Risk-Adjusted Performance of Random Forest Models in High-Frequency Trading
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsWe thank the authors for their fascinating work, which addresses an important topic in machine-learning applications for stock price prediction, specifically in the context of high-frequency trading.
The integration of technical indicators with machine learning methods is an ongoing area of research, and this paper contributes to this field by focusing on random forest regression models and risk-adjusted performance metrics. The paper's strength is that it incorporates advanced risk-reward metrics like the Rachev ratio. It challenges the weak form of the Efficient Market Hypothesis under volatile market conditions. Additionally, the emphasis on risk-adjusted performance metrics adds value to practitioners looking to balance risk and reward in financial models.
However, the paper has significant flaws that require major revision to meet the standards of publication:
1. The literature review is insufficient.
2. The aim and hypotheses of the paper are not clearly articulated, leaving readers uncertain about the specific objectives and the novelty of the paper.
3. The methodology mainly emphasizes well-known technical analysis indicators (without reference links) but provides minimal details about the machine learning modelling process. There is no explanation for why only random forest was chosen over other methods, nor are the tools and libraries used for implementation (e.g., Python, R) mentioned.
4. The authors do not present the algorithm for applying the machine learning model to the stated goals. For example, a flowchart or diagram may enhance clarity, and including files with program code (Python, R and data Sets) would significantly improve the paper.
5. The use of "portfolio" (lines 338, 358, 360, and Section 4.1.1) is unclear. Whether the authors optimize the portfolio structure or analyze a single asset is not evident. This ambiguity should be clarified. If the portfolio structure is not optimizing, why mention the word portfolio and not just asset
6. The machine learning technique is applied only to one portfolio, creating concerns of overfitting. Including other assets or portfolios should be done. This makes the results more robust and scientifically credible.
7. The novelty of the paper is not adequately highlighted. The authors should connect their findings with the study's aim and discuss how the paper advances the field beyond previous recent investigations.
8. Sections 4.2 and 4.3 are overly concise and lack actual economic or financial results. The models are presented without comparison to other methods or tools, limiting their practical significance.
9. The references are not correctly formatted, and their content should be reviewed for completeness and relevance, notably to include more recent studies.
Thus, without addressing the following flaws, the article is not suitable for publication:
1) a current literature review (articles after 2021 and recent ML methodology and tools paper);
2) a clearly defined methodology and criteria for selecting specific machine learning models;
3) conducting research on various (not only one) assets and/or portfolios, identifying persistent dependencies and cross-validation.
In conclusion, while the paper has potential, it requires major revisions to address these issues. I hope these recommendations will help you and improve the quality of the research and its presentation.
Author Response
Thank you for the reviewers' valuable comments. Please find attached a PDF document containing our responses to each point and suggestion.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsAssessing the Impact of Technical Indicators on Machine Learning Models for Stock Price Prediction
This manuscript makes an excellent contribution to the special issue. The authors are very clear about the state-of-the-art and their contribution in this paper. I find the paper very accessible and well-written.
Minor comments which can be addressed before sending the final version to the typesetter:
Eq (1). It might be better to denote log returns "log_return(t)" for consistency of notation.
Eq (3). No time/day index. Do you use last day's interest rate?
Indicators. Time index moves between "(t)" and "subscript t". Is there any reason for this?
"This simulates liquidity constraints and transaction costs, making the simulation more
realistic." I am not sure that the 4% turnover constraint captures TC. I am also not quite sure whether your buys (sells) are at the (average) ask (bid) of the current minute ro do you use the mid-price? Needs to be clarified.
Author Response
Thank you for the reviewers' valuable comments. Please find attached a PDF document containing our responses to each point and suggestion.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThe following major comments must be addressed before publishing the paper.
1. Abstract: "highlighting significant overfitting challenges", instead of writing a general statement, be specific while referring to the results achieved in the paper. In addition, start your abstract with a novelty or knowledge gap that motivated you to write this paper. Moreover, the abstract will be concluded by mentioning the implications of the study.
2. The introduction part lacks key contributions of the study. Instead of mentioning contributions in the literature review section, they must be moved to the introduction section.
3. The literature review section lacks hypotheses; perhaps there is a need to re-write the whole literature review. In addition, I believe a critical review of the literature is missing, and simply mentioning previous research is not enough. There are ample studies on the topic, but recent studies analysis is also missing.
4. Linking with the previous comment, references are not enough. Perhaps incorporating the latest studies would strengthen the context and ensure that the research is aligned with current trends and advancements in the field.
5. The discussion section lacks in depth and detail. It would benefit from a more thorough analysis, particularly by aligning the results achieved in this study with similar studies in the field. In addition, write the implications of the study and provide economic sense to the results.
6. As mentioned in the results section, "These results suggest that, despite a strong in-sample performance, the models struggled with generalization, highlighting the potential limitations of using technical indicators in the given market conditions". However, the minute level data taken is from "April 2024 to September 2024" which limits the outcome of the paper. In addition, High-frequency minute-level data can sometimes include outliers or noise, particularly during off-hours or low-liquidity periods; elaborate further on what authors have done in this case. Finally, it is preferred to have additional time-series cross-validation tests to better evaluate the model's robustness.
Author Response
Thank you for the reviewers' valuable comments. Please find attached a PDF document containing our responses to each point and suggestion.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThank you to the authors for the significant changes made. The paper is ready for publication.
Reviewer 3 Report
Comments and Suggestions for AuthorsThe paper can be accepted, and I have no further comments.
