26 where the distance is measured between the zip codes reported by buyers and sellers. 22 The blue line takes a seller who made an offer at a specific time, and pairs him to every buyer who posted tasks in the preceding 48 hours. The median distance is computed among all such pairs within a city-month. The orange line is just the pairing of tasks and their corresponding offers, and the distance is computed between the zip codes of buyers who posted those tasks and sellers who submitted those offers. The grey line is the pairing of buyers and sellers from successful matches. The figure plots the median distance for the six largest cities over time. 23 None of these measures of buyer-seller distance shrinks as a market scales up. Table 3 reports estimates of the market pricing function, again estimated by ordinary least squares (second column). Price moves very little with the number of tasks and offers. Doubling the number of tasks, holding constant the number of offers, increases the average transacted price by 1.5 percent, while doubling offers decreases it by 1.3 percent. This is perhaps a little surprising from the standpoint of strategic pricing, especially for the auction tasks where buyers choose from competing offers, but it holds true even in a restricted sample of auctions. More details are in the Appendix. The results further confirm that the average price is invariant to market scale: the sum of the price elasticity to tasks (1.5) and to offers (-1.3) is virtually zero. The two results of constant returns to scale in the matching technology and scale invariance of price empirically confirm our earlier assumptions from Section 4. Table 3 does not report the city and time fixed effect estimates. We will return to these estimates in Section 7 where we discuss the differences in platform success across cities. We conclude this section with two observations. First, our identification assumptions provide an over-identifying restriction that we can test because we have assumed that both (S, B) and (σ, β) are independent of the pricing and matching errors (ɛ k, ɛ a ). We can test the latter assumption by running a Hausman specification test where we re-estimate equations 11 and 12 using (S, B) as instruments and compare the set of OLS and IV estimates, testing for their equality. With a χ 2 (65)-distributed test statistic of for the matching function and 0.15 for the pricing function, both tests fail to reject that (σ, β) are independent of (ɛ k, ɛ a ) We compute the geodetic distance, i.e. the length of the shortest curve between the two zip codes, where the input coordinates are assumed to be based on the WGS 1984 datum. The distances are ellipsoidal distances computed using equations from Vincenty (1975). 23 Other cities display similar time trends. 24 We run the original Hausman test and compare the full set of estimates, rather than just the estimates of the elasticities to tasks and offers. 25

27 Our estimation of the matching function leaves very little unexplained. The R-squared of the regression (first column of Table 3) is By construction E(ɛ a n) = 1, and its standard deviation is only About 50 percent of the differences between actual and predicted matches across markets is less than 9. The number of matches formed between tasks and offers is thus very accurately predicted by the Cobb-Douglas matching function. The amount of residual variation in the pricing function is a little higher, given a R-squared of 0.73 (second column of Table 3). However it corresponds to a discrepancy of $3 or less in most markets. 6.2 Gains from Trade We now turn to our estimates of the utility parameters, presented in Table 4. To discuss them, we consider a market with the median number of buyers (B = 447), the median relative number of buyers ( B S = 3.72), and the price and matching parameters from San Francisco in October We consider sellers first. The estimates imply that search costs are relatively low, and, consistent with our earlier evidence, labor supply is highly elastic. Estimates of ˆγ = 0.41 and ˆδ = 0.25 imply that the search costs in the median market are 25 cents for one application, $1 for two, and, per our assumption, continue to increase at twice the rate of offers. The predicted number of offers per seller in that market is 9, corresponding to $21 in individual search costs, and to $4.60 in marginal search costs. We also find that search costs decrease with market size, but only slightly. Holding constant the buyer to seller ratio, an increase in the absolute number of buyers from the 25 th (B = 220) to the 75 th (B = 1, 085) percentile of the distribution of market size increases the number of offers per seller from just above 8 to 10, a 22 percent increase. Search effort is much more responsive to price and the expected match rate than it is to market size (Figure 8). The elasticity of search effort to the offer match rate q s is equal to one by assumption: a doubling in the offer match rate doubles search intensity. Effectively, this implies that the elasticity of tasks supplied is twice as large. Doubling the offer match rate doubles the number of offers submitted, and, because now each offer is twice as likely to be accepted, each seller works fours times as hard. The elasticity of search effort to the price is equal to the inverse of the seller markup. We estimate the seller cost of performing a task to be ĉ = $ In the median market, sellers are paid $48.50, so the supply elasticity to price is equal to Even a $5 price increase 25 In principle, it can happen that the seller opportunity cost c n in a market be smaller than the realized average price paid to sellers. However, c n should always be lower than the expected price reduced by the commission fee 0.8p n in order to rationalize a positive number of submitted offers. That is the case in our estimates. Analogously on the buyer side, p n should always be (and in fact always is) lower than the buyer value for the task v n. 26

28 can raise the number of offers per seller from 9 to 12. Lastly, we find that sellers receive relatively little surplus from completed tasks. The per-task profit is $ = $15.50 in the median market. Given that an offer is expected to match with 30 percent probability and each seller submits 9 offers, the search cost per completed task is $7, or half the task profit. Even if seller search costs are low in absolute dollars, they represent a large share of the per-task profit, and a seller ex-ante expected surplus from the median market is $ Next we turn to the buyer side. 27 We find that the mean arrival rate of tasks is ˆµ = Tasks are cheap to post, with an average cost equal to 1/ˆη = 50 cents, and the buyer value from task completion is ˆv = $70. Given the small cost of posting tasks, in the median market, all needs are posted, so β = In this market, tasks are successfully matched with 61 percent probability, and buyers pay $61 for each completed task. Task demand is inelastic. At equilibrium in the median market, the elasticity of buyers posting rate to the task match rate is basically zero (0.0002), and the elasticity to price is similarly small ( ). Given the low cost of posting, Figure 9 shows that buyers are going to post all their needs for most equilibrium values of match rate and prices. In a standard setting, low elasticity of demand would imply that buyers receive a large share of the surplus. However, a seller willingness to pay for the average task, at $70, is not much higher than the price she actually pays, and a buyer ex-ante expected surplus from the median market is $6. Combining our estimates, we find that the gains from trade for each task are $34, without factoring in seller search costs ($7 per completed task) and buyer posting costs (less than $1). This seems quite plausible given the nature of most tasks, and combined with an elastic supply it means that maintaining a relatively efficient matching process is as crucial for the platform success as attracting a large number of buyers. We discuss the efficiency of the matching process in the next subsection, and buyers growth in Section 7. We conclude this section by examining the fit of our model. Figure 10 compares the actual aggregate number of posted tasks and submitted offers with those predicted by our model for the 26 This is possibly consistent with a young peer-to-peer platform, where the average seller only performs a few tasks to fill in his schedule. Because we focus on market averages, we do not consider the more professional sellers, those who perform tasks on a more regular basis. 27 On the buyer side, the estimation is complicated by the fact that variation in the posting rate is very limited. This implies that the parameter estimates for the mean task arrival rate µ, the cost distribution parameter η, and the value for each task v have to be such that the response of β to changes in the task match rate and price have to be small. Using equation 3, this requires that both β q b = µf ( q b (v p) ) (v p) and β p = µf ( q b (v p) ) ( q b ) be small. The second partial derivative is close to zero only if the density f ( q b (v p) ) is small, given that both the task arrival rate and the task match rate are not small. The buyer net benefit v p (divided by q b ) is equal to the ratio of the two partial derivatives, which are both close to zero. This limits our ability to easily estimate v. 27

29 city of San Francisco. Other cities are in Appendix C. Overall the model does a good job at tracking task and offer activity over time for each city, albeit in some cities supply is consistently underestimated (as in Boston) or overestimated (as in New York). The discrepancies are typically less than 25 percent relative to the actual value, with the only notable exception being the last fews months in New York. 6.3 Benefits of an Elastic Labor Supply When the cost of posting tasks is small, fluctuations in the number of buyers translate in proportional fluctuations in the number of posted tasks, where the coefficient of proportionality is the mean arrival rate of needs. In addition, when buyers have low willingness to pay for tasks, and gains from trade are relatively modest, the range of price adjustment is very limited. Together, these two results imply that the market can clear in only one of two ways: through an elastic labor supply, or through buyers rationing. We found the first to be the dominant equilibrating mechanism on TaskRabbit, and in order to evaluate its benefits, we compare it to the second alternative. We start with a simple exercise to illustrate the intuition. Consider a market with 1,000 posted tasks, and suppose that the number of offers submitted is 1,400. Using our estimates for the matching function, 488 matches would be created out of these two aggregate inputs. Now assume that demand doubles to 2,000 posted tasks. A perfectly elastic supply would lead to a doubling of the number of offers, and would create 930 total matches. 28 Analogously, if demand halved to 500 tasks and supply adjusted downward to 700 offers, the number of matches created would be 256. Regardless of the size of demand, tasks would always match at the same rate. In the alternative scenario, supply is held fixed at 1,400 offers, and equilibration occurs through buyers rationing: when demand is low, it is easier for buyers to find a match, and when demand is high it becomes harder to trade. In the low-demand market (500 tasks), the number of matches created would be 367 and each task would match with a 73 percent probability. In the high-demand market (2,000 tasks), 649 tasks would be matched, implying a 32 percent match rate. Overall, if we compare the total number of matches between the two scenarios, the platform with an elastic labor supply is able to create 11 percent more matches. 29 This is evidently optimal from the platform perspective: since its revenues are a 20% commission on actual matches and 28 Note that doubling the number of tasks and offers does not double the number of matches because of the slightly decreasing returns to scale estimated for the matching function (shown in Table 3). 29 The result comes from the fact that the matching function is concave in both inputs, so that M(b, s )+M(b, s) > M(b, s) + M(b, s ) where b > b and s > s. 28

30 prices barely move, in this simple example having an elastic supply raises its short-term revenue by 11 percent. Since it also raises retention, it benefits the platform by accelerating its growth. Equilibration through seller effort is also optimal from the buyers perspective. Given that the cost of listing a task is low enough that rationing does not prevent them from posting them in the short run, effectively increasing the number of matches by 11 percent raises buyers surplus by the same percentage. An elastic supply creates more matches but it is also more costly, given that sellers costs increase in their search effort at an increasing rate. Again we can use the cost estimates from our model in the simple example above, assuming that offers come from 200 active sellers (so that in the market with 1,000 tasks and 1,400 offers each seller submits 7 offers). The total search costs in the first scenario, where supply fully adjusts to accommodate demand, are 54 percent higher than in the second scenario, where supply is fixed at 7 offers per seller. We now apply this intuition to our context. To do so, we consider all 336 markets and simulate interaction among buyers and sellers under two scenarios: the first considers the labor supply elasticity directly estimated from the model, which implies that the market predominantly equilibrates through seller effort; the second fixes individual supply at 7.58, the average number of offers per seller across our markets, which implies that the market will equilibrate through buyer rationing. We start by measuring the aggregate value of matches created under the two scenarios, which is given by the following formula: 336 n=1 M(B n ˆβ n, S nˆσ n )(ˆv n ĉ n ). A flexible supply allows a 15 percent increase in the value of matches created. The increase in buyer and platform surplus is analogous for two reasons: prices do not adjust much, and buyers posting costs are small. So measuring the platform s aggregate revenue as 336 n=1 0.2ˆp nm(b n ˆβn, S nˆσ n ) results in a 15.5 percent increase in revenue relative to an inelastic supply. Buyers aggregate surplus, or 336 n=1 M(B n ˆβ n, S nˆσ n )(ˆv n ˆp n ) E ( η η ˆq b (ˆv n ˆp n ) ) B n ˆβn, also increases by the same percentage. Sellers aggregate surplus is however reduced by their elasticity, due to the increasing search costs. Relative to providing a constant level of effort, sellers aggregate surplus, defined as 336 n=1 M(B n ˆβ n, S nˆσ n )(0.8ˆp n ĉ n ) 1 2ˆγ(B n ˆβn)ˆδ S nˆσ 2 n, is reduced by 6 percent. If the platform were able to reduce, or even eliminate, seller search costs, the benefits of an elastic supply would be large and positive for sellers as well. We discuss how the platform can achieve this in the conclusions. We now turn to discuss how market efficiency and growth differ by city. 29

31 7 Platform Growth and City Heterogeneity A notable feature of the data is that some cities exhibit striking growth in participation, and others exhibit more moderate growth (Figure 2). In principle, two types of theories can explain differences in how cities attract and retain a large number of users. The first type relies on scale economies and strategic complementarities between the adoption patterns of buyers and sellers. If market frictions were reduced by market scale, we would expect that cities which started off with a large user base grew much faster than cities of modest size, exactly because growth led to more growth. However in Section 6 we estimated only moderate scale economies. Therefore initial differences in adoption cannot explain increasing heterogeneity over time. A second set of hypotheses rely on city differences in facilitating interactions between buyers and sellers. To develop this idea, we show that user attrition is lower in more efficient markets and that markets vary greatly in their matching efficiency, summarized by the fixed effects of equation 11. Combining these results, we see a strong relationship between the rate at which tasks and offers are converted into successful matches and city growth rates. We conclude the section with evidence that relates match efficiency with measures of market thickness at the city level: geographic distance between buyers and sellers, and task specificity. 7.1 City Differences in Growth Platform growth is a combination of adoption and retention of existing users. Given that supply is so elastic that buyers do not have a considerably harder time finding matches when abundant, and given that active buyers seem to post on average the same number of tasks every month, growth depends on buyers participating decisions. Figure 11 plots buyer adoption and retention separately for the 10 largest cities. The left-hand side panel shows the number of new buyers, in log scale, over time. A buyer is defined as new in a city-month if she posts her first task in that city during that particular month. Buyers adopt the platform at a linear rate, different in all cities. At visual inspection, this rate seems to be correlated with the city-specific retention rate. For every city, the right-hand side of Figure 11 plots the share of active users in a month who posted again at least once task in the following three months. San Francisco is successful at both attracting new buyers and retaining current ones, while Philadelphia has both lower adoption and retention rates. 30

32 The literature on innovation diffusion (Young, 2009) has focused on three types of mechanisms leading users to adopt new technologies: network effects, technology improvements, and information diffusion. The first two assume that different users adopt at different points in time because of heterogeneous benefits: early adopters have a large intrinsic value from a new technology, while late adopters join because of scale economies or technical upgrades. We have argued that platform efficiency does not increase with market scale, and for the period under consideration TaskRabbit did not implement major platform changes. Word of mouth and information diffusion, then, seem to be the most plausible alternative in this context, and cities can differ both in the rate at which information spreads and the rate of take-up conditional on receiving that information. For example, in San Francisco adoption might be fast because people there are eager to experiment with new technologies and because current users spread the information at a faster rate, with the second factor possibly driven by a positive experience on the platform. We measure the aggregate effect by estimating the city-specific growth rate: 30 new tc = φ c age tc + ν tc, (13) where new tc is the number of new buyers joining city c in calendar month t, and age tc is the age of the platform in city c at time t. For example, age tc = 1 if month t is the first since TaskRabbit became active in city c. In Appendix D, Table A1 shows the results, and we verify that deviations from the linear adoption rate are not driven by contemporaneous market conditions, in support of our earlier identification assumptions. We compare adoption rates with retention. Retention can be city-specific and, within each city, further depend on current outcomes, match rates and prices: ( ) staytc log = θ 0 X tc + η t + η c + ν tc. (14) 1 stay tc stay tc is the share of users active in city-month t, c who were active again at least once in the following three months within the same city. X tc is two-element vector of relevant outcomes in 30 We assume a linear growth rate different across cities, given Figure 11. It can be rationalized within the Bass model of new product diffusion (Bass, 1969): new tc = φ c + new t 1,c, where new tc is the number of new buyers joining city c in calendar month t. Two things differ from the standard specification. First, the total number of potential adopters is assumed to be large relative to the platform size, which is consistent with the population size of the metropolitan cities relative to the current users on TaskRabbit. Second, we assume that new adopters in the previous month are the only users spreading information, and not adopters of previous months. Each new adopter diffuses information so that exactly one extra adopter joins the platform in the following month. 31

33 city-month t, c: realized buyer match rate and average transacted price. We expect that a high match rate would increase the odds that a buyer will be active again in the next three months, while a high price would drive away more buyers. As with equation 13, Table A3 in Appendix E shows the results, which confirm our hypothesis, and we verify that retention is not driven by expectations on future outcomes, in support of our earlier identification assumption. Figure 14 plots the estimates of φ c (city-specific adoption rate) and η c (retention rate) from equations 13 and 14. A certain correlation exists between the rate at which a city is able to attract buyers and the rate at which it can retain them, although it is by no means perfect. San Francisco and New York are successful on both measures, while Houston, Atlanta and Phoenix lag behind on both. However, in San Diego buyers adopt at a fast rate but are also likely to leave the platform, while in Portland new buyers are just a few but they stay longer. Retention is arguably the decision that is mostly related to the experience on the platform, and indeed in the next section we show that it is associated with how efficiently the platform matches buyers and sellers in each city. 7.2 City Differences in Match Efficiency and Market Thickness For tasks like cleaning and delivery, it is obvious to expect that buyers would care about how easy it is to find a seller willing to provide the service at the desired time and location, and the price to pay for the service. In this section we explore how differently cities perform in this respect. To do so, we take advantage of earlier estimates of the matching and pricing functions (equations 11 and 12). Cities vary widely in the rate at which tasks and offers are converted into successful matches. Figure 12 plots estimates of A c from equation 11, ordering cities from the most efficient (San Francisco) to the least efficient (Miami). San Francisco is 2.37 ( A SF A Miami ) times as effective as Miami in creating matches: out of the same number of tasks and offers, 100 matches are created in Miami while 237 matches are created in San Francisco. Other than San Francisco, among the cities with the highest match productivity are Boston, Portland, Austin, and New York City. The other extreme includes Miami, Denver, Phoenix, Philadelphia, and Atlanta. There is limited heterogeneity in prices across cities, as Figure A3 shows. Here cities are ordered according to their ranking in the match efficiency parameter from Figure 12, and the plot displays estimates of K c from equation Most prices range between $54 and $65, with Denver ($42) and Miami ($68) as outliers K Oct2013 is normalized to The two equations 11 and 12 also include time effects, which we briefly discuss here. Over time, there is a limited 32

34 The most efficient cities are those able to retain the most buyers. 33 In Figure 12 the size of the marker for the match efficiency parameter is proportional to the retention rate η c estimated from equation 14. The cities with high match productivity A c also have high retention rate η c. The next step is to try to understand what can explain efficiency differences across geography. To this purpose we look at two metrics related to market thickness. The first natural candidate is the proximity of buyers and sellers: cities that more easily match tasks and offers might be those where buyers and sellers live closer together and can more easily meet and exchange services. This hypothesis is strongly supported by the data. Figure 15 plots the match efficiency parameter A c and the median geographic distance between buyers and sellers of paired tasks and offers. 34 In cities like San Francisco, Boston, Portland, and New York, which convert tasks and offers into matches at the highest rate, the median distance between buyers and sellers of paired tasks and offers is around 7 miles. At the other extreme, the distance in Philadelphia and Miami is over 20 miles. The second candidate measure of market thickness is related to task specificity. The idea is that an idiosyncratic task, which possibly requires specialized skills on the part of the seller, is harder to match than a standardized cleaning task, for which all that matters is location and time availability of one seller out of many good alternatives. To explore this idea, we look at the share of tasks posted in May 2014 within the top five categories (Shopping and delivery, moving help, cleaning, minor home repairs, and furniture assembly). Figure 16 plots this share of more standard tasks against the match efficiency parameter A c. In San Francisco, Boston, and New York over 60 percent of the tasks are posted within the top five categories, while Dallas, Miami, Atlanta, and Denver all have shares smaller than 50 percent. decline in match efficiency, but not sizable nor statistically significant. Figure A2 plots estimates of A t from the matching equation. The line is fairly flat between Spring 2011 and Summer 2013, with higher variability before and a small downward jump afterwards. This variation coincides with the staggered entry of TaskRabbit into the various cities. Indeed, prior to Spring 2011, only two cities were active, San Francisco and Boston. Around the Summer of 2013, TaskRabbit became active in the nine youngest cities, which are also those with lower match productivity estimates. Instead, there is a substantial increase in transacted prices over time. As Figure A4 shows, estimates of K t monotonically increase, from $30 in June 2010 to just above $60 in May The increase in price seems to closely track the task diversification on the platform towards more expensive tasks. Figure A5 presents the share of posted tasks in the 10 largest categories over time, combining all other categories in an eleventh group. It demonstrates how the period of fastest diversification, where the cumulative share of the top categories fell considerably, occurred in the Spring of 2011, exactly when the average price experienced the highest increase. 33 The most efficient cities also tend to correspond to those where TaskRabbit entered earlier on, and this still holds under different specifications of the matching function, as shown in the Appendix. 34 The correlation is maintained with the two other pairing definitions: median distance between buyers and sellers active around the same time window, and median distance between successfully matched buyers and sellers. 33

35 8 Conclusions In this paper, we have studied the problem of balancing highly variable demand and supply. This is a basic problem of the recently popular online peer-to-peer marketplaces for local and timesensitive services, such as Uber, Airbnb, and TaskRabbit. We have presented a static model of a frictional market where buyers and sellers post requests and offers for services. Our model specifies the possible mechanisms through which the market clears in terms of the elasticity of demand and supply to price, as in standard product markets, and to rationing, as in frictional labor markets. We have applied the theoretical model to study TaskRabbit, a growing platform for domestic tasks. We estimated utility, matching, and pricing parameters using variation in the number of buyers and sellers across cities and over time. The empirical application has allowed us to measure the gains from trade facilitated by the platform, the particular mechanism that equilibrates the market (highly elastic supply), and how market efficiency varies with location and market size. The natural level of market efficiency is modest, although certain cities, such as San Francisco and New York, are largely more efficient than others. The efficient cities are also those which grow the fastest, by attracting new users and retaining existing ones at a higher rate. City differences in the efficiency with which tasks and offers are converted into successful matches seem to be related to at least two sources of frictions: geographic distance between buyers and sellers, and task specificity. The market is able to efficiently accommodate fluctuations in buyers and sellers thanks to a highly elastic labor supply. When demand is capped by an exogenous arrival process of needs for cleaning or furniture assembly, and sellers flexibly adjust their effort in response to changes in relative demand, buyers find it profitable to post all their tasks, and can match at the same rate and price regardless of the number of sellers present in the market. On TaskRabbit, the elasticity of supply is likely due to the fact that sellers might be available to perform tasks only within a defined time window in any given month (say, every Saturday) or within a few miles from their house. A doubling (resp. halving) of demand would thus imply a doubling (halving) of the profitable opportunities for each individual seller, thereby affecting their offer submission. However, it is costly for sellers to search through posted tasks, and with some probability profitable tasks are not found by any seller and are left unmatched, or are found by sellers who turned out to be unavailable to perform them. Indeed, 13 percent of the tasks that did not receive any offers were canceled because the buyer reported having the task done sooner, and 19 percent of the tasks that did receive offers were canceled because one of the parties reported inability to resolve scheduling conflicts. So, what would happen if instead of having sellers search 34

36 for profitable submissions, the platform let sellers directly list their availability on a calendar and within a specific geographic area? This would eliminate search costs of supply, or at least make them independent on the size of demand, and would raise seller welfare while benefiting buyers and the platform by raising task match rates. The benefits of reducing sellers search costs and improving match efficiency provide a rationale for a platform re-design, which actually occurred in the Spring of Figure 17 is a screenshot of the current platform. Buyers can now select the category, location, and time for a given task request, and then either choose among the sellers available to perform that task type at the specified time and location (similar to the auction mechanism prior to the change), or have the platform automatically choose for them (similar to the posted price mechanism). We can actually also use our model estimates on TaskRabbit in a broader perspective, and roughly calculate the value generated by online peer-to-peer markets for domestic tasks. This includes TaskRabbit, as well as other platforms, such as Craigslist, ThumbTack, or Handy. We estimated the value created in each match to be $37, and a monthly average number of tasks per buyer equal to Taking the number of households present in the US cities where TaskRabbit is currently active, and assuming that 20 percent of them will be outsourcing domestic tasks online in the long run, an industry able to match 80 percent of those tasks would generate $920 million in value, and this in 18 cities alone. The advantage of a highly flexible supply is appealing for other peer-to-peer platforms. Consider Uber for example, the fastest growing ride-sharing marketplace. Matches between people wanting a ride and drivers are even more local and time-sensitive that on TaskRabbit. A person at the airport is likely to use alternative transportation if it takes long to find an Uber car (rationing) or the price is too high. Uber relies on having enough cars on the road to ensure reliability during normal and busy times, and achieves this by adjusting its price, balancing buyers response to use substitute services and sellers willingness to provide more rides. An elastic supply would require a lower price increase to adjust its effort and accommodate demand, thus limiting the buyers response to request fewer rides. The same delicate balancing of demand and supply for short-term accommodation occurs on Airbnb. In this case, in order to satisfy demand for accommodation the platform relies on a wide inventory of available listings to adapt to normal and high traveling seasons. Our paper has primarily focused on the short-term balancing of demand and supply. A valuable avenue for future research, facilitated by the large availability of data, would be to study dynamic 35

37 participation decisions of buyers and sellers in more detail than done here. Better understanding the drivers of user adoption and retention can help explain platform competition, both between multiple peer-to-peer marketplaces and between the online marketplace model and the more traditional service providers. 36

38 References [1] Patricia M Anderson and Simon M Burgess. Empirical matching functions: estimation and interpretation using state-level data. Review of Economics and Statistics, 82(1):93 102, [2] Mark Armstrong. Competition in two-sided markets. The RAND Journal of Economics, 37(3): , [3] Patrick Bajari and Ali Hortacsu. Economic insights from internet auctions. Journal of Economic Literature, 42(2): , [4] Frank M. Bass. A new product growth for model consumer durables. Management Science, 15(5): , [5] Olivier Blanchard and Jordi Galí. Labor markets and monetary policy: a new keynesian model with unemployment. American economic journal: macroeconomics, 2(2):1 30, [6] Peter A Diamond. Aggregate demand management in search equilibrium. The Journal of Political Economy, pages , [7] Michael Dinerstein, Liran Einav, Jonathan Levin, and Neel Sundaresan. Consumer price search and platform design in internet commerce, [8] Darrell Duffie and Yeneng Sun. The exact law of large numbers for independent random matching. Journal of Economic Theory, 147(3): , [9] Liran Einav, Chiara Farronato, Jonathan Levin, and Neel Sundaresan. Sales mechanisms in online markets: What happened to internet auctions? NBER Working Paper, (w19021), [10] Glenn Ellison and Drew Fudenberg. Knife-edge or plateau: When do market models tip? The Quarterly Journal of Economics, 118(4): , [11] Andrey Fradkin. Search frictions and the design of online marketplaces. Working Paper, [12] Robert E Hall. Employment fluctuations with equilibrium wage stickiness. American economic review, pages 50 65, [13] John Joseph Horton. Misdirected search effort in a matching market: causes, consequences and a partial solution. In Proceedings of the fifteenth ACM conference on Economics and computation, pages ACM,

39 [14] Kenneth L Judd. The law of large numbers with a continuum of iid random variables. Journal of Economic theory, 35(1):19 25, [15] Pascal Michaillat. Do matching frictions explain unemployment? not in bad times. The American Economic Review, 102(4): , [16] Pascal Michaillat and Emmanuel Saez. A model of aggregate demand and unemployment. Technical report, National Bureau of Economic Research, [17] Paul Milgrom and John Roberts. Comparing equilibria. The American Economic Review, pages , [18] James D Montgomery. Equilibrium wage dispersion and interindustry wage differentials. The Quarterly Journal of Economics, pages , [19] Dale T Mortensen. The matching process as a noncooperative bargaining game. In The economics of information and uncertainty, pages University of Chicago Press, [20] Chris Nosko and Steven Tadelis. The limits of reputation in platform markets: An empirical analysis and field experiment. Working Paper, [21] Amanda Pallais. Inefficient hiring in entry-level labor markets. American Economic Review, Forthcoming. [22] Geoffrey G Parker and Marshall W Van Alstyne. Two-sided network effects: A theory of information product design. Management Science, 51(10): , [23] Barbara Petrongolo and Christopher A Pissarides. Looking into the black box: A survey of the matching function. Journal of Economic literature, 39(2): , [24] Christopher A Pissarides. Short-run equilibrium dynamics of unemployment, vacancies, and real wages. The American Economic Review, pages , [25] Christopher A Pissarides. Equilibrium unemployment theory [26] Paul Resnick and Richard Zeckhauser. Trust among strangers in internet transactions: Empirical analysis of ebay s reputation system. Advances in applied microeconomics, 11: ,

40 [27] Jean-Charles Rochet and Jean Tirole. Two-sided markets: a progress report. The RAND Journal of Economics, 37(3): , [28] Margaret Stevens. New microfoundations for the aggregate matching function, with empirical and theoretical implications [29] Thaddeus Vincenty. Direct and inverse solutions of geodesics on the ellipsoid with application of nested equations. Survey review, 23(176):88 93, [30] E Glen Weyl. A price theory of multi-sided platforms. The American Economic Review, pages , [31] H Peyton Young. Innovation diffusion in heterogeneous populations: Contagion, social influence, and social learning. The American economic review, pages ,

41 Figure 1: The TaskRabbit Platform between January 2009 and May 2014 (a) Example of a posted task. (b) City list of posted tasks for sellers to search. Screenshots from TaskRabbit (accessed on December, 14th, 2013). The screenshots display information available to sellers about posted tasks. 40

Read Article to Me (Mobile Version)

Source URL: Read More
The public content above was dynamically discovered – by graded relevancy to this site’s keyword domain name. Such discovery was by systematic attempts to filter for “Creative Commons“ re-use licensing and/or by Press Release distributions. “Source URL” states the content’s owner and/or publisher. When possible, this site references the content above to generate its value-add, the dynamic sentimental analysis below, which allows us to research global sentiments across a multitude of topics related to this site’s specific keyword domain name. Additionally, when possible, this site references the content above to provide on-demand (multilingual) translations and/or to power its “Read Article to Me” feature, which reads the content aloud to visitors. Where applicable, this site also auto-generates a “References” section, which appends the content above by listing all mentioned links. Views expressed in the content above are solely those of the author(s). We do not endorse, offer to sell, promote, recommend, or, otherwise, make any statement about the content above. We reference the content above for your “reading” entertainment purposes only. Review “DMCA & Terms”, at the bottom of this site, for terms of your access and use as well as for applicable DMCA take-down request.

Acquire this Domain
You can acquire this site’s domain name! We have nurtured its online marketing value by systematically curating this site by the domain’s relevant keywords. Explore our content network – you can advertise on each or rent vs. buy the domain. Buy@TLDtraders.com | Skype: TLDtraders | +1 (475) BUY-NAME (289 – 6263). Thousands search by this site’s exact keyword domain name! Most are sent here because search engines often love the keyword. This domain can be your 24/7 lead generator! If you own it, you could capture a large amount of online traffic for your niche. Stop wasting money on ads. Instead, buy this domain to gain a long-term marketing asset. If you can’t afford to buy then you can rent the domain.

About Us
We are Internet Investors, Developers, and Franchisers – operating a content network of several thousand sites while federating 100+ eCommerce and SaaS startups. With our proprietary “inverted incubation” model, we leverage a portfolio of $100M in valued domains to impact online trends, traffic, and transactions. We use robotic process automation, machine learning, and other proprietary approaches to power our content network. Contact us to learn how we can help you with your online marketing and/or site maintenance.