Skip to main content

Table 10 List of LSGAN parameter

From: Forecasting VaR and ES by using deep quantile regression, GANs-based scenario generation, and heterogeneous market hypothesis

Description

Value

Generator G

 

Conditioning network

 

Size of latent vector Z

2*1

Rolling window b

100

Forward window f

10

1D Conv layers

4

Input and output channels in all Conv layers

2*1

Dense layer output size

1

Simulator network

 

Dense layer output size

f*1

Transpose Conv layers

2

Input channels in the first transpose Conv layer

4*1

Output channels in the first transpose Conv layer

2*1

Input channels in the second transpose Conv layer

2*1

Output channels in the second transpose Conv layer

1

Conditioning and Simulator networks

 

Layers’ activation function

Relu

Conv and transpose conv kernel length

5

Conv and transpose conv stride

2

Discriminator D

 

Convolution layers

4

Input channels for the kth Conv layer

\(2^{k - 1}\)

Output channels for the kth Conv layer

\(2^{k}\)

Layers’ activation function

Leaky_Relu

Conv and transpose conv kernel length

5

Conv and transpose conv stride

2