Carsharing demand estimation: Case study of Zurich area. Date of submission:

Transcription

1 Carsharing demand estimation: Case study of Zurich area Date of submission: 0-- Milos Balac IVT, ETH Zürich, CH-0 Zürich phone: fax: milos.balac@ivt.baug.ethz.ch Francesco Ciari IVT, ETH Zürich, CH-0 Zürich phone: +-- fax: ciari@ivt.baug.ethz.ch Kay W. Axhausen IVT, ETH Zürich, CH-0 Zürich phone: +-- fax: axhausen@ivt.baug.ethz.ch Words: words + figures + tables = word equivalents

2 Balac, M., Ciari, F. and Axhausen, K.W. 0 ABSTRACT Worldwide carsharing has grown significantly in recent years. The traditional round-trip model is no longer the only carsharing model offered. It is now being accompanied by more flexible options like one-way station based, free-floating and peer-to-peer carsharing. Moreover, it has become important to have tools that can estimate both the spatial and temporal demand for carsharing services, providing operators with a good instrument for planning their services. The work presented in this paper makes use of the multi-agent simulation tool (MATSim) to investigate the effects of supply on the demand of the existing round-trip service in the Zurich area. Additionally, the results provide guidance for the possible optimization of the carsharing service. We also present an implementation of a one-way station based model as a part of the MATSim framework and investigate the potentials of one-way carsharing service in the study area. Results show that there is still untapped potential of round-trip carsharing, but that service might need optimization. Furthermore, one-way carsharing, being more convenient, would generate a little less than three times more trips compared to the round-trip option.

3 Balac, M., Ciari, F. and Axhausen, K.W INTRODUCTION Carsharing first appeared as an alternative in Zurich, Switzerland in in a small housing cooperative (). After ups and downs, with the development of the technology and growing awareness of the public it has risen to be a transportation mode worldwide. Carsharing programs today have nearly million members on five continents (). In recent years more flexible carsharing alternatives are accompanying traditional, round-trip carsharing, which require the car to be returned to the station where it was picked up. These include:. One-way carsharing - where a vehicle can be returned at any station. Free-floating carsharing - where a car can be picked and parked on (usually) any public parking spot within the service area. Peer-to-peer carsharing - where privately owned vehicles are available for use by members. These new carsharing alternatives have been growing in the recent years because of their flexibility, but are also facing new problems that need to be addressed and that are limiting the operators. Among these problems are the possibility for the system to become unbalanced during the day and also during a bigger period of time, unavailability of vehicles for the return part of the tour, legal issues with peer-to-peer carsharing etc.. In Switzerland, at the moment, there is only one carsharing operator (Mobility ()) providing round-trip carsharing in the whole country, free-floating in Basel () and is partly involved in the peer-to-peer carsharing (). Having, stations, with,0 cars, it serves nearly,000 members. The paper focuses on the effects of the supply on the demand for the existing round-trip carsharing service and also gives first insights of the effects of replacing round-trip with one-way service, based on an implementation of the multi-agent transport simulation tool (MATSim ()). The results are presented and discussed for the study area of Zurich, varying fleet sizes and the number of members. Even though the share of carsharing trips is marginal - estimated to be well below % in most places with existing carsharing service (see, e.g., (, )), the modal share and number of members is constantly increasing (). Therefore, it is very important to investigate carsharing with a model where supply actually affects the system and the demand. The MATSim framework is a suitable tool for this kind of analysis, as the proposed modeling framework described in the reminder of this paper can capture the relevant interactions. The results obtained on the potential demand and effects of the supply of the carsharing alternatives can be used by policy makers and carsharing operators. RELATED LITERATURE In the recent years, carsharing has attracted attention of many scientists around the globe. The main reasons are the potentially large impacts of carsharing on the environment, transportation, its social effects etc. Some of these impacts include: potential of carsharing to reduce the number of privately owned vehicles (up to vehicles are replaced with one carsharing car (, )), sold vehicles due to carsharing membership (up to % ()), energy consumption (), emission of pollutants (up to % reduction ()) and vehicle kilometers traveled ((0)). In short, one of the main advantages of carsharing is encouraging a more sustainable form of transport. Moreover, the constant growth in the number of operators, members and types of carsharing offered () and the growing presence of the "sharing economy" which is based on "sharing" rather than on "owning" (), has opened many questions and issues that research is trying to tackle.

4 Balac, M., Ciari, F. and Axhausen, K.W From the operational research point of view, the largest attention was given to the possibility of one-way and free-floating carsharing fleets to become unbalanced. Various studies have dealt with this problem in the past ( ) and a good categorization of various relocation strategies can be found in (). However, all these strategies do not take into account that the demand would change with the change of the location of vehicles, which is very important to completely grasp the quality of the service. On the other hand, the literature on the supply optimization and supply-demand relationships is very scarce. An overview of the available literature on the carsharing demand estimation can be found in (). They report that in the current literature, demand estimation has only been addressed for round-trip carsharing and that it is mostly context specific. Moreover, they report that literature on this topic for one-way and free-floating systems is non-existent. Recently, () has addressed this issue for free-floating carsharing in the Berlin area and it showed a large untapped potential of both round-trip and free-floating carsharing services in the simulated area. However, results are based on only one forecast for fleet size and number of members and does not investigate the relationship between supply and demand further. To the best knowledge of the authors only a small number of studies exists dealing with the optimization of the supply side of a car sharing service ((, )). They, however deal only with one-way systems and try to satisfy the current demand with an optimal fleet size, thus not considering how the demand will adapt to the change of the supply. According to () and (0) vehicle availability is one of the most important factors in choosing to become a member of the carsharing program. Moreover, () dealing with one-way carsharing in Montreal area, shows that the sizes of the carsharing stations have a large impact on both availability and usage of vehicles in the area. Building on these results, the purpose of this paper is to fill in the gap in the literature on the demand-supply relationship for round-trip carsharing services, providing a general model that can be used for any situation and area, and to give first insights on the potentials of one-way service, using Zurich as the study area. The model presented is a useful tool for policy makers and carsharing operators to estimate the potential demand of new or changed services. METHODOLOGY MATSim ( was used before to model the supply and demand for round-trip and one-way carsharing (, ). MATSim simulates a synthetic population in a virtual world. The synthetic population is generated using the census data and daily plan (activity chains and mobility tools) for each member of the population is derived from suitable diaries. The virtual world presents the road network - in this study it is a detailed navigation road network. During the iterative process of finding the stochastic user equilibrium each agent can adapt its plan according to its preferences (change transportation mode, departure time, route, location of his secondary activities (shopping and leisure)). The advantage of using an agent-based simulation tool over the traditional four-step models is the opportunity to answer complex scientific questions regarding carsharing user behavior. This comes from the fact that to correctly model carsharing both spatial and temporal location of vehicles is needed which aggregate four-step models can not provide. Both carsharing modes, used in this study, were integrated into the MATSim environment. Implementation of these carsharing alternatives builds on the previous implementations (, ) and provides a significant increase in detail, as will be explained below.

5 Balac, M., Ciari, F. and Axhausen, K.W Simulation Model Transportation alternatives used in the simulation, besides carsharing, are walk, bike, public transport (pt) and car. Walk and pt as modes can be used on a trip level (between any two consecutive activities), while car and bike can be used only on a subtour level (the start and the end location of a subtour need to be the same and a first trip with these modes need to start at home). In the case of round-trip carsharing, individuals are allowed to use it as a mode on the subtour level, meaning that they can pick-up a carsharing vehicle after finishing an activity at a given location and return it after coming back to the same location. The following steps are modeled and simulated:. Agent finishes its activity, finds the closest available car and reserves it (making it unavailable for other agents),. Walks to the station where it has reserved the vehicle,. Drives the car (interaction with other vehicles is modeled),. Parks the car at the next activity,. After finishing his activity the agent takes the car and drives to the next activity,... Before reaching the last activity in the subtour, agent ends the rental and leaves the vehicle at the starting station making it available to other agents,. Walks to the activity,. Carries out the rest of the daily plan. In the case of one-way carsharing (used on a trip level), the steps are similar, but with a few differences:. Agent finishes its activity, finds the closest station with an available car and reserves the vehicle (making it unavailable for other agents),. Walks to the station where it has reserved the car (takes the car and frees a parking spot at the station),. Finds the closest station to his destination, with a free parking spot and reserves it (making it unavailable for others). Drives the car to the reserved parking spot (interacting with other vehicles on the network),. Parks the car on the reserved parking spot and ends the rental,. Walks to the next activity. Carries out the rest of the daily plan. This new implementation of round-trip along with the newly introduced one-way carsharing model addresses previous implementation s limitations by introducing: (a) station capacities including parking spaces for one-way service (b) reservation system (c) physical simulation of carsharing vehicles. Behavioral Model The behavior of agents is evaluated based on the utility function that evaluates each component of the agents daily plan (Equation ()) generating a final score: 0 U plan = m (U act,i + U travel,i ) i= ()

6 Balac, M., Ciari, F. and Axhausen, K.W. 0 where m is the number of activities that agent has in his daily plan. In general, performing activities increases the score (positive utility), while traveling decreases it (negative utility). The utility of an activity is defined as: U act,i = U dur,i + U wait,i + U late.ar,i + U early.dp,i + U short.dur,i () U dur,i is the utility of performing the activity, where the opening times of activity locations are taken into account. U wait,i is the disutility for waiting (i.e. for the store to be opened) and U late.ar,i and U early.dp,i represent the disutility for being late and early respectively. U short.dur,i is the penalty for performing the activity too short. The specific components of carsharing (both round-trip and one-way) travel are: Carsharing constant Rental time fee Time cost of walking (access and egress) In-vehicle travel time Distance cost The utility of traveling, using carsharing, between activities i and i therefore looks as follows: U travel,cs = α cs + β cost,cs Cost t RT + β tt,walk (AT + ET) + β tt,cs TT + β cost,cs Cost d Dist () 0 0 β cost,cs Cost t represents the disutility of time cost for rental. Access and egress walk stages are calculated separately using the underlying navigation network to calculate the shortest time paths and represented with β tt,walk (AT + ET). β tt,cs presents the disutility of traveling while the distance cost is captured with β cost,cs Cost d. Constant α cs captures travel attributes not represented by other components. Equation presents the utility of traveling for all other modes: U travel,i = α mode + β tt,mode TT + β cost,mode Cost d Dist () In this model for calculating the utility of traveling, access and egress times are not calculated but are captured in the constant parameter α mode. The other two components represent the disutility of traveling and the distance cost. These functions are used by the agents to determine, in the iterative process of the MATSim simulation, which mobility option suits them best. A detailed description of the carsharing utility function can be found in (). SCENARIOS DESCRIPTION The study area is the Zurich-Greater area, which is created by drawing a 0km radius circle around the Bellevue square in Zurich s city-center. The area has a population of,,. The agent population is based on the 000. census and 00. national travel diary data. The different types of scenarios used to investigate carsharing alternatives were:

7 Balac, M., Ciari, F. and Axhausen, K.W Base Scenario: Only round-trip carsharing is available, with the current Mobility stations ( stations in the simulated area with cars) and membership assigned to agents, based on the logit model estimated previously by () resulting in, members.. Round-trip scenarios (eight): The effects of different supply size on the round-trip carsharing is observed.. One-way scenario: A complete switch from round-trip to one-way service and its effects are observed for the scenario with original fleet size and the number of members. All scenarios were run for 00 iterations, with a 00% population (meaning that all persons living in the simulated area are represented with a corresponding agent in the simulation). This is the first attempt to run 00% population in MATSim for the analysis of carsharing behavior. Using a smaller percentage of the population is usually done to reduce computation times, but it can produce undesired effects, especially for station-based carsharing as most of the stations have only or cars, which makes scaling down very difficult. Of course, this accuracy comes at the price of higher computation time, which on our servers (0 cores running on.ghz and using 0GB RAM) was approximately days, for each scenario. During simulations, each agent was allowed to change his transportation mode, route and departure time. RESULTS In this section simulation results of the current carsharing situation in the Zurich area is compared with the data obtained from the carsharing operator - Mobility. Next, effects of different supply sizes on the demand is presented and discussed. Finally, results of the switch from round-trip to one-way carsharing are shown and analyzed. Base Scenario Data obtained from the Mobility is from the year 00 and is used to calibrate the simulation module and to determine the quality of the results. The data includes all stations and members in Switzerland, but rental data, within that year, is only for the Canton of Zurich. The rental data includes: id of the vehicle used, rental start and end times and when the vehicle was reserved. Actual simulated area, however, is larger than the Canton of Zurich, with more stations, so the booking data from the Mobility was scaled accordingly, to provide an approximation of the demand side in this area (Table ). The scaling was done by multiplying the number of rentals and used cars with the ratio of the number of cars in the simulated area and the Zurich Canton. TABLE Carsharing supply and demand. Number of Mobility - Canton Zurich Mobility - sim. area MATSim - sim. area Stations: Cars: Members:,,, Rentals/work day: Trips: - -, Used cars: 0 Unique users: - -

8 Balac, M., Ciari, F. and Axhausen, K.W. 0 Table also shows the number of rentals obtained during the simulation with the current fee structure (0.0CHF/km and.chf/hour). A little underestimate of the number of rentals was expected since the Mobility data shows that almost 0 % of the rentals was with the vehicle type "transporter" used to move bulky loads. However, current day plans of the agent population do not differentiate between different types of shopping nor does it includes moving home. Figure (a) shows the comparison of the distribution of rental start times gathered from the Mobility data and observed in the MATSim simulation. Both distributions have three distinguishable peaks and are similar, except that MATSim simulation produces much larger peak in the morning and valleys between the peaks are lower than the ones from the real rentals. Differences between these distributions also arise because in reality users adapt their daily schedule (change order of their activities etc.), which is not modeled at the moment in the simulations (except the length of each activity), to match the availability of cars. Figure (b) shows the distribution of rental length for both data sets. Here the distributions clearly show that the most rentals are very short, however simulation results show a slight peak at 0 hour mark, suggesting that some rentals are used for work activities. The reasons for this could be that the public is not aware that some of them might benefit by using carsharing service for work and/or the calibration of the generalized cost of carsharing travel needs to be slightly adjusted to remove this peak. However, since the peak is not substantial, it was left to be addressed in the future work. (a) Distribution of rental start times. (b) Distribution of rental length times FIGURE Comparison of the base simlation against observed demand.

9 Balac, M., Ciari, F. and Axhausen, K.W. 0 0 These analyses show that the MATSim simulation (even though with mentioned limitations), produces results not very different from the actual behavior of the members of the carsharing scheme. Round-trip Scenarios Before increasing the carsharing fleet, we observed the number of rentals when allowing every agent with a driving license to use the carsharing service. There were rentals, which is only slightly more than the number in the base scenario. This shows, that the demand is definitely limited by the current size of the supply. To see the effects of the fleet size on the number of users, we tested additional fleet sizes - with to times larger fleets. Results for all four fleet sizes are summarized in Table. The increase in the number of rentals and trips is smaller than the increase of the supply and is reaching its limits with - times the original supply size. This was expected since not all members have a daily schedule that is suitable for round-trip carsharing, this finding is supported by the Mobility data which shows that only about 0% of users on each day use the service also on the next day. Moreover, it is noticeable that there are increasingly more unused cars during the day with growing supply. Looking at Figure we can see that the distribution of trip distances (between two consecutive activities) for private car and carsharing are different. The distribution for private cars is left skewed with a very long tail with an average of.0 km per trip. However, carsharing trips have smaller shares for the first km than car trips, but larger ones from - km, with no trips longer than 0 km, resulting in an average of. km. This is expected, since the high distance prices for carsharing is limiting that demand. FIGURE Distribution of trips by distance. Analyzing these results one must have in mind that the number of members was held constant and only current membership holders were affected by the increase of the fleet. However, these findings show that there is still space for the improvement of the current carsharing service, for instance by placing the unused cars into the areas of higher demand, and/or by increasing the number of available cars. On the other hand, these improvements might be limited by the

10 Balac, M., Ciari, F. and Axhausen, K.W. constraints of the space available in a city for placing new or expanding the existing stations. Nevertheless, these findings can be used by operators to steer their development and efforts in the right way. TABLE Round-trip carsharing simulation statistics with different fleet sizes for a constant membership. Variable Original x fleet x fleet x fleet x fleet Number of cars:,.,, Number of rentals/work day: 0,,0,0,0 Rentals increase over original[%]: Number of trips:,,0,,0, Trips increase over original[%]: Average trip distance[km]: Number of used cars: 0,,,, Number of unique users:,,,0,0 0 The rental start times distributions are presented in Figure (a) for original, three times larger and five times larger fleet sizes (for the sake of readability other two were omitted here, since they do not differ from the others). Distributions for all three fleet sizes are very similar. Interestingly with the increased supply size the much larger morning peak disappeared and distributions are much more similar to the Mobility data. This can be explained by the fact that having more cars available compensates the previously mentioned non-modeling of users changing their daily schedule (removing one or more activities, changing the order of activities) to match the availability of cars. The larger fleet gives users more freedom, so they can keep their original daily schedule. Observing rental durations for increasing fleet sizes, it can be noticed that shorter rentals are increasing, the peak around hours has disappeared and percentage of 0 hours rentals is smaller (Figure (b)). These all can be accounted to the increased number of cars, availability and more freedom for users, thus allowing for more shorter rentals.

11 Balac, M., Ciari, F. and Axhausen, K.W. 0 (a) Distribution of rental start times. (b) Distribution of rental length times FIGURE Rental start times and durations for round-trip carsharing with different fleet sizes. 0 Figure shows the distribution of trips, done by round-trip carsharing, by purpose. Even though, return to home trips before ending the rental are usually not counted as a purpose, they are included here to have a better comparison later with one-way carsharing where home as purpose is regularly reported. The distributions are similar for all five fleet sizes, supporting the interpretation that with the larger fleet sizes more users with similar activity-chains are able to use carsharing. The only major change is the increase in the shopping trips. This is expected as there are more shorter rentals as the supply increases. Number of trips with a purpose to work might seem high, but taking into account that work activities reached by any mode, on average, last hours and minute and those reached by round-trip carsharing only hours and minutes, this is not surprising.

12 Balac, M., Ciari, F. and Axhausen, K.W. FIGURE Trips by purpose for round-trip carsharing and different fleet sizes. 0 Round-trip Scenarios with an increased number of members As previously mentioned, one of the most important factors of becoming a member of the carsharing service is car availability. Therefore, the membership model mentioned earlier is used to estimate the number of members for the larger fleets. Statistics on the demand for each fleet is presented in Table. While additional cars bring an increase in the number of members, the fact that the cars are placed at the already existing stations is limiting this increase. Nevertheless, the statistics in Table show that round-trip carsharing has still an untapped potential in the Zurich area. Doubling the fleet size of the current stations would more than double the number of rentals. On the other hand, with increasing the size of the fleet there is an increase in the unused cars which suggests that serious optimization is required to have an optimal service. This would also increase the profitability of the service for the operator. TABLE Round-trip carsharing simulation statistics with different fleet sizes and increased number of members. Variable Original x fleet x fleet x fleet x fleet Number of cars:,.,, Number of members:, 0,,,, Number of Rentals/work day: 0,,,,0 Rentals increase over original[%]: - 0 Number of trips:,,,,00, Trips increase over original[%]: - 0 Number of used cars: 0,0,,0, Number of unique users:,,0,, Figures and show that the rental start times and duration and distribution of trips by purpose, for each supply size is very similar to the base scenario. This suggests that new users are behaving and using carsharing in the same way as the ones in the base-scenario. This shows

13 Balac, M., Ciari, F. and Axhausen, K.W. that by increasing the supply size, more people that have similar daily routines to the ones in the base-scenario, can make use of the round-trip carsharing service. (a) Distribution of rental start times. (b) Distribution of rental length times FIGURE Rental start times and durations for round-trip carsharing with different fleet sizes and matching number of members.

14 Balac, M., Ciari, F. and Axhausen, K.W. FIGURE Distribution of trips by purpose for round-trip carsharing with different fleet sizes and matching number of members. 0 One-way Scenario To grasp the potential of one-way carsharing in Zurich a complete switch from round-trip to one-way was performed while keeping the current number of stations and cars, but adding one additional parking space per car at each station. The pricing structure for round-trip was also used for one-way carsharing for better comparability. Moreover, Kaspi et al. () shows that letting users to reserve the parking spot at the start of their rental increases the performance of the system. Therefore, in the simulations an available parking spot, closest to the destination location, is assigned to each user upon the start of the rental. In the future work, the focus will be on the development of a separate membership model for one-way service and also on analyzing the demand based on different fleet sizes. Table shows the results. TABLE One-way carsharing demand for original fleet and membership size compared with round-trip service. Variable Round-trip One-way Number of cars: Number of trips/work day:,, Average trip distance [km]:..0 Number of cars used: 0 Number of trips per used car:.. Number of unique users:, Table shows that there is a large potential for one-way carsharing with almost three times more trips conducted than with round-based carsharing. Moreover, utilization of vehicles was close to two times larger, almost. trips per used vehicle were made with one-way carsharing service. This increase in the number of trips comes from the much bigger flexibility of one-way

15 Balac, M., Ciari, F. and Axhausen, K.W. carsharing, since the person can return the vehicle at any station and not having to pay for the car during his activities. Looking at the Figure, it can be seen that the start times distribution is similar to the one for round-trip carsharing, but there are some rentals after :00 which corresponds to renting after, late night, leisure activities. This is an additional benefit of one-way service, since members can use this option at night when the public-transport is not very frequent. Schmöller and Bogenberger () in their analysis of the free-floating carsharing in Germany, which is a more flexible type of one-way service, observe exactly the same effect. FIGURE Distribution of rental start times for different operating schemes. 0 As expected work and home dominate the distribution by purpose of trips made with one-way carsharing (Figure ). Unlike for round-trip service, average length of work activities after the trip with one-way carsharing vehicle is higher than the average ( hours and 0 minutes), implying that the service is convenient for people having longer work hours. Moreover, having higher percentage of work is expected, since with one-way carsharing, users do not have to keep the rented car during the duration of their activities in otder to have it for the return part of the trip.

16 Balac, M., Ciari, F. and Axhausen, K.W. FIGURE Distribution of trips by purpose by operating scheme CONCLUSION The results indicate that there is still untapped potential for round-trip carsharing. Increasing the fleet size would mobilize more users among the current members. However, this increase saturates when the supply reaches four times the original fleet size. This increase in the supply allows users more shorter rentals than before, because of the increased availability of the cars, but it comes at the price of more vehicles being unused during the day. Therefore, removal of the unused vehicles and possibly increasing their numbers at the stations with high demand is necessary. On the other hand, assuming that the increased supply would also increase the number of members, we observed that a larger number of users emerge, that have daily plans with a sequence of activities where using round-trip carsharing is suitable. It is also observable that the number of unused cars is also increasing which is undesirable for the operator. Therefore, it seems that better assignment of vehicles to the stations and also finding new spots for stations is necessary for the optimization of the service. However, one needs to have in mind that the simulations presented here simulate only one work-day and that it might be that some cars are used during other days or during the weekends when users have different activity-chains. Finally, it looks that increasing the fleet size by the factor of two brings more rentals/car when the number of members is increased accordingly, which is important for the cost recovery of the service provider. Furthermore, we see that by replacing the round-trip with one-way service has a big potential in Zurich area. The results show that one-way may provide a much better option, being much more convenient for users and generating a little less than three times more trips than round-trip service. Additionally, as expected it is more convenient for making work trips than round-trip service. However, this switch from round-trip to one-way service comes at the price of larger requirements for parking spaces and as the literature suggests of re-balancing the system during the day to maintain the desired performance. MATSim as a simulation tool, allows us, though currently with some limitations, to investigate sophisticated policy measures and to give answers that the traditional four-step model cannot. Improvements, presented in this paper, over the previous version of the carsharing implementation and also by using 00% of the population, some of the limitations have been addressed. Having this in mind, all findings presented in this study, can serve as an insight for the operators of the potentials of carsharing service in the Zurich area. Moreover, they can serve

17 Balac, M., Ciari, F. and Axhausen, K.W as the basis for the fleet optimization (removing unused vehicles, increasing the number of cars where the users where not be able to reserve the car, etc.), which will be a part of our future work. ACKNOWLEDGMENTS The authors would like to thank Mobility, for providing the data used in this research. REFERENCES. Harms, S. and B. Truffer () The emergence of a nation-wide carsharing co-operative in Switzerland, Technical Report, EAWAG,University of Twente, AE Enschede, Netherlands.. Shaheen, S. A. and A. P. Cohen (0) Carsharing and personal vehicle services: worldwide market developments and emerging trends, International Journal of Sustainable Transportation, ().. Mobility (0) Sharing Service in Switzerland, webpage, April 0, mobility.ch.. CatchaCar (0) Free-floating Car Sharing Service in Switzerland, webpage, November 0, sharoo (0) Peer-to-peer Car Sharing Service in Switzerland, webpage, November 0, Balmer, M., M. Rieser, K. Meister, D. Charypar, N. Lefebvre and K. Nagel (00) MATSim- T: Architecture and simulation times, in A. L. C. Bazzan and F. Klügl (eds.) Multi-Agent Systems for Traffic and Transportation Engineering,, Information Science Reference, Hershey.. Haefeli, U., D. Matti, C. Schreyer and M. Maibach (00) Evaluation carsharing, Energy and Communications, Bern, Switzerland.. Kortum, K. (0) Driving smart: Carsharing mode splits and trip frequencies, paper presented at the rd Annual Meeting of the Transportation Research Board, Washington, D.C., January 0.. Lane, C. (00) Phillycarshare: First-year social and mobility impacts of carsharing in philadelphia, pennsylvania, Transportation Research Record,,. 0. Cervero, R., A. Golub and B. Nee (00) City carshare: longer-term travel demand and car ownership impacts, Transportation Research Record,, Shaheen, S. A., M. A. Mallery and K. J. Kingsley (0) Personal vehicle sharing services in north america, Research in Transportation Business & Management,,.. Barth, M., M. Todd and L. Xue (00) User-based vehicle relocation techniques for multiplestation shared-use vehicle systems, paper presented at the 0th Annual Meeting of the Transportation Research Board, Washington, D.C., January 00.. Kek, A. G., R. L. Cheu, Q. Meng and C. H. Fung (00) A decision support system for vehicle relocation operations in carsharing systems, Transportation Research Part E, ().

18 Balac, M., Ciari, F. and Axhausen, K.W Weikl, S. and K. Bogenberger (0) Relocation strategies and algorithms for free-floating car sharing systems, Intelligent Transportation Systems Magazine, () 00.. Jorge, D. and G. Correia (0) Carsharing systems demand estimation and defined operations: a literature review, EJTIR, () Ciari, F., B. Bock and M. Balmer (0) Modeling station-based and free-floating carsharing demand: a test case study for Berlin, Germany, paper presented at the rd Annual Meeting of the Transportation Research Board, Washington, D.C., January 0.. Uesugi, K., N. Mukai and T. Watanabe (00) Optimization of vehicle assignment for car sharing system, in B. Apolloni, R. Howlett and L. Jain (eds.) Knowledge-Based Intelligent Information and Engineering Systems, vol. of Lecture Notes in Computer Science, 0, Springer Berlin Heidelberg.. Correia, G. H. d. A. and A. P. Antunes (0) Optimization approach to depot location and trip selection in one-way carsharing systems, Transportation Research Part E, ().. Millard-Ball, A., G. Murray, J. t. Schure, C. Fox and J. Burkhardt (00) TCRP report 0: Car-Sharing: Where and How It Succeeds, TCRP, Transportation Research Board, Washington, D.C., Le Vine, S. (0) Strategies for personal mobility: A study of consumer acceptance of subscription drive-it-yourself car services, Ph.D. Thesis, Imperial College London, London.. De Lorimier, A. and A. M. El-Geneidy (0) Understanding the factors affecting vehicle usage and availability in carsharing networks: A case study of communauto carsharing system from montréal, Canada, International Journal of Sustainable Transportation, ().. Ciari, F. (0) Sharing as a key to rethink urban mobility: Investigating and modelling innovative transport systems, Ph.D. Thesis, ETH Zurich, Zurich.. Ciari, F., N. Schuessler and K. W. Axhausen (0) Estimation of carsharing demand using an activity-based microsimulation approach: Model discussion and some results, International Journal of Sustainable Transportation, () 0.. Ciari, F. and C. Weis (0) Carsharing membership in Switzerland: modeling the influence of socio-demographics and accessibility, Arbeitsberichte Verkehrs- und Raumplanung,, IVT, ETH Zurich, Zurich.. Kaspi, M., T. Raviv and M. Tzur (0) Parking reservation policies in one-way vehicle sharing systems, Transportation Research Part B, () 0.. Schmöller, S. and K. Bogenberger (0) Analyzing external factors on the spatial and temporal demand of car sharing systems, Procedia-Social and Behavioral Sciences,,.