Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1]
Granger Causal Inference in Multivariate Hawkes Processes by Minimum Message Length
Authors: Katerina Hlaváčková-Schindler, Anna Melnykova, Irene Tubikanec
JMLR 2024 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct a numerical study comparing the proposed algorithm to other related classical and state-of-art methods, where we achieve the highest F1 scores in specific sparse graph settings. We illustrate the proposed method also on G7 sovereign bond data and obtain causal connections, which are in agreement with the expert knowledge available in the literature. |
| Researcher Affiliation | Academia | Kateˇrina Hlav aˇckov a-Schindler EMAIL Faculty of Computer Science, University of Vienna Vienna, Austria, and Institute of Computer Science Czech Academy of Sciences, Prague, Czechia Anna Melnykova EMAIL Laboratory of Mathematics University of Avignon, Avignon, France Irene Tubikanec EMAIL Department of Statistics University of Klagenfurt, Klagenfurt, Austria |
| Pseudocode | Yes | Algorithm 1: MMLH: Causal inference in exp-MHPs by MML Input : Dimension p, data x Output: Estimate ˆγ = [ˆγ 1 , . . . , ˆγ p ] Γ := Γ1 . . . Γp 1 for each i {1, . . . , p} do 2 for each γi Γi = {0, 1}p do 3 ˆθi argminθi Θγi log pγi(x|θi) log πγi(θi) ; 4 ˆcγi I(x; ˆθi; γi) (eq. 31); 6 ˆγi argminγi Γi ˆcγi; 8 return ˆγ = [ˆγ 1 , . . . , ˆγ p ] ; |
| Open Source Code | Yes | MMLH is coded in R, using the package Rcpp (Eddelbuettel and Fran cois (2011)). A sample code is available at: https://github.com/IreneTubikanec/MMLH |
| Open Datasets | Yes | We illustrate the proposed method also on G7 sovereign bond data... It is publicly available at: http://qed.econ.queensu.ca/jae/2018-v33.1/demirer-et-al/ |
| Dataset Splits | No | For each country we roll a one-year window over the respective data and register an event if the latest value of the window is among the top 20% of values in the rolling window. The resulting number of events registered for each country is roughly 500. Following Jalaldoust et al. (2022), we assume that this data is observed within a time horizon of T = 400 of the investigated exp-MHP, i.e. an instance of a short time horizon. |
| Hardware Specification | Yes | The code was run on p parallel cores of a HPC architecture located at the University of Klagenfurt (AMD EPYC 7532, 2.4 GHz, 32-core processor). |
| Software Dependencies | No | MMLH is coded in R, using the package Rcpp (Eddelbuettel and Fran cois (2011)). |
| Experiment Setup | Yes | In both settings we set all non-zero αij-parameters to 0.55, and consider µi = 0.5 and βij = 1. Furthermore, we set the parameter of the uniform prior (28) and exponential prior (30) to b = 105 and c = 10 5, respectively, choosing thus very flat and little-informative prior distributions... Moreover, each entry µi of the baseline vector µ is drawn from U([0.5, 1.0]) and all βij are again set to 1. Here, we further set the prior parameters b and c to 4 and 0.3, respectively, reducing the penalty strength on structures γi with a larger number of non-zero entries. |