nonlincausality 1.1.11

Python package for Granger causality test with nonlinear (neural networks) forecasting methods.

nonlincausality

Python package for Granger causality test with nonlinear forecasting methods (neural networks).

This package contains two types of functions.

As a traditional Granger causality test is using linear regression for prediction it may not capture more complex causality relations. The first type of presented functions are using nonlinear forecasting methods (neural networks) for prediction instead of linear regression. For each tested lag this function is creating 2 models. The first one is forecasting the present value of X based on n=current lag past values of X, while the second model is forecasting the same value based on n=current lag past values of X and Y time series. If the prediction error of the second model is statistically significantly smaller than the error of the first model then it means that Y is G-causing X (Y➔X). It is also possible to test conditional causality using those functions. The functions based on neural networks can test the causality on the given test set (which is recomended). The first type of function contains: nonlincausalityMLP(), nonlincausalityLSTM(), nonlincausalityGRU(), nonlincausalityNN() (and additionally nonlincausalityARIMA(), as there is no other Python implementation of ARIMA/ARIMAX models for causality analysis). NOTE: currently MLP, GRU and LSTM functions might not work properly, but by using NN function can replace all those three.

The second type of functions is for measuring the change of causality over time. Those functions are using first type functions to create the forecasting models. They calculate the measure of the causality in a given time window (w1) with a given step (w2). The measure of change of the causality over time is expressed by the equation:

Those functions can operate with multiple time series and test causal relations for each pair of signals. The second type of function contains: nonlincausalitymeasureMLP(), nonlincausalitymeasureLSTM(), nonlincausalitymeasureGRU(), nonlincausalitymeasureNN() and nonlincausalitymeasureARIMA().

Author

Maciej Rosoł [e-mail], [e-mail]
Warsaw University of Technology

Reference

Maciej Rosoł, Marcel Młyńczak, Gerard Cybulski
Granger causality test with nonlinear neural-network-based methods: Python package and simulation study.
Computer Methods and Programs in Biomedicine, Volume 216, 2022
https://doi.org/10.1016/j.cmpb.2022.106669

Example usage

Assume that there are two signals X and Y, which are stored in the variable data, where X is in the first column and Y in the second. The variable data was split into data_train (first 70% of the data) and data_test (last 30% of the data). Then to test the presence of causality Y➔X for the given lag values (defined as a list e.g. [50, 150]) the following functions can be used (all arguments are examples and may vary depending on the data.).

MLP

results = nonlincausalityMLP(x=data_train, maxlag=lags, Dense_layers=2, Dense_neurons=[100, 100], x_test=data_test, run=1, add_Dropout=True, Dropout_rate=0.01, epochs_num=[50, 100], learning_rate=[0.001, 0.0001], batch_size_num=128, verbose=True, plot=True)

GRU

results_GRU = nonlincausalityGRU(x=data_train, maxlag=lags, GRU_layers=2, GRU_neurons=[25, 25], Dense_layers=2, Dense_neurons=[100, 100], x_test=data_test, run=3, add_Dropout=True, Dropout_rate=0.01, epochs_num=[50, 100], learning_rate=[0.001, 0.0001], batch_size_num=128, verbose=False, plot=True)

LSTM

results_LSTM = nonlincausalityLSTM(x=data_train, maxlag=lags, LSTM_layers=2, LSTM_neurons=[25, 25], Dense_layers=2, Dense_neurons=[100, 100], x_test=data_test, run=3, add_Dropout=True, Dropout_rate=0.01, epochs_num=[50, 100], learning_rate=[0.001, 0.0001], batch_size_num=128, verbose=False, plot=True)

NN

results_NN = nonlincausalityNN(x=data_train, maxlag=lags, NN_config=["l", "dr", "g", "dr", "d", "dr"], NN_neurons=[5, 0.1, 5, 0.1, 5, 0.1], x_test=data_test, run=3, epochs_num=[50, 100], learning_rate=[0.001, 0.0001], batch_size_num=128, verbose=False, plot=True)

ARIMA

results_ARIMA = nonlincausalityARIMA(x=data_train, maxlag=lags, x_test=data_train)

Change of causality over time

For a deeper understanding of the dependency between the signals, the change of causality over time might be studied using the above-mentioned functions. The example usage for MLP neural networks:

results = nlc.nonlincausalitymeasureMLP(x=data_train, maxlag=lags, window=100, step=1, Dense_layers=2, Dense_neurons=[100, 100], x_test=data_test, run=5, add_Dropout=True, Dropout_rate=0.01, epochs_num=[50, 100], learning_rate=[0.001, 0.0001], batch_size_num=128, verbose=False, plot=True)

Conditional causality

nonlincausality package also allows to study conditional causality (with signal Z).

results_conditional = nlc.nonlincausalityMLP(x=data_train, maxlag=lags, Dense_layers=2, Dense_neurons=[100, 100], x_test=data_test, run=5, add_Dropout=True, Dropout_rate=0.01, z=z_train, z_test=z_test, epochs_num=[50, 100], learning_rate=[0.001, 0.0001], batch_size_num=128, verbose=True, plot=True)