naïve model which predicts that the next return is equal to the last known return. All values are multiplied by 100