Table 4.

Parameters tested in the ML models and parameters leading to the best model

Model	Hyperparameters used	Hyperparameters of the model with the best performance in the validation sample
RF	Number of trees: 500, 1000, 1500	Bitcoin: 1500 trees, 50.0% of the variables sampled at each split
	Percentage of variables sampled at each split: 50.0%, 33.3%	Litecoin: 1500 trees, 50.0% of the variables sampled at each split
		Ethereum: 500 trees, 50.0% of the variables sampled at each split
RF-binary	Number of trees: 500, 1000, 1500	Bitcoin: 500 trees, 33.3% of the variables sampled at each split
	Percentage of variables sampled at each split: 50.0%, 33.3%	Litecoin: 1000 trees, 33.3% of the variables sampled at each split
		Ethereum: 1500 trees, 50.0% of the variables sampled at each split
SVM	Kernel: radial, linear, polynomial	Bitcoin: polynomial kernel, gamma = 0.20
	Gamma: 0.05. 0.10, 0.20	Litecoin: radial kernel, gamma = 0.05
		Ethereum: radial kernel, gamma = 0.10
SVM-binary	Kernel: radial, linear, polynomial	Bitcoin: radial kernel, gamma = 0.20
	Gamma: 0.05. 0.10, 0.20	Litecoin: polynomial kernel, gamma = 0.10
		Ethereum: radial kernel, gamma = 0.10

This table presents all combinations of hyperparameters used in the experiments. The hyperparameters of the model with the best performance in the validation sample were then used to define the trading strategies in the test sample. For Random Forests (RF), the remaining hyperparameters were kept at the defaults of the randomForest R package. For Support Vector Machines (SVM), the remaining hyperparameters were kept at the defaults of the e1071 R package