## Reviewer 3 ### Summary This paper proposes a method combining NCDEs with DFMs to nowcast GDP. According to experiments on real-life datasets, the model performs well on sudden drops and irregular or missing data. ### Strengths and weaknesses The paper is well written, and contributes a novel framework combining NCDEs and DFMs for nowcasting GDP, which enables better nowcasting of irregular time series to solve the real-world problems that irregular times and lagged factors often interfere with GDP nowcasting. And the proposed scheme outperforms the baselines and can be applied to predict various economic indicators to help make better decisions, which has significant practical value in real life. However, there exist several problems, which must be solved before it is considered for publication. - The abstract fails to state the key contents of the paper clearly. The proposed model is a fusion of NCDEs and DFMs. However, in the abstract, the author only focuses on NCDEs and mentions DFMs in one stroke, which may cause readers to misunderstand the content and methods of the paper or ignore one of the key parts. It is recommended that the author should explain more clearly the key features and uses of the DFMs to ensure that readers can obtain a comprehensive understanding when reading the abstract. - The contribution of the article is not obvious enough, and the novelty of the paper seems limited that the fusion method simply stacks two models without an in-depth exploration of the complementarity and synergy between these models. The paper also doesn’t clearly and detailedly explain why combining NCEDs and DFMs helps better deal with lagged and missing data in theory. - The experiment in this paper is not convincing enough. Some of the baselines compared to the proposed model are relatively out of date. And since the authors choose to use MLP as one of the methods to design the CDE function at last, it’s more reasonable and comprehensive to add baselines related to MLP models like DFM-MLP. - The structure of the model is not clearly illustrated. Figures in the paper do not provide intuitive demonstrations of how important parts in the model work. - The paper lacks an explanation of the advantages and disadvantages of the models mentioned in related works. The authors need to highlight the advantages of the proposed model compared to existing models. - When elaborating the conclusions of section 4.6, NCDEs are missing from the ranking. ### Questions for the authors 1.There is a model called DFM-NCDEs in the baselines of your experiment. What’s the difference between it and your proposed model as they both seem to be a fusion of DFM and NCDEs? 2.From your paper, it seems that your model simply combines two technical methods and is not innovative enough. Can you explain the novelty of your proposed framework? **Reproducibility:** Yes ### Overall evaluation Weak Reject: Technically solid paper where reasons to reject, e.g., poor novelty, outweigh reasons to accept, e.g. good quality. ### Confidence Quite confident. I tried to check the important points carefully. It is unlikely, though conceivable, that I missed some aspects that could otherwise have impacted my evaluation. ## Reviewer 5 ### Summary This paper proposes a novel method for GDP nowcasting by combining NCDEs and DFMs. The key idea is to use DFMs for estimating latent factors and NCDEs for capturing temporal dependencies and idiosyncratic disturbances. Experiments on real world GDP datasets validate the effectiveness of the proposed method compared with other baselines. Extensive analysis on the experiment results were carried out thereafter. ### Strengths and weaknesses - Strength (1) The paper leverages DFMs for latent factor estimation and NCDEs to capture temporal dependencies and idiosyncratic disturbances. (2) Experiment results show great improvements on real world datasets. - Novelty: poor Since NCDEs and DFMs have already been successfully applied for time series nowcasting, a simple integration of NCDEs and DFMs may not be novel enough. - Quality: fair The proposed method makes sense: employing two models (NCDEs and DFMs) to capture model different aspects of the time series. - Clarity: poor (1) Writings: overall, the writing is quite poor making it hard to follow for general audience, including but limited to typos/notation confusion, vague motivation, poor explanation for the experiments, etc. (2) Motivation for combining NCDE and DFM is not clear. (3) notations and typos: a. eq. (2) how can the vector $\bm{\Lambda z_t}$ plus a scalar results in a scalar $y_t$? Similar situation applies to eq. (11). b. please use bold character for matrices, e.g., $X$ in eq.(1), $A_i$ in the paragraph after eq.(3), etc. - Significance: poor Although the empirical experiment parts seem convincing, its significance is hampered by its poor novelty and clarity. More analysis and explanation regarding empirical performance are recommended to better justify the proposed method's significance. ### Questions for the authors More explanations are needed for experiment results: (a) What is ‘DFM-NCDEs’ in table 1? What the difference between ‘DFM-NCDEs’ compared with the proposed ‘NCDENow’? (b) Compared with ‘DFM-NCDEs’, what contributes most to the significant performance improvement of the proposed ‘NDCENow’? (c) In table 1, it seems that the integration of DFM and NCDEs (i.e., DFM-NCDEs) performs poorer than the pure DFM and pure NCDEs. For example, ‘DFM’ outperforms ‘DFM-NCDEs’ on Korean GDP in terms of MSE, and ‘NCDEs’ performs better on UK GDP in terms of other metrics. I think contradicts to the assumption of the paper that DFM and NCDEs are complementary in achieving better performance? **Reproducibility:** Yes ### Overall evaluation Reject: For instance, a paper with poor quality, limited impact, poor reproducibility, mostly unaddressed ethical considerations. ### Confidence Somewhat confident, but there's a chance I missed some aspects. I did not carefully check some of the details, e.g., novelty, proof of a theorem, experimental design, or statistical validity of conclusions. ## Meta-review **Promoted to phase 2:** No