Dynamic Pricing on Round-Trip Carsharing Services: Travel Behavior and Equity Impact Analysis through an Agent-Based Simulation
Round 1
Reviewer 1 Report
The paper proposed a methodological approach in investigating the impact of dynamic pricing programme on the demand and supply performance of carsharing services. The research described in the paper is innovative and useful. There are several questions and comments need to be addressed before being able to be published.
Page 1, Line 34, there could be more information about the benefits of carsharing services from a wider perspective, for example, road efficiency, air pollution and emission, and social inclusion.
Page 2. Line 68, the more in-depth review about the existing research about using MATSim to simulate pricing strategies should be provided here. What has previous research found? What are the limitations of the existing research about the simulation of pricing strategies? How did these limitations shape the current study?
Page 2. Line 70, this paragraph the authors mainly focused on how dynamic pricing applied in the airline industry and discussed how it could be applied car sharing field. There are a number of existing studies about the application of dynamic pricing in carsharing, electric vehicle sharing, parking, automotive industry, which are more relevant to this current study but has not been reviewed at all here. Such a review is important and should be added.
Page 2 to 3. Section 1 introduction should be improved and reorganised. It is currently a mix of background information, some literature review, and some methodological information (i.e. page 3 Line 98 onwards). To make it clear, it would useful to add a more detailed review of existing literature, add one paragraph discussing the ‘research gap’ identified from reviewing the existing research and move the methodological information to section 2.
Page 3 Line 96, there should be more information about the explorative analysis.
Page 4. Line 184, how the study area was determined and why?
Page 5. Line 195, there should be more details about the specific recruitments for the individual to become members to use the services.
Page 12 Discussion, how could the simulation of dynamic pricing of car-sharing adapt to the trend of upcoming connected and autonomous vehicles?
Page 18, Conclusion, there could be more information about the limitation of the study. What about the implications of the current study on key stakeholders of the carsharing industry, i.e. operators, policymakers, end-users?
Author Response
Please see the attachment.Author Response File: 
Author Response.docx
Reviewer 2 Report
The authors propose an interesting question about how pricing policies affect carsharing markets. They decide to build a sophisticated model and try to formalize many details from the supply and demand of these services. The risk of this type of model is that it is often too complex to understand the logical implications that explain the results. In the current manuscript, I have a lot of problems understanding the model completely. I summarize my main concerns in the next three points.
First, the core of the model is the agent-based transport simulation tool MATSim. MATSim implements ad hoc assumptions about traffic dynamics, modes of transportation, and people mobility, among others. I think these assumptions should be explained if you want to understand how the model runs. I know that MATSim is a complex tool, but a suitable summary of its mechanisms is needed. In particular, the parameterization, the initial state, the schedule of actions agents do at each tick, and how the steady-state is reached. These data can be added as supplementary information of the paper and published in a web repository, for example, the MATSim config.xml, network.xml, and population.xml files. How can I replicate the results if I don't know the model and its parameterization precisely? Reproducibility is mandatory in science.
Second, it is very important to justify other assumptions added to the MATSim by the authors. For example, the four VOT classes selected, the geographical distribution, i.e. coaxial and radial, the pricing functions, i.e. ABDP (line 302) and TBDP (Figure 5), the supply of 75 vehicles distributed over 17 stations (line 329), the utility function (equation 3), the carsharing scoring function (equations 3). Moreover, the last two equations have several variables and parameters that are not explained. On the other hand, there is no sensitivity analysis of these and other parameters, which weakens any conclusion.
Third, there is a validation problem. After spending a lot of hours parameterizing the model and running a simulation, how do the authors validate the results? Do the authors have carsharing data of Berlin that can be used to validate results? Or, can the results and the logic of the model assumptions be validated by stakeholders? Even assuming a complete understanding of the model, considering the number of particular assumptions, it is very difficult to generalize the results to carsharing markets.
Finally, some sections of the manuscript should be rewritten to avoid self-plagiarism. In particular, the introduction section and part of the methodology section is a republished text of the manuscript published in Procedia Computer Science.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 3 Report
The paper presents an evaluation of dynamic car-sharing pricing using an agent-based approach. The study is very well written, methodology is consistent and conclusion are supported by results.
I have just a few comments which could help the authors to improve the quality of the paper:
1) The use of the word "equity" is not very clear in the evaluation: I think the concept should be better highglighted in the introduction of the paper. Is this "equity" (and consequently the VOT) also related to social consideration? Or it is just a matter of trip purpose? Please clarify it.
2) Among the different agent-based simulators presented in lines 158-171, the authors should also include NetLogo, used in several transport studies, e.g.:
- Giuffrida, N., Le Pira, M., Inturri, G., Ignaccolo, M., Calabrò, G., Cuius, B., D'Angelo, R. & Pluchino, A. (2020). On-Demand Flexible Transit in Fast-Growing Cities: The Case of Dubai. Sustainability, 12(11), 4455; and other studies by Inturri et al. on demand responsive transport systems;
- Gatta, V., Marcucci, E., Le Pira, M., Inturri, G., Ignaccolo, M., & Pluchino, A. (2020). E-groceries and urban freight: Investigating purchasing habits, peer influence and behaviour change via a discrete choice/agent-based modelling approach. Transportation Research Procedia, 46, 133-140; and other studies conducted by same authors on freight deliveries and public participation in transport planning.
3) Several parameters are presented in section 2.2 for the evaluation of the agents' score (and throughout the paper), but most of the time parameters' notation is not defined. E.g., in Equation 2 the parameter S_dur,q is intended to be linked to the utility of performing an activity, but its notation (dur) does not reflect the final meaning of the term. Please clarify this issue by defining all the notations.
4) other minor comments:
- in line 105: remove brackets from VOT
- in line 523: I guess "pick" was meant to be "peak"
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
I am not a matsim modeler so I am not really qualified to comment on the modeling details. Overall it is an interesting paper and seems soundly developed. Though I admit I've had trouble following some of the logic, and understanding how the setup and assumptions lead to the results. I will put most of that on me as I am not trained to fully understand this approach, but any additional explanation of linkages between inputs/model structure/outputs to help the reader would be useful. Also a table of acronyms would be great, there are so many.
I have just a few overarching questions/comments:
- I cannot find anywhere that you explain your choices for the levels of the value of time. Your highest value is Eur 4 per hour, which seems very low. Is there a reason for this? Maybe at higher values, one just isn't sensitive to the variation in prices? Or are not typical users of car sharing? This should be explained.
- the value of time seems to be applied only to the decision whether or not to use a carshare, not the time frame for using it. That is, you do not consider 30 or 60 minute time shifts in trips to gain the cheaper time. Is this correct? If not please clarify and discuss a bit more. If so, it would be interesting to see what happens if time shifting were included, and how this relates to the availability of cars. Do minor adjustments in price result in too much shifting, relative to cars available?
- Try as I might (in my time available) I cannot understand the fundamental difference in your two types of spatial distribution of VOT. Maybe you need a figure that shows how this looks within one sector of the city, how the two systems compare. Or maybe there just needs to be some more explanation. But any further discussion of how these are different and how you expect them to result in different shifting patterns, ex ante, would be helpful.
- You indicate that in general your results show that for high values of time, there is less response. But figure 9 shows that for TBDP, the highest value of time (4 Eur) drops back to a similar level as lower values of time. Deserves some discussion.
I don't intend any of these to require any changes in analysis, only interpretation/explanation/discussion in a revision.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
The authors have improved the manuscript responding to some of my main concerns. Now, the model is much clearer, parametrization files are at a public repository, and the main assumptions of the model are explained better. They have rewritten the introduction section, reducing the self-plagiarism with the paper published in Procedia Computer Science. The only criticism I maintain is the validation of the model. The authors propose a theoretical model that is not validated by empirical data or experienced stakeholders, so the conclusions are weak and, considering the complexity of the model, difficult to generalize. In short, the paper shows a significant work of modeling and simulation, which results are valid only for the set of theoretical assumptions of the model. The editor will assess if this contribution is enough to publish the paper in this journal.
Reviewer 3 Report
The authors responded porperly to all my comments and suggestions.
