How Users Drive Value in Two-sided Markets - Platform Designs That Matter (Empirical Findings)

5 5.1 Empirical Findings Parameter Estimates In this section, we present the parameter estimates. As aforementioned, we leverage the fact that Groupon o↵ers products and services of di↵erent categories in di↵erent markets, which allows us to estimate di↵erent types of network e↵ects in a quasi multi-platform context. We estimate the consumer-side model (Equation 2) for search and experience goods, respectively, and present the results in Table 6. Similarly, the estimates for the merchant-side model (Equation 5) for search and experience goods are presented in Table 7. For each table, we present the fixed e↵ect estimator and the results based on the DGMM estimator after applying first-di↵erencing and the instruments. In the subsequent section, we interpret the coefficients based on the DGMM results. On the consumer side, the SS 1 is estimated to heterogeneous between expe- rience and search goods. While the SS 1 is positive and significant for experience goods (0.096, p < 0.05), it is insignificant for search goods (-0.034, p>0.10). As hypothesized, because of the higher uncertainty prior to consumption, more existing consumers can shed light on the quality of goods on the platform and attract more future consumers, yielding a positive SS 1 . However, the quality uncertainty for search goods can be reduced through information search (for example, doing online searches or reading consumer reports), the size of existing consumer base is no longer found important for attracting future consumers. The pattern is di↵erent for the CS0 . The e↵ect of merchants on consumers is 35 Table 6: Parameter Estimates for Consumer-Side Model FE Experience Goods Est Lagged consumer base ( 1 ) Merchant base ( 1 ) Deal price (⇢1 ) Rival consumer base (✓1n ) Rival merchant base (✓1m ) Returning merchants Deal duration Week Week-square Sample size R-square Number of instruments Hansen test of overidentification Di↵erence-in-Hansen test for consumer base Di↵erence-in-Hansen test for merchant base Di↵erence-in-Hansen test for price (SE) 0.030** (0.013) 1.297*** (0.029) -0.286*** (0.016) 0.073*** (0.012) -0.013 (0.035) -0.418*** (0.074) 0.102*** (0.010) -0.007*** (0.001) 2.9E-6 (4.4E-5) 5,324 0.723 Est 0.096** 1.439*** -0.656*** 0.075** -0.057 -0.278** 0.142*** -0.008*** 2.6E-5 2 Est (SE) (0.039) (0.072) (0.098) (0.019) (0.050) (0.114) (0.028) (0.001) (5.0E-5) 5,212 110 (101) = 101.88, p = 0.457 2 (50) = 19.81, p = 1.000 2 (9) = 2.89, p = 0.968 2 (51) = 14.58, p = 1.000 FE Search Goods Lagged consumer base ( 1 ) Merchant base ( 1 ) Deal price (⇢1 ) Rival consumer base (✓1n ) Rival merchant base (✓1m ) Returning merchants Duration Week Week-square Sample size R-square Number of instruments Hansen test of overidentification Di↵erence-in-Hansen test for consumer base Di↵erence-in-Hansen test for merchant base Di↵erence-in-Hansen test for price DGMM DGMM (SE) -0.018 (0.012) 2.012*** (0.040) -0.383*** (0.019) -0.005 (0.023) -0.118* (0.067) -0.533*** (0.074) -0.036*** (0.013) -0.013*** (0.001) -6.6E-5 (8.3E-5) 4175 0.554 Est -0.034 1.970*** -0.448*** -0.032 -0.180 -0.325*** 0.043 -0.014** -1.8E-5 (SE) (0.036) (0.128) (0.100) (0.047) (0.137) (0.113) (0.029) (0.005) (2.1E-4) 3,801 110 (101) = 94.20, p = 0.671 2 (50) = 27.66, p = 0.996 2 (9) = 5.50, p = 0.789 2 (51) = 28.51, p = 0.995 2 FE is the fixed-e↵ect estimator and DGMM is the first-di↵erence GMM estimator. *** p<0.01, ** p<0.05 36 Table 7: Parameter Estimates for Merchant-Side Model FE Experience Goods Est Lagged merchant base ( 2 ) Consumer base ( 2 ) Deal price (⇢2 ) Rival consumer base (✓2n ) Rival merchant base (✓2m ) Returning merchants Week Week-square Sample size R-square Number of instruments Hansen test of overidentification Di↵erence-in-Hansen test for consumer base Di↵erence-in-Hansen test for merchant base Di↵erence-in-Hansen test for price SE 0.190*** (0.016) 0.276*** (0.005) 0.107*** (0.008) 0.013** (0.006) 0.151*** (0.017) 0.155** (0.035) 0.003*** (0.000) -6.0E-5*** (2.1E-5) 5,324 0.8773 Est 0.417*** 0.423*** 0.411*** -0.023** 0.131*** 0.173** 0.002** 6.6E-5** 2 Est SE (0.053) (0.018) (0.050) (0.011) (0.028) (0.067) (0.001) (2.7E-5) 5,242 111 (103) = 106.05, p = 0.399 2 (10) = 5.64, p = 0.844 2 (50) = 17.32, p = 1.000 2 (51) = 12.35, p = 1.000 FE Search Goods Lagged merchant base ( 2 ) Consumer base ( 2 ) Deal price (⇢2 ) Rival consumer base (✓2n ) Rival merchant base (✓2m ) Returning merchants Week Week-square Sample size R-square Number of instruments Hansen test Di↵erence-in-Hansen test for consumer base Di↵erence-in-Hansen test for merchant base Di↵erence-in-Hansen test for price DGMM DGMM SE 0.024 (0.017) 0.276*** (0.005) 0.153*** (0.008) 0.024** (0.010) 0.103*** (0.028) 0.194** (0.031) 0.012*** (0.001) 1.9E-4*** (348E-5) 4,175 0.834 Est 0.242*** 0.380*** 0.196*** -0.002 0.117** 0.170*** 0.009*** -5.1E-5 SE (0.057) (0.023) (0.052) (0.018) (0.050) (0.051) (0.002) (8.7E-5) 3,801 111 (103) = 97.56, p = 0.633 2 (10) = 0.31, p = 1.000 2 (50) = 31.75, p = 0.979 2 (51) = 25.00, p = 0.999 2 FE is the fixed-e↵ect estimator and DGMM is the first-di↵erence GMM estimator. *** p<0.01, ** p<0.05, * p<0.1 37 found to be positive and significant for both categories. The CS0 for experience goods is estimated to be 1.439 (p<0.01), and the CS0 for search goods is estimated to be even 37% higher than the former (1.970, p<0.01). On the merchant side, the SS 1 is estimated to be positive and significant for experience goods (0.417, p<0.01) and search goods (0.242, p<0.01). The CS0 of consumers on merchants is also estimated to exist for both product markets. The CS0 for experience goods is estimated to be 0.423 (p<0.01), and the CS0 for search goods is estimated to be 0.380 (p<0.01). In sum, on the merchant side, the SS 1 for experience goods is found to be approximately 1.7 as large as that for search goods, while the CS0 is qualitatively similar between these two categories. This is again consistent with our hypothesized e↵ects. In the experience-goods category, merchants benefit from the UGC that existing consumers have contributed, because it can help reduce quality uncertainty and create demand. Thus, the SS 1 of the merchants is found to be larger for merchants of experience goods. In contrast, merchants o↵ering search goods do not benefit as much from existing UGC, and the SS 1 in the search-goods category is significant but weaker. Next, we present the results for control variables. Not surprisingly, price is estimated to have a negative e↵ect on consumer base and a positive e↵ect on the merchant base. For experience goods, the consumer price elasticity is estimated to be -0.656 (p<0.01), and the merchant price elasticity is estimated to be 0.411 (p<0.01). For search goods, the consumer price elasticity is estimated to be -0.448 (p<0.01), the merchant price elasticity is estimated to be 0.196 (p<0.01). 38 Furthermore, after controlling for SS 1 and CS0 , the number of returning mer- chant is found to be negatively related with consumer base for experience goods (-0.278, p < 0.05) and search goods (-0.325, p < 0.01), indicating that consumers overall respond more positively to new merchants. This finding is consistent with the “switching risk reduction” hypothesis in (Li et al., 2011): because user reviews are more helpful for informing consumers for experience goods, consumers find it less risky to try new products rather than sticking with known products, resulting in a negative association between returning merchants and the consumer base. In other words, consumers respond more positively to new merchants than to existing merchants. 5.2 Robustness Analyses To verify the validity of our results, it is critical to assess the extent to which the assumptions are met for the DGMM estimator. The first important assumption is that the instrumental variables used for identification should be exogenous to the errors. We present the results from the Hansen test of overidentification restrictions. On the consumer side, the Hansen test statistics are and 2 2 (101) = 101.88 (p = 0.457) (101) = 94.20 (p = 0.671) for experience and search goods, respectively. On the merchant side, the test statistics are goods and 2 2 (103) = 106.05 (p = 0.399) for experience (103) = 97.56 (p = 0.633) for search goods. These results indicate that the null hypothesis of that all IVs are exogenous should not be rejected at 0.05 level. Furthermore, Tables 6 and 7 include the Di↵erence-in-Hansen test for each endogenous variable in our models. The results indicate that the exogenous IV 39 assumption is not rejected for all endogenous variables. Table 8: Results of First-stage Regression Consumer Side Experience Goods Search Goods Endogeneous Variable R2 Partial R2 F Statistic p Consumer Base Merchant Base Consumer Base Merchant Base 0.808 0.889 0.413 0.539 0.016 0.012 0.026 0.010 4.723 24.801 6.074 15.189 p<0.001 p<0.001 p<0.001 p<0.001 R2 Partial R2 F Statistic p 0.810 0.456 0.911 0.527 0.014 0.002 0.013 0.026 17.277 2.190 3.983 6.590 p<0.001 p<0.001 p=0.053 p<0.001 Merchant Side Experience Goods Search Goods Consumer Base Merchant Base Consumer Base Merchant Base Next, we present the first-stage diagnostic statistics for the instrumental variables (see Table 8). To examine the strength of the instruments, the first-stage model was fit regressing the endogenous consumer base and merchant base on its respective instruments, controlling for all the exogenous variables including the market fixed e↵ects. We begin by the consumer side model. For experience goods, the first-stage R-square was higher than 0.80 for both endogenous variables, which was perhaps due to the market fixed e↵ects and week e↵ect. After removing the e↵ect due to the exogenous variables, market and time e↵ects, the first-stage F statistic was 4.723 (p < 0.001) for the consumer base and the F statistic was 24.801 (p < 0.001) for the merchant base. For search goods, the overall R-square was lower, but the first-stage test statistics were still significant (F = 6.074, p < 0.001 for consumer base and F = 15.189, p < 0.001 for merchant base). Although the first-stage F statistics were modest for the endogenous consumer base, we would like to emphasize that the IV estimates were substantially di↵erent across the two types of products: the SS 40 1 was positive and significant (0.096, p < 0.05) for the experience goods and insignificant for search goods (-0.034, p > 0.010). Thus, the heterogeneous SS 1 between expe- rience and search products is not driven by the di↵erence in the explanatory power of the IVs. The first-stage statistics are qualitatively similar for the merchant size model. The F statistics are modest for both endogenous variables but the p-values were all smaller than 0.1, even after controlling for all the exogenous variables, the time e↵ect, and the extensive set of market fixed e↵ects. We conduct another robustness analysis to validate our estimates of the network e↵ects. Specifically, we replace the focal market’s SS 1 and CS0 e↵ects with the non- local counterparts. For the new SS 1 , we add up the customer size and merchant size (for Equations 2 and 5 respectively) in the previous period across all the markets on Groupon other than the focal market. The CS0 variables are constructed similarly. Because these variables reflect the (lagged) customer size and merchant size of the platform in other markets, they capture the network e↵ects due to non-local influences (if any). The parameters are presented in Tables 18 and 19. From the results, we see that the non-local network e↵ects are all estimated to be close to zero and weaker than the local network e↵ects, indicating that consumers and merchants on Groupon are attracted bySS 1 and CS0 within the same local market. This analysis also provides additional validation for the IV estimates of our main models. The non-local variables should be uncorrelated with the local market political advertising and weather conditions, and thus our instruments would not provide the needed variation for identification. As expected, the corresponding coefficient estimates are much reduced, providing validation evidence for using the instrumental 41 variables to identify the local network e↵ects. 5.3 The Analysis of A User Growth Strategy In this section, we calculate CLV 2 and M LV 2 based on the parameter estimates from the foregoing models and examine the e↵ectiveness of the user growth strategy. 5.3.1 The Customer Lifetime Value in Two-sided Markets To calculate CLV 2 and M LV 2, we first transform the unit free elasticity coefficients from Table 6 and 7 to the unit denominated marginal coefficients needed for marginal analysis in Table 9. We then use the calculation of CLV 2 as an example for demonTable 9: Marginal E↵ects Of Independent Variables Marginal E↵ects Experience Goods Search Goods ˜1 0.096 0 ˜2 ˜1 ˜2 0.416 320.7 0.002 0.242 365.4 0.002 ⇢˜1 ⇢˜2 -32.7 0.092 -3.455 0.008 stration. Since the marginal cost of serving an additional consumer is almost zero for digital platforms, ⇡ n of Groupon equals to its gross profit generated by a consumer in each period. Note that Groupon’s gross profit is not directly observed in the data. Thus, we use its revenue multiplied by the gross margin to approximate it. First, we use half of the gross billing as a proxy for revenue according to Dholakia (2011) 12 . Gross billing is the total sales on Groupon, which is calculated from the transaction volume and deal price. Note that the transaction volume is used as a proxy for the 12 It is also consistent with the ratio of revenue to gross billing in Groupon’s 2012 annual report. Groupon’s gross billings and revenue in North America in 2012 is $2,373.153 million and $1,165.7 million, respectively, yielding a ratio of gross billing to revenue of 49.12%. 42 consumer base. Thus, we use the increased consumer base multiplied by the deal price to calculate the gross billing. Second, in Groupon’s 2012 annual report, the gross profit accounted for less than 70% of Groupon’s revenue. In later years, this percentage decreased to around 50%. We use 70% of Groupon’s revenue to represent the gross profit, which sets an upper bound of CLV 2. According to Equation 7, we calculate CLV 2 in experience-goods and searchgoods markets, respectively. The same logic applies to the M LV 2. Table 10 reports the CLV 2 and M LV 2 in experience-goods and search-goods markets. According to our calculation, a consumer is worth $62 and a merchant is worth $31,654 in the experience-goods market. In the search-goods market, a consumer is worth $54 and a merchant is worth $26,338. Although the CS0 of experience goods and search goods are all positive and significant, the CLV 2 and M LV 2 in the experience goods market are larger than those in the search goods market, as a result of the di↵erences in SS 1 . In the experience-goods market, the SS by 1.15 times, and the SS 1 1 of consumers amplifies the CLV 2 of merchants amplifies the M LV 2 by 1.66 times. In the search-goods market, due to the absence of a significant SS relative weaker SS 1 1 of consumers and a of merchants, the value created on both sides of the market has less opportunity to be amplified over time. The positive shocks that a market may receive in one period quickly dissipate and do not result in persistent value creation. Table 10: Estimated Customer Lifetime Value Customer Lifetime Value CLV 2 M LV 2 Experience Goods($) Search Goods ($) 62 54 31,654 26,338 43 5.3.2 The User Acquisition Cost To boost growth, platforms can consider marketing to increase their consumer base, hoping that the incremental consumer base could lead to future increases in the consumer base and merchant base. In this section, we consider two commonly used marketing strategies: (1) advertising to buy consumers, and (2) implementing price discounts. We then compare the CLV 2 computed in the previous section to the consumer acquisition cost. Note that this analysis can be applied to both consumer side and merchant side. However, platforms’ cost to acquire merchants is typically unavailable to researchers, which is the case for Groupon. In its annual report, Groupon reported neither the expense of acquiring merchants nor the growth of its merchant base. As a result, our cost-and-benefit analysis focuses on the consumer side. Intervention – Acquisition Through Marketing. We first look at the scenario when the platform uses advertising marketing as the user-growth strategy. The consumer acquisition cost (CAC) in this case is calculated as the total marketing expense divided by the total number of acquired consumers (Gupta et al., 2004; McCarthy and Fader, 2018). In its 2012 annual report, Groupon reported a total marketing expense of $336.85 million with 7.3 million consumers acquired worldwide, yielding an acquisition cost of $46 per new consumer globally. In North America, the marketing expense was $105.9 million, but the report did not include the total number of consumers acquired in North America. Based on the proportion of gross billing in North America ($2.37 billion) relative to worldwide sales ($5.38 billion), we esti- 44 mated the number of new consumers in North America as 3.22 million, which yields an estimated cost of $33 to acquire a new consumer. Table 11 reports Groupon’s net CLV 2 in North America, which is equal to CLV 2 minus CAC. It represents a net CLV 2 of $29 in the experience-goods market but a net CLV 2 of $21 in the search-goods market, which implies that the consumer acquisition through marketing is less e↵ective in the search-goods market than the experience goods market due to di↵erent CLV 2 in these markets. Note that fixed costs and operating costs associated with serving consumers are not included in this calculation. If we amortize these costs to each consumer, the net CLV 2 will become negative. Table 11: Cost-and-Benefit Analysis of Marketing Acquisition Experience Goods Search Goods 62 33 29 54 33 21 Consumer Lifetime Value ($) Consumer Acquisition Cost ($) Net Consumer Lifetime Value ($) Intervention – Price Cut. In a two-sided market, a price cut can stimulate greater participation on the discounted side and the other side of the market through various network e↵ects. We analyze the cost and benefit of a price cut on each side of the market. We use the formula of the marginal e↵ect of the deal price in Section 5.2.1 to calculate the increase in the consumer base and merchant base for the promotion on each side, which is then multiplied by the CLV 2 and M LV 2 to obtain the total increased value. Table 12 illustrates these results. For experience goods, a $1 cut in the deal price brings 33.2 additional consumers and 0.063 additional merchants, yielding a 45 total increased value of $4,035. Based on the average transaction volume of each market, the promotion cost is calculated to be $3,959. Thus, the net gain from deal promotion is $ 76. For search goods, a $1 cut in the deal price brings 4.03 additional consumers and 0.008 additional merchants, yielding a total increased value of $435. Based on the average transaction volume of each market, the promotion cost is calculated to be $704. Thus, the net gain from deal promotion is $ -269. Due to the di↵erence between the CLV 2 and M LV 2 in the experience-goods and searchgoods markets, the deal promotion for experience goods is cost-e↵ective but the deal promotion for search goods is not cost-e↵ective. Table 12: Cost-and-Benefit Analysis of Price Promotion 6 $1 price cut Experience Goods Search Goods Increase in consumer base Increase in merchant base Total increase in value ($) Promotion cost ($) 33.2 0.063 4,035 3,959 4.03 0.008 435 704 Net gain ($) 76 -269 Managerial Implications: Platform Designs to Enhance User Stickiness For platforms with a strong CS0 but a weak SS 1 , platform design choices that enhance user stickiness may o↵er greater leverage than common marketing strategies that bring users to a platform. In this research, we define user stickiness as the ability of the platform to continually attract users in each period. We first calculate 46 the CLV 2 and M LV 2 in simulations when the user stickiness is enhanced, and then discuss two methods of enhancing user stickiness. One method considers platform design choices such as improved reputation systems that are derived from user network e↵ects. The second method considers the design choices that are not derived from user network e↵ects but from the improved platform quality. For example, the development of a merchant information management system reduces the risk of disintermediation and reroutes transactions through the platform rather than conducting them o↵-platform. The direct implications of this design intervention is that users become stickier on the platform, leading to increases in CLV 2 and M LV 2. Lastly, we discuss the feasibility of these platform design choices as compared to a similar platform. 6.1 Simulating Enhanced User Stickiness We use the CLV 2 and M LV 2 in the experience-goods market as an example to simulate the impact of changing the strength of SS 1 ). We simulate three scenarios of 1: 1 of consumers (i.e., modifying 0.1, 0.5, and 0.9, and compute the corre- sponding updated CLV 2 and M LV 2. Similarly, we simulate the impact of changing the strength of SS When the SS 1 1 of merchants by modifying 2. Table 13 presents the results. of consumers or merchants changes, it has a substantial impact on CLV 2 and M LV 2. Boosting 1 from 0.096 to 0.5 would improve the CLV 2 by 1.8 times and the M LV 2 by 1.4 times. Boosting 1 from 0.096 to 0.9 would improve the CLV 2 by 9 times and the M LV 2 by 5.2 times. Boosting 2 from 0.417 to 0.5 would improve the CLV 2 by 1.1 times and the M LV 2 by 1.2 times. Boosting 47 1 Table 13: CLV 2 and M LV 2 in Response to Changing the Strength of SS SS 1 of Consumers ( 1 ) Consumer value ($) Merchant value ($) SS 1 of Merchants ( 2 ) Consumer value ($) Merchant value ($) 0.1 0.096(observed) 0.5 0.9 62 31,728 62 31,654 111 45,086 557 165,317 0.1 0.417(observed) 0.5 0.9 51 20,505 62 31,654 67 36,908 211 184,541 1 from 0.417 to 0.9 would improve the CLV 2 by 3.4 times and the M LV 2 by 5.8 times. Note that changing CC Similarly, changing the M M 1 1 has a greater impact on CLV 2 than on M LV 2. has a greater impact on M LV 2 than on CLV 2. It is worth noting that participant acquisition marketing becomes more cost-e↵ective as any inter-temporal same side network e↵ect SS 6.2 1 increases. Discussion of Platform Design Choices The scale of these e↵ects raises the question of whether enhancing user stickiness is feasible. Could managerial decisions concerning platform design have large strategic growth implications? Evidence from another daily deal platform suggests that these outcomes are indeed possible. Meituan launched in 2010, two years after Groupon, as a daily deal platform that used an identical business model which leveraged group buying power to o↵er coupons and volume discounts to price-sensitive consumers. In contrast to Groupon’s strategy that ran several years of losses in an e↵ort to buy user growth, Meituan adopted platform designs that focused on enhancing user 48 Figure 4: Average Units per Consumer: Groupon vs Meituan stickiness and sustaining network e↵ects. We first summarize two types of platform designs that enhance user stickiness, then compare the performance of Groupon and Metiuan. Enhancing User Stickiness Through Network E↵ects. First, Meituan merged with Dianping (a Chinese Yelp) to o↵er more UGC on the platform. As new transactions generated more user-review data, UGC improved recommendations, created positive feedback, and increased user stickiness. Second, Meituan rewarded consumers for generating ideas and for repeat purchases. It then used this data to help merchants identify new products and situate new stores, which, again, increased their dependence on Meituan. By design, it helped users create more value for other users. Enhancing User Stickiness Through Factors Other Than Network E↵ects. Meituan invested heavily in delivery and reservation systems. It expanded its services to handle logistics and deliver not only food to consumers but also supplies to restau49 Figure 5: CLV 2: Groupon vs Meituan ($) rants. Using technology to reduce friction encouraged restaurants and consumers to rely on Meituan, e↵ectively enhancing user stickiness. Comparison of Meituan and Groupon’s Performance. Figure 4 and Figure 5 show the average units per consumer13 and CLV 2 of Groupon and Meituan,14 respectively. We consider the average units per consumer as a direct indicator of user stickiness. Groupon’s average units per consumer gradually decreases while Meituan’s average units per consumer continues to increase. A number of factors may have contributed to the divergent CLV 2 of Meituan and Groupon yet it is worth noting that the average units per consumer of Meituan is 8.3 times that of Groupon in 2020, and CLV 2 of Meituan is 12 times that of Groupon. 13 In Groupon’s and Meituan’s financial reports, the average units per consumer are the number of purchases per consumer during the reporting period. 14 Meituan appeared in stock market listings in 2018, so we can only collect data and compare CLV 2s of Meituan and Groupon from 2018-2020. 50 7 Limitations & Research Opportunities There are several directions for future research. First, platforms have multiple ways to improve their long-term value. Should a platform invest in user growth, the design of network e↵ects, key assets, brand, or other factors? It is worth gathering real data on the comparative costs and benefits of these strategies before the platform makes investment decisions. Second, in our empirical setting, we include the competition e↵ect from the primary competing platform. Although this should capture the competition in many markets, we acknowledge that, in some markets, there could be other major platforms as a key player. Future analysis can improve the competition analysis. Third, our CLV 2 model is not based on accounting profit in each future period. Following Zhang et al. (2012) and Li et al. (2020), we identify users’ impact on future profits. Since we want to estimate the value created by users, our model is suitable for this research. However, our model cannot capture the impact of increased platform quality on future profits, it may underestimate the CLV 2 when platform quality increases during the observation period. It would be valuable to develop a CLV 2 model that can capture the impact of network e↵ects as well as platform quality on future profits. Lastly, the strengths of the inter-temporal e↵ects and within-period e↵ects can di↵er strongly across di↵erent markets. We have shown that they vary by product categories on one platform, and in future research, we aim to identify and compare the inter-temporal e↵ects across multiple platforms. 51 8 Conclusions We study how users drive nework value in two-sided markets. Introducing time, we develop a discrete-choice model to estimate both within and across period network e↵ects, which have been under-studied in empirical research (Jullien et al., 2021). Our model underscores, in particular, the importance of same-side intertemporal attraction. For certain platforms, weak same-side e↵ects SS 1 yield poor user stick- iness and the impacts of strong cross-side e↵ects CS0 may not persist. Based on our estimates, we develop a model to calculate consumer (CLV 2) and merchant (M LV 2) lifetime value in two-sided markets. By comparing CLV 2 and user acquisition costs, a platform can examine the e↵ectiveness of di↵erent user growth strategies. We theorize that the mechanism behind heterogeneous SS 1 is the “product learning” e↵ect: products with higher uncertainty garner a higher value from user contributions (Hagiu and Wright, 2020; Li et al., 2017). Our theoretical simulations and empirical estimates confirm di↵erent levels of SS 1 between search and experience goods. We then discuss platform strategic responses when cross-side e↵ects CS0 are strong but same-side e↵ects SS 1 are weak. Results show that stickiness enhanc- ing platform designs can increase CLV 2 and M LV 2. After users become stickier, user acquisition marketing becomes more cost-e↵ective. Data on customer lifetime value at Meituan demontrates that enhancing user stickiness is feasible and can be designed. Overall, our findings remind managers not to overemphasize user growth when the user stickiness is poor and instead focus on platform designs that increase user to user stickiness. We believe these results o↵er the following three contributions. First, theoretical 52 models of network e↵ects have largely been static. Empirical models of network e↵ects have largely been absent or have struggled to address the reflection problem. Extending these works, our model addresses dynamic network e↵ects, including both within and across period attraction, for two-sided markets. We address the reflection problem by using prior period instruments for both sides as well as current period instruments a↵ecting only one side. Main results are economically and statistically significant. In cases where our instruments are not strong, we contrast within market estimates to demonstrate key results hold. Second, the customer lifetime value literature has omitted network spillovers in CLV calculations (Gupta and Mela, 2008). Our research provides a straightforward set of tools for making these estimates, accounting for how CLV2 and MLV2 a↵ect each other. This can also account for the lifetime value of “free” customers, who attract other sides of the market but do not themselves pay. Of necessity, these calculations are interdependent when network e↵ects are present. Our results show spillovers are, in fact, present confirming the need for such a model. Third, we identify factors that can strengthen or weaken each of the four sets of network e↵ects in our study. As illustration, selling experience goods raises the importance of same-side customer adoption whereas selling search goods does not. By contrast, both matter for merchants’ network attraction. Other salient factors include a competitor’s customer and merchant bases, own and rival prices, and local versus non-local interaction. Equipped with the ability to estimate attraction, retention, and now cross-market spillovers, managers and investors can design marketing and information systems interventions that o↵er greater lifetime value leverage. Be- 53 yond the suggestive examples of Amazon, SAP, Airbnb and others, case analyses of Meituan and Groupon provide helpful guidance. 54