 Okay, so why don't we get started? I'm really happy to welcome everyone. This is our second try at the economics of platform seminar series So it's to drive it to lose school of economics digital center We have way Lee and Carnegie Mellon presenting information transparency multi-homing and platform competition and natural experiment in the daily deals market. It's joint with Fong Zhu who is also joining us here today? But way is gonna take it away Okay, then you share my screen now Just actually just before you go forward I just said the format is going to be she's going to speak for about 40 minutes I'm gonna unmute everyone so you can unmute yourself and the expectation is you're gonna stay mute at least for the first 40 minutes If you have questions, you can chat them in and I will handle the questions And and maybe calling you but hopefully we're gonna try and keep moving through the first 40 minutes And then we'll be taking questions in the last 20 minutes. Okay, so that's our that's our format for today. Okay. Okay. Yep Thanks Mark for moderating the talk and thank you for the great opportunity to share my work with Fung Zhu on information transparency and multi-homing So information transparency Is commonly observed now on platforms But for instance, if you think about Uber to display the performance of the drivers through the rating Airbnb displays reading and reviews of the hosts Airbnb not only are reviews and readings displayed it also this I think I'm muted now. Yes, so it just yeah go yeah, it's perfect gone. Okay. Yep, so So you see on the right hand side, this is a typical Groupon deal So besides the price discount rate and duration of the deal It also displays how many deals have been sold so far So good things about displaying information, especially if information transparency is that it attracts consumers and facilitate matching between sellers and buyers For instance on Groupon studies have shown that a display in new sales can encourage consumer hurting behavior and increase sales However, if you think about the downside of information transparency is that not only can consumers see the information But also rivals can see it. For instance, even social can use information to identify high-quality merchants and try to poach them and Attacks them to multi-home So that's bringing us to another common phenomenon in industry, which is multi-homing So my home is now commonly observed in public markets, but actually because of the low adoption and switching costs Think about the same consumer can visit both Groupon and leaving social to get new deals Also, he or she might have both Uber and Lyft on his smartphone Similar thing on the merchant side, right? The same merchant can put up deals on both Groupon and leaving social and also the same driver can drive for both Uber and Lyft So when combining multi-homing and information transparency together You see that rivals now can leverage this opportunity to poach high quality merchants from the focal platforms and ask them to multi-home on the rival platform and This reduces exclusivity and reduce the differentiation between the two platforms because now they're more overlapping merchants on both sides And that's kind of that can hurt the focal platform That's why in practice many platform attempt to prevent rivals from multi-homing using the right strategies For instance, Uber and Lyft actually encourage their drivers to single home Game console providers also offer incentives to top-run game publishers for sending exclusive contracts So this is particularly relevant because in the gaming industry the heat game can drive sales of the console Alibaba also discouraged its merchants from multi-homing by actually prioritizing single-homing merchants through their ranking algorithm Recently eBay also sued Amazon for exploiting eBay's internal messaging system to viewers top sellers and also sell on Amazon Now despite all these Strategies in pure evidence on how these strategies impact multi-homing and competitive dynamics is limited and remains unclear How these strategies can impact competition So what do we know about multi-homing just a brief overview of their literature So there are a few empirical literature on multi-homing and studies have shown that model platforms should prevent their own users from multi-homing and model homing indeed makes dominance of the focal platform life-likely and Also exclusivity the opposite of model homing benefits and some platforms So most of the work on model homing is actually on the theory side Most of the work either abstract from a platform's role or assume that platform uses price as the only tool to influence model homing behavior But as you can see from the previous side these days from are active in terms of taking actions to prevent model homing from happening So price may not only be the only tool Also, most of the work restrict model homing to be only on one side, but actually because of the complexity of the problem Some of a few work do allow model homing to be on both sides both consumer side and the merchant side But then found that the equilibrium model homing exists only on one side So in our work, we're going to highlight Platforms active active role in terms of influencing model homing behavior and also allow both sides to not at home Okay, let me now talk briefly about our approach So it's going to start with a game theoretic model to illustrate key trade-offs of other players and then derive hypotheses to be tested Importantly we allow for model homing on both sides of the market And we're going to model decision-making of consumers platforms and merchants Now given that policies I'm going to implicitly test Hypothesis by leveraging an exogenous position from Groupon So what it does is essentially it limits the level information transparency on the focal platform on Groupon Which can have an impact on rebels and we're going to look at how the policy affects rebels model homing behavior In this case leaving social the largest largest rebel going to zoom in on leaving social and look at how leaving social is not at home behavior changed and Then how does the policy impact industry variety? Consumers model model homing behavior and also finally how the policy impact rebels profitability After coming for the revenue the change in customer base and also merchant acquisition costs So here is a specific policy setting we're really can't So Groupon went IPO in 2011. So in October 2011 Groupon changed how the deal sales information was displayed on a steel counter So if you look at the left-hand side of the graph, this is original how the deal was displayed in the steel counter It shows the exact number of deal sold. So here is 275 But then after the policy change, it actually shows an inaccurate numbers over some number So what it does is that I should announce in one of their blog posts that they randomly round out that percentage of the actual sales And display a rounded number and actually they change the percentage of rounding from time to time So that is never possible for outsiders to guess So the intention of the policy was actually to prevent outsiders for instance financial analysts from estimating Groupon's revenue And this is this can possibly hurt the power of the company on its journey to go in public So essentially what the policy does is that it makes information transparency less available Which can possibly prevent or raise the cost of multi-homing for the rebels So this policy change is actually a very desirable design for our policy for our research Because the policy was not intended to deter multi-homing It was likely to be exogenous to multi-homing related factors And later on we actually turned out a series of analysis to show that exogenous actually exists Okay, so just give you a quick preview of the results before we go into the details Out of the policy change Groupon limited its information transparency And that makes living social the rebel copy fewer Groupon deals and have to increase its efforts to source more new deals As a result, you see the merchant side multi-homing decreased because of the policy As a result because living social amount of source more new deals the industry deal variety increased And we also identify interesting seesaw effects in that although merchant side model homing decreased consumer side model homing increased There were more model homing and living social exclusive consumers on website And overall living social customer value increased But then this come at a cost because it's more costly to source new deals Such that living socials model homing a merchant acquisition cost increased Taken together the policy change hurts living socials profitability Okay, uh, let me pause here and see whether there any questions clarification questions Okay, so now let me move on to the theory model Because the paper is mainly on the empirical side I'm going to be relatively brief on the theory model try to illustrate the key trade-offs of other players Instead of getting to the details of notations and derivations So we model the players decisions first of all the platform decision of multi-homing So we think about the benefits and costs of multi-homing On the one hand, multi-homing reduces deal uncertainty because now essentially, uh, you just look at the past sales information of that merchant And you can predict that the popularity of the merchant So it reduced uncertainty And also because the merchant has worked with Groupon before it's likely that it's less costly or they're more willing to work again So the acquisition cost of the merchant can be also lower. So that's the benefit However, the downside of model homing is that as we as we just talked about it reduced the differentiation across different platforms any intensifies competition So you can imagine that a equilibrium platform having set up both to multi-home and copy from the existing pool And search for new deals, but then add some uncertain popularity So we're going to model the decision of how many deals to copy versus search for for the platform to maximize its profits And here we're going to account for different values of model homing versus single-homing consumers And also allow for different costs of copying versus searching deals So the idea is that search deals, which are new It will be probably more costly to acquire than copying from an existing pool And also allow Groupon and living social to have different costs And you can imagine that the policy change reduced the information transparency So that's going to raise the rival's cost of copying in our model Now the second player consumers then make the decision of adoption of platforms So this is driven by the classic indirect network effects So in that consumers care about the quality and also the quantity of the deals And it allows them to value copied versus search deals differently And you know that consumers will have incentive to model home if the platforms are more differentiated Meaning there are fewer overlapping deals across Groupon living social Now finally the third player merchants They also make platform adoption decisions And that's also driven by the indirect network effects in that they care about the number of consumers on each platform And will allow them to value model homing versus single-homing consumers differently And merchants have incentive to model home if there were fewer overlapping consumers So that's the two platforms are more differentiated So here's a time of timing of the game At the beginning think about as a period zero There's a set of merchants that have worked with Groupon before so their popularity is known So their sales are already a public information. This is essentially the information transparency part And then we assume that Groupon is the stackerberg leader He decides how many deals to copy from this pool and how many new deals to search for And we make this assumption that Groupon is a stackerberg leader because over 90% of the time Groupon either entered city first Or entered city the same time as living social And then living social decides how many deals to copy and search for as well Merchants given that they have approached by other platform Then decide whether to work with the platforms anticipating the number of consumers on each platform Similar thing for consumers Consumers decide what which platform to use accounting for the number of merchants on each platform So you can imagine that in equilibrium all these players Groupon and living social make their optimal decisions of copy versus search Merchants and consumers make their own optimal adoption decisions accounting for the model homing incentives Now we make one assumption on the merchant side. So Merchants have heterogeneous popularity So we're going to assume that the most popular merchants will always be approached by both platforms So when we solve for the platforms copy and searching strategies Uh, we're going to look at only the moderately popular merchants So the idea here is that uh, the most popular merchants think about a best seller or a top popular restaurants So their dual popularity can be commonly known probably from other sources other channels And it's likely that the the benefits of of working with them is always higher than the cost So that merchants a platform always have incentive to work with them But then for the moderately popular merchants, they're more likely to be affected by the policy because their information Their popularity is less likely to be known by other channels So this dual counter for the information transparency unfold platform become more important For your quick question. Are the uh, are they choosing simultaneously group on living social? Uh, so group on good question. So group on uh, is a stacker book leader and linear social follows But then consumers and merchants are going to make decisions simultaneously Did that answer the question? Okay. Yeah, all right Okay, so when we solve for the Equilibrium and then change the the cost of the copy and see how the results change. So here are the hypotheses derived So in terms of living social because it's more costly for living social to copy from group on You can imagine that living social is going to model home fewer group on deals and increase its efforts to source more new deals And remember we assume that Everybody always copy the most popular merchants, but then leaving social copy fewer moderately popular deals So the average deals that leaving social copy from group on is going to increase And remember because living social copy a fewer and search for more deals The energy wide deal variety is likely to increase and leaving social is going to contribute more to the deal variety On consumer side consumers model homing response can be mixed right on the one hand Because the two platforms are more differentiated that leaving social have more unique deals Consumers have incentives to model home more But on the other hand the the search deals the new deals the popularity is unclear is not guaranteed So if the search deals are not often enough high quality Then consumers might find is not worthwhile to model home and visit leaving social So our hypothesis goes if the search deals are often of high quality Then consumers will be more likely to Multi-home and also there will be more exclusive consumer selling in social just because they can benefit more from the new deals And as a result, there will be fewer exclusive group on consumers Okay, just a little more about the difference between copying and sourcing or developing new deals What is the what exactly is the difference? Yeah, so copying essentially is not at homing remember Information transparency says there's at the beginning there's going to be a set of merchants who have worked with group on so their deal sales information is known So copying from that pool essentially means working with those merchants again. So that is multi-homing. So I'm going to use actually copying and multi-homing kind of interchangeably in this setting And searching for new deals would mean just you know go outside of that pool and search for new deals And those deals will be unique on leaving social And there's different costs Right exactly at different costs. So the policy is going to affect the cost of copying or multi-homing, right? Because When when the the the information becomes less less clear Then it's more costly for leaving social to identify good quality merchants, right? So the cost of copying is going to increase after the policy and But then the search searching cost remains the same per merchant, right? So given that the cost of copying is higher and cost of searching is the same Leaving social have been send it to copy fewer and search more Okay, that's good Yeah, great excellent. Now. Let's talk about the empirical side So we have data from two sides the merchant side and consumer side on the merchant side model homing We have data from ebay, which is a market research company that aggregates all deals During this 36 months The policy happened in the 22nd month so that leave us with 20 21 pre and 15 post policy months So for every single deal we observe the sales the category of the deal for instance, whether it's a beauty fitness Or entertainment observe the price the discount rate duration Where the market where the the deal was launched the market which platform launched it and merchant information And we focus on the top 100 cities with the most cumulative number of deals Deal sales And then we're going to focus on us markets and also markets that have experienced both pre and post policy months And that leave us with 160 000 merchants and 600 000 deals So you see that group on it is the number one player in the market followed by leaving social and other sites Now on the consumer side, we observe consumer side by browsing records from comscore during 24 months period Which leave us with nine purie and 15 post policy months So for every single set visit we observe the exact time stamp The machine ID that can be used to identify the unique consumer and house the demographics and where they live the zip code We're going to focus on consumers who live in the cities that we observe in the first dataset And also consumers who have visited at least once leaving social warm out or group And that give us a 6 000 consumers and 13 000 set visits Okay, let's take a look at some data pattern So here i'm tabulating the typical price discount duration sales of group on deals leaving social deals and other sites Also the past experience of the merchants So for instance here you can see the first column says a typical group on merchant Has has worked with group on point nine four times in the past Has worked with leaving social point two nine times in the past and has worked with other sites 1.202 times in the past So this seems to suggest that multi homing exists on the merchant side Also, if you look at overall 60 percent of merchants have offered group on deals 50 percent have offered leaving social deals and 75 percent have offered deals on other sites So these numbers sum up to over a hundred again suggesting that model homing behavior exists on the merchant side Okay, any questions? All right, so let's talk about a hypothesis testing So we first look at the impact of a policy on leaving socials model homing strategy Hypothesis one says after group on policy change, leaving social is going to multi home fewer group on deals and search for new New deals. So here let's look at some model free evidence first I'm plotting the percentage of leaving social deals at multi home group on so Every observation is a city month and averaging across cities and probably over time So here there are two curves because you were to ensure That the result is not driven by different cities in their different life cycle. For instance, some cities have Group on enter earlier some cities have group on later I'm going to group cities into two two types The solid line represents cities that group on enter before the 10th month And the dash line represents cities that group on entered after the 10th month They can imagine that solid line represent more mature cities. The dash line represent more younger or less mature cities And vertical vertical dash line represent when the policy change happened So as you can see a percentage of leaving social deals that copy group on increase before the policy and decrease after the policy Which is consistent with our policies that leaving social model home fewer group on deals And this is further validated by running a regression of the percentage of model homing deals in category j market m time t on leaving social On a post policy dummy and post policy dummy interact with a linear city specific time trend So here t m is a city specific time trend that potentially capture cities different cities in their different time life time cycle and also controlling for time fix effects Market characteristics category fix effects month of the year fix effects and so on So as you can see from the first column of the first of the table The interaction term on the the coefficient on the interaction term is negative So suggesting that indeed after the policy Leaving social reviews to model homing of group on Okay Can you start talk about uh, was other things changing with this policy Is that the only thing that changed and also kind of is the ipo right around like where does the ipo and timing on on these regressions? and then another question is like how how truthful how sure are we that the Group on posting accurate information when they say they've sold this many. I mean, is it just marketing or is that a real Are you confident that that number is a real number? uh, great question actually for the first question about Whether the policy is indeed exogenous. So this is not driven by anything other than model Things that are for instance ipo rather than, you know, deter Model homing if you can wait for a few more slides at the end we can have a conductor or bus and check Uh, we actually list a series of analysis that test whether this policy is the exogenous Essentially, the idea is that if the policy was actually to deter the model homing instead of, you know Exogenous the model homing then we should expect other policies related model homing to also change right for instance Group on can offer more favorable deal terms to the merchants when negotiating with them We're offering more favorable commission rates to the merchants and we find that that's not a case Looks like the change in the the essentially what we identified here is really just from the change of of the deal counter uh, and the second question is about You better understand quickly it's about whether indeed it is a group on deal counter That uh, that drives model homing and also comes uh, the the rebels change in the policy. Is that correct? So that part yeah, I think I think it's you know, they're posting this number more than 150 sold Right, and I guess is it accurate or is it just a made-up thing? Oh, so what they claim in their blog post is that This number essentially is intentionally rounded down So it's definitely a reflective of the actual How the actual number because you will see that they appear in different magnitudes Some says over 150 something over a thousand So the magnitude can can can probably give a signal that there are indeed real differences behind the actual sales It just a bit randomly round it down so that you can never guess what is the exact number Okay And then can you see how many people are on living socials like how many customers living social has you know that number Yeah, so we have data on the consumer side as well. Remember we have a consumer's white browsing records so Here although the caveat here is that we observe how many consumer visit group or visit the social or visit both of them Right, so we group them as group on exclusive living social exclusive or a lot of home consumers But we do not know exactly how many consumers actually bought from these websites So if the question is uh, how many accurate consumer do we know? Well, uh, we do know something from group house of where it's 10 k philings But we do not know that from living social here The underlying hypothesis that uh, the number of website visits is going to be proportional or at least reflective Proxy for the actual consumers they have These total number of visitors for living social change during this time too Yeah, yeah, exactly And you're going to see that we test this empirically and show that uh, living social consumers indeed change and actually increase the policy Sounds great. Okay, excellent All right, so besides the change in number of consumer number of merchants not at home We also look at the quality where the average sales of deals and living social not at home from group Remember the idea is that uh, because they continue to copy the most popular merchants But reduced the the moderately popular merchants Then the average sales of deals that copied should increase So here again, uh, we're using uh, we're plotting the log sales of model homing deals and other deals Over time, right? So you can think about other deals as a benchmark to control for any overall change in the popularity of living social deals And again, because uh, to try to ensure that this is not driven by the different life cycle of cities We're flawed by two types of cities earlier mature cities and younger cities As you can see for both cities after the policy change There's a consistent larger gap between model homing deal sales and other deal sales seems to suggest that our policies hold And we'll further conduct a different depth regression analysis by regress per deal sales of a living social deal On a post-policy dummy the post-policy dummy interact with the model homing dummy, which equals one if it is a model homing deal And also control for a category of fixed effects characteristics of the deals market demographics and also month of year fixed effects So here only existence this control is to say well it equals one If this merchant has worked with living social before and we control for the only existence or past experience as well As you can see from the first column the interaction term the coefficient on the interaction term is positive So this suggests that uh living so model homing deal indeed increased Sales after the policy change and the results are robust if we replace the market characteristics demographics with market fixed effects All right So remember living social copy fewer deals and search more for new deals So we can imagine that in the true wide deal variety can increase up the policy Intuitively consumers value the unique deals that appear on a website during a specific time period So here we count a deal to work towards a variety deal If the same merchant has not put up deals on any other platforms in the past three months So we also change the cutoff to two months six months the results are very robust So here i'm flooding the percentage of variety deals among all deals As you can see the percentage decreased before the policy Potentially because of an exhausting merchant pool right so as the the platform grows There are fewer and fewer merchants who have ever not worked with any platform before or who have worked only only with one platform But then the percentage starts to increase after the policy seems to suggest that the variety indeed increased after the policy change And this is consistent with the annual regression of variety deals on post policy dummy The post policy done interact with the city's specific linear time trend Controlling for city fixed effects and month of the year fixed effects as you can see from the first column this table The interaction term is positive suggesting that the deal variety indeed increased after the policy So hypothesis three says that living social contribute more to the deal variety So we can replace this dv of the regression with living socials contribution here measured as the percentage of Variety deals that come from living social among all deals all all variety deals So as you can see again, the interaction term is positive suggesting that indeed living social contribute more to the variety after the policy Okay, now let's talk about the consumer side Remember after the policy change high policies four five success There will be more multi-homing consumers and living social exclusive consumers While there will be fewer group on exclusive consumers So we look at the percentage of group on consumer living social consumer and multi-homing consumers Over time so as you can see from the first plot the solid line represents group on So the the percentage of group on exclusive consumers decreased And the percentage of living social exclusive consumers increased right on the right hand side Multi-homing consumers also increased So this is again consistent with our hypothesis that living social benefited from more multi-homing consumer and more exclusive consumer who live in living social So what's interesting is that not only do living social benefit from more consumers They also benefit from more set visits per consumer So here we zoom in on multi-homing consumers Okay, we look at the fraction of their set visits to group on versus to living social because they visit both sides And we plot what's a fraction of set visits that goes to living social versus group on in this block As you can see after the policy They visited group on less and visited living social more So this suggests that being social not only benefit from more consumers But also from the same consumer visit living social more Okay, so far we know that a living social's profitability is impacted in several ways So first of all living social have to multi-home fewer group on deals But then the average deal sales of multi-homing deals increased So the total revenue after accounting for this quality versus quality change might increase or decrease So this is an empirical question But then living social do benefit from the consumer side There were more multi-homing consumer and living social exclusive consumers And the value of the customer base might increase so it suggests a higher profitability But remember this concept cost because living social have to source more new deals A higher more cost So the acquisition cost the merchants might increase which can negatively impact profitability So what is the overall profitability impact? So let's start with the raw raw Revenue just look at the number of deals sold together for living social So we request the total revenue Also times the commission rate of living social We request that on the post-policy dummy a post-policy dummy interact with the linear city specific time trend Also controlling for city fixed effects the month of the year fixed effects And if you look at the first column the interaction term is negative Well, this suggests that the policy Hurt living social if you account for just the raw revenue they put in their pockets However, we need to remember that we also need to account for an increase in customer base because they can represent future revenue in future profit So how to calculate the value of customer base? While the marketing concept of clv customer lifetime value coming handy So clv of the total customer base you call the number of customers times the clv per customer So we observe the number of living social customers directly from outcome score data And we scale it up to the full population For customer lifetime value So which essentially is what's the average cumulative sales per customer? We obtain that from group on 10k filings because living social was not public And that number was around 3.5 per customer And later on we change the number to be higher and lower and it doesn't change the results so If you just look at the change in customer base use clv as a dv to run the previous regression this regression Just replace the dv with the customer life value and turns out that the interaction term is policy Which suggests that living social do benefit from increased clv increased customer base value And now we can add this change in clv back to the original raw revenue and get an updated revenue measure And when we use this as the dv to run the regression, it turns out that the interaction term becomes positive Suggesting the policy change actually helps living social when accounting for increased customer base So remember we need also account for the merchant acquisition costs So in this industry a merchant acquisition cost mainly is sales force expense Because it's very important for the for the platforms to have a group sales sales guy to knock on the merchant stores Negotiate contract with them and a persuade to put up bills So this is actually one of the major costs for the daily deals platform Again, because we do not observe living socials Directly observe their sales force expense because it's not public We look at group on sales for expense from their 10k and s1 filings And the idea is that the sales force is like sales force expense is likely to be Affected by the number of merchants or the number of deals and also how many deals are new because again search New deals can be more costly to acquire So we first progress the total acquisition cost of the group on deals On number of merchants for group on a number of fraction of new merchants for group on We got these estimates beta 1 beta 0 beta 2 And these are essentially coefficients that can proxy how the total acquisition costs can be effective And we take these estimated coefficients now plug in the living socials number of new merchants and fraction of new merchants We can have an estimate of living socials acquisition costs So if you do use living socials acquisition costs as dv to run the regression Now we get a positive coefficient on the interaction term We suggest that after the policy indeed we see living sort of have an increased merchant acquisition costs Now we're finally uh ready to put everything together. So we used to draw a revenue At the increase in customer customer value subtract the acquisition cost of the merchant to get More thorough more overall profitability measure and we use as We use it as the dv and run the regression and turns out that interaction term is negative So so that's in that the policy change hurts living social after accounting for both increased customer base and increased acquisition costs Okay, so So far we use the lv to be 3.5 and assume that essentially living social group. I have the same acquisition cost So it's a very robust if we very very clv to be one or five Or allow living social to have higher or lower acquisition costs than group And intuitively the policy impact is more positive if the customer value is higher Right, because that's well how living social can benefit and the policy impact is more negative If the acquisition cost of the merchant is larger and in fact the policy impact is going to disappear when uh clv is the clv of the The customer is high enough and much acquisition cost is low enough This actually suggests that different cities or different industries with different magnitude of Customer value and acquisition cost of merchants might have different results of the policy impact Industries where consumers are particularly valuable and the acquisition costs of the merchants are relatively low You can expect that the gain out with the loss for the rival platform So which means it's actually more difficult for the focal platform To drive out its rivals because of their two-sided model homing and the seesaw effect Okay, so remember we'll talk about the entire policy change Analysis relies on the desorginity of the policy So if indeed the top policy was to determine that homing, we should expect to see changes in group on other deal Strategies to relate to model homing around the same time not just this counter change, right? So we first test whether there's any change in the deal terms for instance discount rate and duration offered to the merchants And we run a regression of a log deal sale a discount rate and log duration On post-policy dummy and interaction term and we find that coefficients are insignificant Meaning there's no significant or systematic change of the deal terms that group on offer before and after the policy also group on might be able to offer more favorable commission rates to the merchants and persuade to persuade them to do the last model homing and work exclusively with group on So although we do not have data directly about group on commission rate We do obtain data on leaving socials commission rate from a market research company So as you can see before and after the policy, there's no meaningful change in the systematic change in the commission rate that leaving social offer to the merchants Which might suggest that a group on might not have systematically changed the commission rate either Now second robustness check If the group on deal counter is indeed used by the rival's model home meaning all the change we have observed so far Is indeed due to the deal counter change Then the policy change should not affect leaving socials model home behavior behavior towards other sides Right because group on change is still counter but others do not So we should expect that a leaving socials model homing towards other side should not be affected by the policy change And if you look at the percentage of leaving social deals that model home other sides Over time you see a consistent trend. There's no systematic change in that trend Also, if the deal counter again is indeed used by rivals in model home Then other side should also be affected by this policy because the policy change should also affect other sides model homing behavior towards group on Right. So we now also plot the percentage of other deals that other sides that model home group on And you see a very similar trend as leaving social right before the policy There's an increase in percentage of model homing deals And after the policy you see a decrease Which seems again to be consistent with the fact that indeed everything is driven by group on still counter change And finally if leaving social indeed use group on sales information Then we should expect this effect to be moderated by deal sales uncertainty Well, how valuable the sales information is the intuition is that the more valuable the sales information where the higher the 30 years Then the the information become more valuable the policy impact should be higher And we further render different different regression in the original analysis But then use uncertainty as a third As a third control and you see that the the interaction the triple interaction term is positive Which seems to suggest that indeed deal sales uncertainty moderates The policy impact Okay So to conclude Uh, we find that limiting information transparency on the focal platform can reduce rival's model homing on the merchant side And that benefits industry by increasing the industry wide variety But then also as some cost right leaving social benefit more from model homing behavior on the consumer side And overall leaving social still get hurt because of how profitability decreased And we highlight this seesaw effect in that reduced model homing on the merchant side can increase model homing on the consumer side It means that the platform needs to be really cautious about whether and how to disclose information Because uh, they need to come forward the changes on the consumer side as well and merchant acquisition costs So it becomes probably more difficult for the focal platform to dominate the market given the existence of two-sided model homing All right, I think that's all I have All right, okay. That was great. Thank you very much Thank you. Now what's going to happen is uh, people can raise their hand with their zoom Uh button and uh, I'll call on people or or maybe I'll call on people and And then we will take the question Um And so for the first one, I think julian has a question If anyone wants to raise their hand or somehow you don't know how to raise your hand And you just want to chat in or put up your video and wave your hand up and you can raise your physical hand And that's fine. Whatever you want to do is fine. But um, Let me uh, I have to unmute people, I believe They can do it themselves if they want Okay, uh, well actually Yeah, julian's here here. Why don't julian, why don't you take the first? Sure, thanks I had a question about the robustness check one Yeah, um and You I mean you're saying that Group on and living social shouldn't well, they didn't adjust their deal terms and commissions Yep In response to the policy change and that was sort of evidence that this is an exogenous change But I would have thought if whether the change is exogenous or endogenous if there was a if there was um You know this policy change basically affected a parameter of the model, which is the cost of living social copying Group on deals, right? So given that affected the cost Of living social copying shouldn't have that then led to Living social and group on optimally adjusting their deal terms and commissions So I sort of whether it's exogenous or endogenous policy change or change I would have thought it would still lead to those kinds of effects in terms of changing commissions and deal terms And therefore I don't really understand why the The fact that you don't observe it those changes implies that it was somehow an exogenous change Okay, so I guess our uh, uh intuition is that uh, the policy because we try to convince you that the policy is not intended to determine that homing so this this uh pattern of you know change in deal terms or commission rate is Sometimes consistent with the fact that If they are uh intention was indeed to determine that homing they should also change these ones. So these are not like Uh, there's policy response to the change in policy But rather whether they change these things simultaneously were Around the same time with the policy change So either way we do not see any change in deal terms of commission rate Which means that intention the intention probably is not to term out of homing because there's no change at all Besides just the change in deal pattern, which is consistent with their original claim Is there any change afterwards like Let's see for instance, you see the the the graph So the vertical line says this is the policy change, right? So after the policy, there's no change systematically in terms of commission rate Right, so my I guess my point is if they don't if they don't change it even in the long run after the change Then I should they change it At the same time like the theory would say if there's some interaction between the two variables Then eventually they would change the commission rate, but you're not seeing that change in the long run And therefore the absence of it and absence of it at the same time doesn't really prove anything Oh, right. So uh, so you if you look at the time We know we're looking at this so this is actually six months before the policy and six months after the policy So, uh, you can imagine that as you said After the policy there might be longer long run, you know long run responses to the change in competition dynamics because although the policy was not Intended return model homie. Indeed. There's changes in model homing behavior in in the marketplace So they might have other responses, but after long time after the policy change So here we're more likely to be just, you know, just looking at within this time window relatively Six months before and after we look at whether their systematic change in in in their deal terms and commission rate so, uh, I can definitely agree with you that if there's, uh Hispanic response it can happen probably We do not have it on after beyond for instance beyond six months of the of this period So could potentially happen during a longer time horizon And also just add to waste response and Julian so so also part of our objective is also just to make sure that This observed dynamic changes are due to the kind of the reduction in the information transparency, right? So so there might be lots of other Actions right taken by group on at the same time So so even though our mechanism we're trying to argue is because of the dual counter change If a group point is also doing bunch of other things that at the same time that it's not very clear whether it's It's because of the information transparency or because of other actions taken by group. So this is also partly to kind of just beef up some confidence around the mechanisms Yep Louise All right, thanks So I already had a private exchange with fang But I thought it might be useful to sort of bring this out So one thing that we didn't have a chance to see in detail because there's not so much time is theoretical model But it seems to me that the the authors do have a good theory model that Brings together all the different Moving parts and if that's the case then one thing that would seriously encourage them is to try to Not necessarily estimate a theoretical model. I know that that's you know, Social estimation is a different animal altogether, but to at least calibrate that theoretical model based on the results from the from the Difference and differences analysis because then it would allow us to get a much Better feel for what the size of the effects are and what where they're coming from As opposed to just rely on on survey reduced form regression analysis, which as as kara mentioned earlier I mean it's always fraught with a series of problems, you know, there could be many other things happening at the same time, etc etc Having at least a calibration of this theoretical model. I think would In my opinion would increase the value of the Exercise substantially because you give you at least an idea of what are the orders of magnitude effects We're talking about in here and why they're happening even like I said if you don't estimate the model per se Yeah, that's a general point about a lot of this literature by the way, not just about your paper I think sometimes we're so close to doing something that's substantially better than just A times is reduced for regression And and and we don't I'm not sure exactly why not especially if you have a theory model Yeah, yep great point Oslo Yeah, thank you So I have a question more like a comment on uh, you you said that after the policy change more There is an increase in the multi homing on the consumer side And then you think it to the you said that the u-document and reaction to merchants less multi homing Consumers react by multi homing more Uh, I'm not sure that you can really conclude uh, and maybe with this causal statement because What the policy changes also at the same time uncertainty on the deal making it through so Uncertainty about whether this can indeed be at the end a deal or not also increases when there's this So regardless of what merchants do Consumers might multi home. So I I'm not sure that you can really conclude in saying that This is documenting what you suggested And I have another point, uh, which was about this customer life value calculation that you showed I mean, I understand that this is something coming from marketing But it's very important in this context because those consumers who are moving to Living social might have very different value per customer than an average consumer So there happened, you know, they are more marginal types. So they might have lower customer value So I'm not sure that taking the same customer value per customer would be a right Uh, you know, probably you're overestimating the value of those consumers moving to living social Okay. Yeah, thank you for coming. Uh regarding your first one You're exactly right in the sense that Uh, the search deal quality actually determines a lot of things, right? If the search as as you can see from our policies He actually relies on that the quality of the search deals are of enough quality So in that case that consumers will more likely to multi home So our hospital testing is actually, you know, uh, based on this because Uh, uh, to your medical model, we cannot conclude Exactly that the consumers are more likely to model home For sure because it depends on the sales of the quality of the search deals And the hypothesis testing surface in pure for evidence to show that indeed this is the case So it looks like in practice the search deals are of you know, high quality So that's uh the common uh response to your first point and regarding the cov analysis agree that consumers model home consumer single-home consumer might have different values um And uh, and this essentially is the first kind of a first cut analysis to look at average consumers But I want to do do want to bring out one point is that Even if a model home consumer would do it's actually assume that a model home Single-home consumers have different values But then that's potentially because model home consumers are likely to be avid deal seekers So they probably uh have higher sales than single-home consumers But remember their attention is split right? So model home consumers have to split their sales across multiple platforms while single-home consumer Shop only from that platform. So, uh, it's actually not clear Whether after you come for the split of revenue or split of wallet Whether the per customer clv is still very different between uh model home consumer versus single-home consumer We do have some data raw data from from the customer Uh, uh, routing behavior on com score. So, uh, model home consumer remember correctly visit seven times on average In the month across all problems While single-home consumer visit two or three times a month, I believe so it's roughly like two times or three times of Of a single-home consumer. So that's my my estimate, but you're right So a more accurate analysis can be, you know, break down customers by model home versus single-home consumer Although there's one concern which is, you know in our analysis Some model single-home consumers become model homing after the policy change, right? So if we we say that, uh, they have a different values, then we might be assuming essentially the same consumer have different values before and after Which again is it's not that consistent with the saying that, uh, you know, consumer might have their own intrinsic tastes So there seems to be some trade-off in terms of the assumptions we're making Uh, hi here, uh, I'm curious Uh, I'm curious to know a little bit more about the copy coping strategies because you sort of Rely on the assumption that you copy after sales on the competing platforms are realized Uh, and so do you have evidence of these sort of staggered? Uh adoption of the same deals by the same merchants across different platforms Uh, oh, yeah. So that's a great question. So one, uh, one Industry detail here is that in fact, uh, so the same merchant cannot put up deals simultaneously on two problems So this is part of the kind of, uh, uh, uh, uh, uh, sales arrangement of the deal terms So it says you cannot, well, you're working with Groupon, you cannot work with other platforms So in fact, what happens in practice is that model homing do happen kind of sequentially So first they work with Groupon and then the deal sales information is known, right? And afterwards, uh, if leaving social would like to copy we're not at home They approach the same merchant with that information in mind as well So so kind of leaving social kind of as a stackable follower already know what happens before on Groupon's website So in practice, this is actually indeed what's happening. There's kind of stackable mover first mover versus a follower And you have sort of at least anecdotal evidence that the deals that are copied are the more successful ones Yeah, so that's a great question. So I actually have a backup slide here. So We do look at, uh, what's the probability of being, can you see the stack, hopefully? So, uh, so the probability of being copied by by leaving social actually decrease with sales rank So essentially the more popular deals on left hand side are more likely to be copied and the less likely The less popular deals are less likely to be copied. So there indeed is a kind of a decrease in trend in terms of ranking of the sales If I can if I can kind of chime in to that to Kira's question But uh, do you know that after the deal is copied by leaving social is it More successful is it's a more profitable for living social because maybe the second deal on the same For the same merchant may actually be less profitable Yeah, so first of all, we do account for that if you're one of our regressions the Regression on the different different regression So here we do account for only existence Which is said essentially said if this deal have worked with living social before to account for this a dummy Just in case that if the same merchant pull out deal twice the second one might not be as successful at the first one So we do account for that. Um Although our paper I remember There's other papers actually came at all in 2017 They do find that the same merchant if they put up deal the second time on another platform The sales can be lower. So that's what they found in their paper Although we do not directly speak to that but that's what I remember from from reading their paper So there it seems to indeed indeed exist a kind of a decline in sales For the same merchant over time for if they put up deals on multiple sides Yeah, so that's consistent with your intuition. Yep Julian I think has a question Um, yeah, I had a question Um about implications Yeah to other marketplaces And it would seem like in standard marketplaces if They become less transparent with their information. So imagine airbnb doesn't show full information or amazon doesn't show full information Then you would expect consumers would be directly hurt by that and would be less likely to use the platform So they face this trade-off that they would like to limit copying by rival platforms on their merchants But at the same time they want to Keep the information there for their consumers And it seems to me in your paper, you don't have that trade-off. Is that right because in the model at least in the model This you didn't really mention any direct effect on the consumers in terms of With less information they get less value from the platform You're right in the sense that uh, yeah So in the theory model, we do not allow for essentially consumers to be hurt because of uh, Limited information transparency. So you're right. Uh on the main practical side I think the overall effect to identify it's kind of the overall the net effect of for what it describes So consumers can be hurt by the more, you know, inaccurate information But then uh, the change in the overriding can be a positive force in driving them to, you know, continue with the website So I think the results will identify here the empirical Result is the net effect of the two Yeah, although, uh, I do not have the data to disentangle the two. Yeah, you're right Right and I guess in the so then in the theory, would it be the case that Groupon would always find this change profitable because there's no Sort of negative direct effect on its consumers Oh, if you remember Groupon actually gets hurt in terms of the theory Uh, in the theory. Oh in the theory actually the hypothesis that Uh, the uh group on actually get hurt because there were fewer exclusive consumers on Groupon So overall effect on Groupon is negative Uh That's a good question. I think the We we actually I don't think we have derived hypothesis related to Groupon's profitability Uh, the reason is that the policies are derived mainly to use to test empirically But empirical evidence on Groupon's profitability can is limited because in our data if you remember The the data about the quality of deals after the policy essentially is blurred, right? So we only know over a hundred or over a thousand would not know the exact number of Groupon deals So so that prevent us from using the data to back out Groupon's profitability change So that's one limitation of the because um, it uh Groupon no longer displayed the accurate information of their quality So we can never actually do this type of analysis on Groupon's profitability Right, but it'd be interesting if there is a trade-off for Groupon even without direct effect on consumers That would be kind of interesting Right, exactly. Yep. So what we do know from the theory is that Groupon Uh, have fewer exclusive consumers and because leaving social copy fewer and search for more Groupon actually copied more and search less Because they're less worried that uh, they will be over that between leaving social group Okay, if there are no more questions. So I'm going to um End the formal part of the discussion if people want to stay on and talk to Hui and and Bang further you're welcome to do so. I'm going to put uh, Alexander to corny on As the host for that part, but I'm just going to unmute everyone and you can give a round of applause to Um, we thank you very much Thank you mark. Thank you I probably want to mute yourself Unless you're going to say something That's a lot everyone. Thanks everyone for this