Multiscale Causal Structure Learning
Authors: Gabriele D'Acunto, Paolo Di Lorenzo, Sergio Barbarossa
TMLR 2023 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To solve the resulting (non-convex) formulation, we propose an algorithm termed MS-CASTLE, which exhibits consistent performance across different noise distributions and wavelet choices. We also propose a single-scale version of our algorithm, SS-CASTLE, which outperforms existing methods in computational efficiency, performance, and robustness on synthetic data. Finally, we apply the proposed approach to learn the multiscale causal structure of the risk of 15 global equity markets, during covid-19 pandemic, illustrating the importance of multiscale analysis to reveal useful interactions at different time resolutions. |
| Researcher Affiliation | Academia | Gabriele D Acunto EMAIL DIAG, Sapienza University of Rome Centai Institute, Turin, Italy Paolo Di Lorenzo EMAIL DIET, Sapienza University of Rome Sergio Barbarossa EMAIL DIET, Sapienza University of Rome |
| Pseudocode | Yes | Algorithm 1 MS-CASTLE 1: procedure MS-CASTLE(Y, L, λ, ρ, γ, r, t, γmax, maxiter) |
| Open Source Code | No | The paper does not provide an explicit statement about releasing its own source code, nor does it provide a direct link to a code repository for the MS-CASTLE or SS-CASTLE methodology. It only mentions packages used for baseline methods. |
| Open Datasets | Yes | We consider data concerning 15 global equity markets at daily frequency. To focus on covid-19 pandemic period, we restrict our attention to observations from January 2020, the 2nd to April 2021, the 30th... The data has been downloaded from Stooq2. 2The website is reachable at https://stooq.pl/. |
| Dataset Splits | No | The paper describes the generation of synthetic data with varying parameters (e.g., T={100, 500, 1000}, N={10, 30, 50}), and the use of real-world financial data from a specific period. However, it does not specify any training/test/validation splits for these datasets required to reproduce experiments. |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., GPU models, CPU types, memory amounts) used to run the experiments. |
| Software Dependencies | No | The paper mentions using 'lingam and Tigramite Python packages' for baseline comparisons but does not specify the version numbers for these or any other software dependencies used for their own MS-CASTLE/SS-CASTLE methodology. |
| Experiment Setup | Yes | Regarding the initialization of W and Z, they must be initialized in order to satisfy the primal feasibility conditions of Problem (6), i.e., to satisfy the two constraints. A possible choice is to set them equal to zero. Regarding the dual variables of the augmented Lagrangian in Equation (8), α and β... they might be initialized to zero as well... The hyper-parameters ρ and γ represent the augmented Lagrangian penalty parameters... Algorithm 1 MS-CASTLE(Y, L, λ, ρ, γ, r, t, γmax, maxiter)... More precisely, we let λ assume values in the set {.001, .005, .01, .05, .1, .5}... The chosen values for λ are given in Appendix D. |