Deep Learning Pricing of Processing Firms in Agricultural Markets

Round 1

Reviewer 1 Report

The title of the article raises the research problem of "deep lerning" in relation to intelligent pricing on the contemporary food processing market. Certainly, the proposed wording and wording should be considered concise and correct. It is also interesting from a scientific (economic) point of view and market relevant. The Abstract section (6-18) has been assessed against the IMRaD standard. It indicates the importance of the purchasing policy of agricultural processing companies on the markets of agricultural production inputs and identifies a research gap in the form of the limitations of ABM agent models. However, the Method used was not indicated, and limiting the Results only to the issue of transport costs (16-18) may turn out to be insufficient (especially since in such a case this aspect should be highlighted in the title itself). It is absolutely unacceptable not to include Keywords (0-0) in a scientific article.

In part "1. "Introduction" (19-105), in the first paragraph there was an introduction to the macroeconomic approach to the agricultural market, but the last sentence also included the purpose of the article. This is a significant structural error. This part always has a separate two-part system. Additionally, the author indicates that the goal will be “/…/is to discover computational solutions by means of artificial intelligent methods/…/” (30-31), to specify “/…/to improve the explanation/…/” (31), which does not correspond to the purpose declared in Abstract “/…/to simulate/…/” (13). So what is the actual purpose of the article: explanation or simulation? But the closest to the truth is the goal contained in section "6.Conclusion and further discussion" (462-504). Additionally, the topic of "game theory" was introduced in the phrase "/.../interaction of firms takes places in one stage games/..../" (47), and later constantly present in the content of the article. And yet, as the author himself writes, “/…/In this paper we opt for using reinforcement learning methods, which are based on deep neural networks (DNNs)/…./” (68). The declared research assumptions are based on the distinction “/…/we constitute two types of agents: agents who we name supervised agents throughout and agents who we name unsupervised agents/…/” (78-79) to bring the considerations back to “/…/ the targeted games Nash equilibria/…/” (95). Therefore, the author points to the following sequence of considerations, i.e.: "price theory", "multi-agent techniques", "characteristics of market participants", "supervised and unsupervised agents" and "flexible and non-elastic market environments". With such a declaration, summation becomes crucial!

In part "2. Background” (106-198), another doubt appears, i.e. the author points to “means of agricultural production”, i.e. fertilizers, plant protection products, industrial feed or seed. It is a pity that he uses the term "food processor agents" because food processing covers a very wide area of activity. There are companies involved in the processing of both animal and plant products. Products can be low-processed or the result of advanced technological processes. At this stage, perhaps only further reading of the article can resolve the doubt. The distinction between learning methods “/…/supervised, unsupervised and reward based learning/…/” (145-146) is a good starting point for further considerations, although not fully exploited in the research process.

In part "3. Market environment" (199-232) presented the adopted methodology for researching this part of the market, and not, as it might suggest, its description with the possible identification of key variables. The title is therefore inadequate to its content.

In part "4. Learning model” (233-374) describes the distinguished types of agents, i.e. unsupervised and supervised, as part of a sequential game for selecting the most profitable pricing policy. This is the correct part that has great cognitive and discussion potential. Game theory is therefore an accepted methodology of research optics.

In the section "5.Simulation results" (375-461), although "the processor agents" (378) is mentioned again, it is without definition. However, the presented (simulation) results are valuable (Figure 1. and 2.) because setting a two-variant transport cost rate, regardless of the type of market, causes oscillation around monopsonistic optimal discriminatory prices.

Based on these achievements, it is much easier to move on to part "6. Conclusion and further discussion” (462-504), especially since the author finally indicates the precise goal he set for the article, i.e. “/…/We designed experiment runs to examine if the deep learner agents converge to Nash or close to Nash policies/ …/” (470-471). However, this part does not sufficiently expose the actual research results and therefore does not provide significant discussion potential. To sum up, you should first autocorrect the Abstract and add Keywords, which would facilitate the positioning of the adopted research intentions. Secondly, the purpose of the article should be clearly and consistently specified and the title of part 3 should be changed, as it concerns the adopted research assumptions. At the same time, difficulties in the economic reception of the article result from the fact that the "deep lerning" formula allows for dynamic pricing strategies, but based on a number of variable factors, e.g. demand, supply, competition prices, availability, seasons, etc. Therefore, intelligent price creators are created, but the algorithm itself must be better than a human (a qualified trader). It is also a pity that the rules for creating prices are not taken into account, e.g. in the case of rigid rules - Rule Based, in the case of conditional rules - capacity, market situation, weather, competitive system, etc. And in customer segmentation - e.g. RFM (Recency, Frequency, Monetary). However, the article after corrections has significant potential for conducting scientific discourse.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The presentation is very interesting. The quality of the study is very good.

I believe that the interest for this topic is high, therefore this study comes in handy. 

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf