Empirical Analysis of Automated Stock Trading Using Deep Reinforcement Learning
Round 1
Reviewer 1 Report
This manuscript is working on the idea to design an automatic trading algorithm on financial markets by means to obtain higher return in comparison with the appropriate indices on the markets. The algorithms, which the manuscript applies belongs to the domain of reinforcement learning. It has been compared the outgoing results from 5 algorithms, belonging to the domain of reinforcement learning: Advance Actor Critic (A2C), Deep Deterministic Policy Gradient (DDPG), Proximal Policy Optimization (PPO), a combination of these three algorithms, named Ensemble strategy. An extension of the last algorithm by the authors with four additional algorithms: Actor-Critic Kronecker-Factored Trust Region (ACKTR), Soft Actor Critic (SAC), Twin Delayed DDPG (TD3 and Trust Region Policy Optimization (TRPO) was named Remake Ensemble. These five algorithms were tested. As input data the values of three markets were used and their corresponding indices. Mainly the returns, evaluated by the five tested trading algorithms and the values of the indices were compared. The algorithms were also assessed according to 5 criteria, which gives a vector form of the assessment. The results of the five algorithms are compared towards the behavior of the three indices from the markets: Down Jones, Korean and Japan market indices. The results give reason to the authors to conclude that each algorithm has to be appropriately parametrized towards the particular market.
I find that these results are interesting and probably can help the practician in financial investments. My personal feeling is that the paper uses a large number of notations, which are not easy to remember. This makes difficulties in reading the paper. My general concern is about the opportunity to check and repeat the presented results. I find reasonable the paper to provide illustrations of the calculations, how the different algorithms evaluate the values of the applied five criteria. This can be done as appendix towards the current manuscript. Thus, the reader can use and/or check how the given results are obtained and can apply this methodology to other markets.
I recommend minor revision of this paper.
Author Response
We greatly appreciate your valuable comments. We have revised our manuscript according to your comments.
For our response, please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
I appreciate your effort to contribute to the literature with the study. the content of the study and your methodology are innovative. I believe that your findings will contribute to the literature.
Author Response
We greatly appreciate your valuable comments. We have revised our manuscript according to your comments.
For our response, please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
The theoretical background is weak. Although some relevant references are mentioned in the paper, they did not support the research, which is a problem. One alternative could be to present previous findings in Section 2. In addition, these findings can be useful for comparing performance metrics in Section 6. One theory that can be discussed is the Efficient Market Hypothesis of Fama (1970) as the portfolio performance is compared to the market indices.
There are many papers that explain what is a technical indicator. Please, prefer scientific materials as reference such as [1-3].
Max Dragdown does not exist. Do you mean Max DRAWDOWN? I recommend a detailed review/proofreading after concluding the manuscript.
Sharpe Ratio and other evaluation metrics are not clear. Please, show all formulas. What is the risk-free rate?
Figure 3 does not reflect clearly the index cumulative return. Please, inform that in the caption.
What's the reason for using two data sources? Why not yf only?
Lastly, the contributions and implications should be provided somewhere.
References: *APA style
1. Chong, T. T. L., Ng, W. K., & Liew, V. K. S. (2014). Revisiting the Performance of MACD and RSI Oscillators. Journal of risk and financial management, 7(1), 1-12.
2. Cohen, G., & Qadan, M. (2022). The Complexity of Cryptocurrencies Algorithmic Trading. Mathematics, 10(12), 2037.
3. Mndawe, S. T., Paul, B. S., & Doorsamy, W. (2022). Development of a Stock Price Prediction Framework for Intelligent Media and Technical Analysis. Applied Sciences, 12(2), 719.
Author Response
We greatly appreciate your valuable comments. We have revised our manuscript according to your comments.
For our response, please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
The paper is ready to publish now. Congratulations!
