SEPNET is a tool to estimate the formation of links in a production network. The program is a simulation algorithm of the well known exponential random graph model ERGM. The code is based on the fixed-density Markov Chain Monte Carlo sampling and the stochastic approximation of the MLE. Details were discussed in Snijders et al. (2006) and Morris et al. (2008).
SEPNET implements ERGM for production network:
- SEPNET considers the gravity model assumption: Firms are connected based on their geographic distance and their respective performance (Sales, Profit, Productivity, etc.).
- SEPNET adds attributes related to the economic activity of firms (Industrial sector, Major bank, etc.)
- SEPNET is a flexible tool to add new economic statistics (Ownership relations, Bankruptcy, Financial Risk, etc.).
- SEPNET includes the most important network statistics (K- Triangles, K- stars, K- Two paths, etc.).
-
Download files: .cpp and .DAT for data.
-
The program uses the C++ Eigen library.
-
Eigen should be installed in your framework, e.g. directory C:/Eigen.
-
To compile from your shell or cmd command: g++ -I C:/Eigen ERGMGithub.cpp -o ERGM.
-
Then, simply execute as follows: ERGM.exe
Parameters in SEPNET considers mainly the chain size of the MCMC. Following, we describe each parameter with its suggested value.
-
number_variables: It counts the number of attributes; 16 in our case.
-
theta: The vector of estimated parameters, with size equal to number_variables.
-
IFD_size: the size the fixed-density MCMC sampling. It depends on the network size, see Lusher et al. (2013).
-
a_r is the gain factor that controls the largeness of the updating steps of parameter theta. Lusher et al. (2013) suggests a value of 0.1.
-
initial_simulation: the size of iteration to calculate the initial theta. This parameter is suggested to be equal to 3 + 7*number_variables.
-
burnin: It defines the size of the burn-in stage.
-
number_phases: The maximum number of iterations before convergence.
SEPNET offers two main outputs, as shown in the example directory "Example_Outputs".
- Theta_Collection.txt: A file that contains the final estimation of the ERGM parameters. Each column is related to an attribute of the model.
- network_sim.txt: A file that contains a simulated network with the FD_MCMC based on the estimated parameters. The user can simulate multiple networks in order to achieve the Godness of Fit analysis of the model.
Krichene, H., Chakraborty, A., Fujiwara, Y., Inoue, H. and Terai, M. Tie-formation process within the communities of the Japanese production network: application of an exponential random graph model. Applied Network Science 4(5), (2019). https://doi.org/10.1007/s41109-019-0112-9
Krichene, H., Arata, Y., Chakraborty, A., Fujiwara, Y. and Inoue, H. How Firms Choose their Partners in the Japanese Supplier-Customer Network? An application of the exponential random graph model (2018). RIETI DP 18-E-011. https://www.rieti.go.jp/jp/publications/dp/18e011.pdf