Asset_Pricing_with_Neural_Networks_Significance_Tests

This repository contains code, data, and results of the paper Asset pricing with neural networks: Significance tests, co-authored by Hasan Fallahgoul, Vincentius Franstianto and Xin Lin

Version

Python 3.8.10

Tensorflow 2.8.0

Keras 2.8.0

Numpy 1.24.1

Pandas 1.3.3

matplotlib 3.4.3

Download data

download data from https://drive.google.com/drive/folders/1LRgT72VbG0w3EkTcJXfqY33fIMuSmvKX?usp=drive_link

put Pc50 and Pc100 inside the folder data

Description of folders

1. code includes:

   build_nn_model.py, initializes and compiles a NN model.

   discretization.py, generates tau_h^(max) samples for X data.

   test_statistics.py, calculates gradients and tstats of variables.

   test_variable_significance.py, identifies significant variables, calculates p-values and significance frequenies.

   reproduce_results.py, produces Table 2, Figure 2, and Figure 3 from data uploaded.

2. data includes:

   test statistics and scaled tau_h^(max) samples for the training set.

   scaled tau_h^(max) = tau_h^(max)*U(C',ϵ_n)^2, see Theorem 3.2.3 in paper.

3. output includes:

   .csv and figures produced by reproduce_results.py.