Optimal Estimation of Multivariate ARMA Models
Authors: Martha White, Junfeng Wen, Michael Bowling, Dale Schuurmans
AAAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | An empirical evaluation demonstrates the benefits of globally optimal parameter estimation over local and moment matching approaches. An experimental evaluation demonstrates that globally optimal parameters under the proposed criterion yield superior forecasting performance to alternative estimates, including local minimization for ARMA estimation and moment-based estimation methods for state-space models. |
| Researcher Affiliation | Academia | Martha White, Junfeng Wen, Michael Bowling and Dale Schuurmans Department of Computing Science, University of Alberta, Edmonton AB T6G 2E8, Canada {whitem,junfeng.wen,mbowling,daes}@ualberta.ca |
| Pseudocode | Yes | The overall estimation procedure is outlined in Algorithm 1 for G(B) = ||B||2 F ; the approach is similar for the other regularizer, but with an outer line search over ρ. Algorithm 1 RARMA(p, q) |
| Open Source Code | No | The paper does not provide any explicit statements about the release of its source code or links to a code repository. |
| Open Datasets | Yes | To see how our method performs on real-world data, we experimented on two real-world datasets from IRI.5 These climate datasets consist of monthly sea surface temperature on the tropical Pacific Ocean (CAC) and the tropical Atlantic Ocean (Atlantic). 5http://iridl.ldeo.columbia.edu/SOURCES/ |
| Dataset Splits | Yes | For RARMA, because of the temporal structure in the data, parameters were chosen by performing estimation on the first 90% of the training sequence and evaluating model performance on the last 10% of the training sequence. |
| Hardware Specification | No | The paper discusses computational complexity and runtimes but does not provide specific details about the hardware (e.g., CPU, GPU models) used for experiments. |
| Software Dependencies | No | The paper mentions 'built-in Matlab implementations' for AR and N4SID, but does not provide specific version numbers for Matlab or any other software dependencies. |
| Experiment Setup | Yes | The lag parameters p and q were selected according to standard criteria in time series. For AR and ARMA, the parameters p and q are chosen using BIC and AICc... We use a robust loss, the Huber loss, for RARMA, which is easily incorporated due to the generality of RARMA. |