Can Artificial Neural Networks Be Used to Predict Bitcoin Data?

Round 1

Reviewer 1 Report

Manuscript Number: automation-2051445

This manuscript tried to study the prediction of stock return series using computational intelligence and/or artificial intelligence specifically the artificial neural network (ANN) and the results were compared to the random walk. The article is interesting and well prepared; however, it is required minor amendments to be considered for the publication.

Comments for author File: Comments.docx

Author Response

Title of Paper: May Artificial Neural Networks be used to Predict Bitcoin Data?

Manuscript Number: automation-2051445

This manuscript tried to study the prediction of stock return series using computational intelligence and/or artificial intelligence specifically the artificial neural network (ANN) and the results were compared to the random walk. The article is interesting and well prepared; however, it is required minor amendments to be considered for the publication.

General comments:

1. Your references within the manuscript are numbered; however, in the reference list, they are in alphabetical order. Please follow the Journal format. Nevertheless, MDPI is usually using numbering the reference based on where their appear in the text. Therefore, you need numbering from 1 and follow to the last reference.

For instance, Line 19, it should be amended to [1, 2, 3] NOT [1, 4, 27].

2. Line 74, 80, 93, 282, 283, 292,… need to be adjusted to be well located in the paragraph.

Line 74:….

A trading system is a system or a set of rules that with no human interaction, generates trading signals. These trading signals determines when to take a market position and what position to take. The trading rules have to be absolute to make the trading system statistically testable. This makes the trading system testable on historical data. In order to make profit on a trading system, one needs to perform better than most other traders since trading is a zero-sum game [24].

Line 80 ….

A problem with trading system signals that are not carried out automatically are human emotions. Emotions of fear and greed could lead a human trader not to carry out the signaled trades. This can make the trading systems useless.

3. Line 115 to 117, is it a complete quote?

Reply:

Yes

Line 282, 283, …

The first difference means actors are passive between processing messages, in contrast to agents that can be active continuously, by observing an environment. This environment may also include other agents. The second difference means that communication between actors is done only through message passing, and for an actor to send a message to another actor the address is needed. Agent A and B may, for instance, communicate by letting the agent A modify something in the environment that agent B is observing. Agent B can then act by observing certain conditions in the environment. The address of the agent is therefore not needed in this case.

Line 292, ….

The platform consists of four independent parts: data collector, trading system runner, an application programming interface (REST API) and a web application. The reason for dividing the platform into these separate parts is to make it decoupled, so that each part can be exchanged without interacting with the other parts and make each part independently testable. While testing, each part other parts do not need to run, only database entries used by the tested part is needed. These database entries can be added manually to test certain functions and situations. The platform is not tied to a specific financial instrument.

Reply:

Line 115 to 117, is it a complete quote? Yes

5. Please change Table 1 from a picture to an editable Table.

Reply: I think this is ok. I cannot see any reason for to do this. My co-author has answered earlier and has demonstrated the code for this. If the reviewer insists to have changed the Table 1 I think my co-author has to do it. .

6. Your input data for ANN is weights of some indicators. Is it possible to specify the number of datasets?

Reply: Is this correct Asgeir?

7. How did you test the training data? Did you use the Backpropagation?

Reply: Yes Backpropagation is used.

8. Kindly put the complete name for the first appearance of the acronyms in line 393 for ROC, %R, DI and AO.

Reply: ROC = Rate of Change and it is now defined earlier in the paper

9. In line 414-417 while difference signs for the ranges are used? [ ] , ( )?

Reply: [] means a closed interwall and () means an open interwall.

10. Line 456, I think between “80 and F or” we need a full stop.

Reply:

 … long position when the indicator is below 20 and taking a short position when the indicator is above 80. For the Disparity Index a rise above zero is taken as long position, and when it falls below zero, a short position is taken [7]. The Relative Strength Index indicator has a range from 0 to 100 and is normally traded by taking a long position when the indicator is below 30 and taking a short position when the indicator is above 70. When the Aroon Oscillator rises above zero, a long position is taken, and when it falls below zero, a short position is taken. The Moving Average Convergence Divergence indicator consists of two moving averages, one slow and one quick. When the quick one crosses above the slow, a long position is taken and when the quick one crosses below the slow, a long position is taken. 

11. Did you compare the results with “random walk”?

Reply:

Yes we have done that. See the reference list. [3]

Reviewer 2 Report

This paper studies the possibility of artificial neural networks (ANNs) for predicting bitcoin data. The authors performed comprehensive analyses on the systems and discussed how ANNs could be used for related predictions. This is a nicely written paper with the topics very interesting. My only comment is that the authors can give more discussions of other machine learning methods and discuss the possibilities of other machine learning methods as well.

Author Response

This paper studies the possibility of artificial neural networks (ANNs) for predicting bitcoin data. The authors performed comprehensive analyses on the systems and discussed how ANNs could be used for related predictions. This is a nicely written paper with the topics very interesting. My only comment is that the authors can give more discussions of other machine learning methods and discuss the possibilities of other machine learning methods as well.

Reply: This will be done in a continuing paper, to compare the results obtained in this paper by using other machine learning methods.

Academic Editors’ comments:

The paper is interesting, well written, and well structured enough. The aim of the contribution is clear and the application field is original. Nonetheless both reviewers raised some (minor) concerns. In particular, more details on how the ANNs are selected among the other ML tools and some insights on their performance when compared to other possible prediction methods are required. Therefore authors are invited to take into account the received comments.

Reply: it will require another paper to do it.