Jinpu Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Jinpu Li has a total of 9 co-authors. Here are the top co-authors by number of shared publications:
- Feng Dong
School of Economics and Man- agement, Tsinghua University
7 shared publications Yi Dong
4 shared publications- Haoning Sun3 shared publications
- Jiahong Gao
Wenlan School of Business, Zhongnan University of Economics and Law, China
2 shared publications Ronghua Sui
School of Business Administration, Chongqing Technology and Business University
2 shared publicationsZhenyang Xu
2 shared publications
Latest Publications
- Open Access
- 13 October 2026
- Published by Elsevier in SSRN Electronic Journal
The Supply of Patience: Population Growth and Bubble Capacity
Population growth is slowing, reducing the young cohorts that purchase long-lived claims from older owners. We show that this demographic shift can substantially reduce the capacity of an economy to sustain rational bubbles when household patience responds endogenously to asset returns. Slower cohort growth lowers the return anticipated by young households, shifts consumption toward youth, and reduces saving. As a result, bubble capacity falls relative to output whenever saving rises with the anticipated return, in contrast to the fixeddiscount benchmark where this ratio is unchanged. When the saving response is sufficiently strong, bubble capacity also falls per young household. The same endogenous-patience mechanism weakens the crowding-out of capital by bubbles and makes current bubble prices respond to demographic news. It also tightens the rollover capacity of public debt and raises the confidence required to sustain a risky bubble, linking demographic change to both fiscal space and bubble survival.
- 13 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessThe Supply of Patience: Population Growth and Bubble Capacity
Population growth is slowing, reducing the young cohorts that purchase long-lived claims from older owners. We show that this demographic shift can substantially reduce the capacity of an economy to sustain rational bubbles when household patience responds endogenously to asset returns. Slower cohort growth lowers the return anticipated by young households, shifts consumption toward youth, and reduces saving. As a result, bubble capacity falls relative to output whenever saving rises with the anticipated return, in contrast to the fixeddiscount benchmark where this ratio is unchanged. When the saving response is sufficiently strong, bubble capacity also falls per young household. The same endogenous-patience mechanism weakens the crowding-out of capital by bubbles and makes current bubble prices respond to demographic news. It also tightens the rollover capacity of public debt and raises the confidence required to sustain a risky bubble, linking demographic change to both fiscal space and bubble survival.
- Open Access
- 11 October 2026
- Published by Elsevier in SSRN Electronic Journal
Runnable Bubbles
A rational bubble can make banks less prone to runs at moderate exposure but more fragile once their holdings become large, even if the bubble’s continuation value is fully preserved. We study an overlapping-generations economy in which future buyers sustain the bubble’s value, while a system-wide run must be financed by cash already in the market. At moderate holdings, the bubble replaces capital that would be sold at a discount and raises liquidation coverage, the cash available in a run net of demandable debt. Once outside cash is exhausted, however, larger bubbly positions generate no additional sale proceeds and displace recoverable capital. Liquidation coverage is therefore hump-shaped in bank holdings, so sufficiently large positions restore the run equilibrium. A bubble that corrects dynamic inefficiency can thus make a run-proof bank runnable. With recurring bubble creation, a growing bubble can pass through a finite run-proof window. The same reversal arises when bubbly collateral expands credit and investment. Limiting bank holdings to run-state market capacity and expanding credible fiscal support both widen the run-proof region.
- 11 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessRunnable Bubbles
A rational bubble can make banks less prone to runs at moderate exposure but more fragile once their holdings become large, even if the bubble’s continuation value is fully preserved. We study an overlapping-generations economy in which future buyers sustain the bubble’s value, while a system-wide run must be financed by cash already in the market. At moderate holdings, the bubble replaces capital that would be sold at a discount and raises liquidation coverage, the cash available in a run net of demandable debt. Once outside cash is exhausted, however, larger bubbly positions generate no additional sale proceeds and displace recoverable capital. Liquidation coverage is therefore hump-shaped in bank holdings, so sufficiently large positions restore the run equilibrium. A bubble that corrects dynamic inefficiency can thus make a run-proof bank runnable. With recurring bubble creation, a growing bubble can pass through a finite run-proof window. The same reversal arises when bubbly collateral expands credit and investment. Limiting bank holdings to run-state market capacity and expanding credible fiscal support both widen the run-proof region.
- Open Access
- 9 October 2026
- Published by Elsevier in SSRN Electronic Journal
Lemon Data Ecosystem
Developers of large language models license human-written content to train their models, while users increasingly produce content with these models. In a dynamic model of the market for training data, a better model attracts more users, increasing the value of better performance but also raising the share of generated content in future data sources. Much of this content echoes what the model already knows and concentrates on common tasks, reducing both the information and the task diversity contained in the data. We show that information dilution alone causes the value of a data source to decline gradually, whereas the loss of task coverage can produce a market freeze. Once the generated share crosses a threshold, developers stop buying from the data pool even at a zero price, possibly while most records are human, while verified data become relatively more valuable. Successful AI can therefore reverse the usual positive data network effect because greater adoption can reduce the value of the data available to train future models. Market freezes are temporary as knowledge depreciates and adoption falls, but strong feedback between model performance and data quality can destabilize the steady state and generate recurring booms and freezes. Certifying informational provenance, meaning whether a record contains observations beyond existing model knowledge, eliminates the freeze. By contrast, information about who or what produced a record provides only an imperfect signal of its informational value. Information about record origin and task coverage can each restore trade, and the two can be complements.
- 9 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessLemon Data Ecosystem
Developers of large language models license human-written content to train their models, while users increasingly produce content with these models. In a dynamic model of the market for training data, a better model attracts more users, increasing the value of better performance but also raising the share of generated content in future data sources. Much of this content echoes what the model already knows and concentrates on common tasks, reducing both the information and the task diversity contained in the data. We show that information dilution alone causes the value of a data source to decline gradually, whereas the loss of task coverage can produce a market freeze. Once the generated share crosses a threshold, developers stop buying from the data pool even at a zero price, possibly while most records are human, while verified data become relatively more valuable. Successful AI can therefore reverse the usual positive data network effect because greater adoption can reduce the value of the data available to train future models. Market freezes are temporary as knowledge depreciates and adoption falls, but strong feedback between model performance and data quality can destabilize the steady state and generate recurring booms and freezes. Certifying informational provenance, meaning whether a record contains observations beyond existing model knowledge, eliminates the freeze. By contrast, information about who or what produced a record provides only an imperfect signal of its informational value. Information about record origin and task coverage can each restore trade, and the two can be complements.
- Open Access
- 9 October 2026
- Published by Elsevier in SSRN Electronic Journal
Runnable Bubbles
A rational bubble can make banks less prone to runs at moderate exposure but more fragile once their holdings become large, even if the bubble's continuation value is fully preserved. We study an overlapping-generations economy in which future buyers sustain the bubble's value, while a system-wide run must be financed by cash already in the market. At moderate holdings, the bubble replaces capital that would be sold at a discount and raises liquidation coverage, the cash available in a run net of demandable debt. Once outside cash is exhausted, however, larger bubbly positions generate no additional sale proceeds and displace recoverable capital. Liquidation coverage is therefore hump-shaped in bank holdings, so sufficiently large positions restore the run equilibrium. A bubble that corrects dynamic inefficiency can thus make a run-proof bank runnable. With recurring bubble creation, a growing bubble can pass through a finite run-proof window. The same reversal arises when bubbly collateral expands credit and investment. Limiting bank holdings to run-state market capacity and expanding credible fiscal support both widen the run-proof region.
- 9 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessRunnable Bubbles
A rational bubble can make banks less prone to runs at moderate exposure but more fragile once their holdings become large, even if the bubble's continuation value is fully preserved. We study an overlapping-generations economy in which future buyers sustain the bubble's value, while a system-wide run must be financed by cash already in the market. At moderate holdings, the bubble replaces capital that would be sold at a discount and raises liquidation coverage, the cash available in a run net of demandable debt. Once outside cash is exhausted, however, larger bubbly positions generate no additional sale proceeds and displace recoverable capital. Liquidation coverage is therefore hump-shaped in bank holdings, so sufficiently large positions restore the run equilibrium. A bubble that corrects dynamic inefficiency can thus make a run-proof bank runnable. With recurring bubble creation, a growing bubble can pass through a finite run-proof window. The same reversal arises when bubbly collateral expands credit and investment. Limiting bank holdings to run-state market capacity and expanding credible fiscal support both widen the run-proof region.
- Open Access
- 27 August 2026
- Published by Elsevier in SSRN Electronic Journal
Networked Dollarization: Stablecoin Adoption and Cross-Border Reallocation
Dollar stablecoins combine a dollar claim with liquidity supplied by a payment network. User adoption and merchant acceptance reinforce each other, creating an adoption threshold that switches an offshore dollar balance sheet on and off. Crossing the threshold reallocates developing-country saving from domestic capital and local money toward Treasury-backed dollar claims, shrinking the local monetary-revenue base and expanding dollar-side rents under the maintained intact-peg pass-through rule. Greater peg risk can eliminate the high-adoption equilibrium, while local inflation lowers the network strength needed to sustain it. Better domestic payment infrastructure raises the coordination hurdle for widespread stablecoin use.
- 27 August 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessNetworked Dollarization: Stablecoin Adoption and Cross-Border Reallocation
Dollar stablecoins combine a dollar claim with liquidity supplied by a payment network. User adoption and merchant acceptance reinforce each other, creating an adoption threshold that switches an offshore dollar balance sheet on and off. Crossing the threshold reallocates developing-country saving from domestic capital and local money toward Treasury-backed dollar claims, shrinking the local monetary-revenue base and expanding dollar-side rents under the maintained intact-peg pass-through rule. Greater peg risk can eliminate the high-adoption equilibrium, while local inflation lowers the network strength needed to sustain it. Better domestic payment infrastructure raises the coordination hurdle for widespread stablecoin use.
- Open Access
- 27 August 2026
- Published by Elsevier in SSRN Electronic Journal
Big Push Toward Big Data
Economic activity generates data, those data improve forecasts, and better forecasts raise activity. Individual firms' data-processing decisions can also respond to economic activity and the behavior of other firms. This paper studies when these feedbacks become a macroeconomic data flywheel. Firms learn about persistent demand from raw data generated by economy-wide activity or by their own operations, and they choose how effectively to convert those data into usable signals. We show that multiple steady states can exist and that the economy can become stuck in a low-production, low-data trap. A Big Push policy supplies public signals at a labor cost and has state-dependent crowd-in and crowd-out effects. Provision crowds in private activity when endogenous information is scarce but crowds it out once information is abundant. A permanent push can eliminate the low-data state while drawing productive labor away from the surviving high-data state. A temporary push of sufficient strength and duration can lead the economy out of the low-data trap.
- 27 August 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessBig Push Toward Big Data
Economic activity generates data, those data improve forecasts, and better forecasts raise activity. Individual firms' data-processing decisions can also respond to economic activity and the behavior of other firms. This paper studies when these feedbacks become a macroeconomic data flywheel. Firms learn about persistent demand from raw data generated by economy-wide activity or by their own operations, and they choose how effectively to convert those data into usable signals. We show that multiple steady states can exist and that the economy can become stuck in a low-production, low-data trap. A Big Push policy supplies public signals at a labor cost and has state-dependent crowd-in and crowd-out effects. Provision crowds in private activity when endogenous information is scarce but crowds it out once information is abundant. A permanent push can eliminate the low-data state while drawing productive labor away from the surviving high-data state. A temporary push of sufficient strength and duration can lead the economy out of the low-data trap.
- Open Access
- 10 March 2026
- Published by Elsevier in SSRN Electronic Journal
The Data Trap: When AI Fails
We model the co-evolution of AI and the economy using a task-based dynamic framework where data quality is an endogenous variable. In our model, next-generation AI capability is a function of training data quantity and quality. The data pool mixes real data from economic activity with synthetic data from incumbent AI. The automation process itself also generates partially synthetic data, creating a data-AI feedback loop. Our analysis shows this mechanism can lead to two traps from different perspectives. First, a development trap may arise in the steadystate. An over-reliance on synthetic data can degrade quality, trapping the economy into a long-run equilibrium with lower automation and output. Second, the system can fall into a dynamic trap of persistent boom-bust cycles. Instability in the feedback loop can prevent convergence, trapping the economy in oscillations that may evolve through period-doubling bifurcations into chaos. We then examine policy responses. A countercyclical levy on synthetic-data use, indexed to model performance or output, stabilizes endogenous cycles without depressing long-run levels. Caps on the synthetic share work in the same direction, while AI adoption caps may be counterproductive by reducing real-data generation. A complementary source-side policy is public investment in data infrastructure that expands the supply of high-quality real data, albeit subject to a crowding-out trade-off in production labor. These results highlight the importance of protecting the information environment and sustaining high-quality real data.
- 10 March 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessThe Data Trap: When AI Fails
We model the co-evolution of AI and the economy using a task-based dynamic framework where data quality is an endogenous variable. In our model, next-generation AI capability is a function of training data quantity and quality. The data pool mixes real data from economic activity with synthetic data from incumbent AI. The automation process itself also generates partially synthetic data, creating a data-AI feedback loop. Our analysis shows this mechanism can lead to two traps from different perspectives. First, a development trap may arise in the steadystate. An over-reliance on synthetic data can degrade quality, trapping the economy into a long-run equilibrium with lower automation and output. Second, the system can fall into a dynamic trap of persistent boom-bust cycles. Instability in the feedback loop can prevent convergence, trapping the economy in oscillations that may evolve through period-doubling bifurcations into chaos. We then examine policy responses. A countercyclical levy on synthetic-data use, indexed to model performance or output, stabilizes endogenous cycles without depressing long-run levels. Caps on the synthetic share work in the same direction, while AI adoption caps may be counterproductive by reducing real-data generation. A complementary source-side policy is public investment in data infrastructure that expands the supply of high-quality real data, albeit subject to a crowding-out trade-off in production labor. These results highlight the importance of protecting the information environment and sustaining high-quality real data.
- 13 November 2025
- Published by Emerald Publishing in Industrial Management & Data Systems
- Vol. 126 (9) , 1-30
Optimal data pricing decisions of competitive two-sided platforms with heterogeneous data costs
Purpose: In the digital economy, two-sided platforms increasingly operate as data intermediaries, establishing vital revenue streams and competitive differentiation. This study aims to explore how competing two-sided platforms, acting as data intermediaries, strategically optimize their collection and pricing of user-generated data under cost heterogeneity and varying affiliation behaviors. Design/methodology/approach: This study uses the Hotelling model to construct a multi-stage game framework, analyzing data collection and the related pricing decisions for competitive platforms. Through equilibrium analysis and practical validation, we reveal how duopolistic platforms strategically select full/partial data collection regimes while optimizing pricing, explicitly incorporating per-unit collection costs and cross-sided network effects. Findings: A significant cost advantage prompts differentiated collection strategies, whereas smaller cost gaps favor consistent partial collection by both. Under partial collection, the cost-disadvantaged platform gains disproportionately from increased data intake, driving more aggressive price adjustments. Rising marginal value of data usage can lead platforms to lower price to expand their user base and leverage higher per-unit data utility for greater profitability. Consumer affiliation predominantly shifts optimal pricing, while provider affiliation significantly alters collection decision. Originality/value: This study innovatively analyzes platform pricing from the perspective of data intermediation. Departing from the most existing literature focused on product/service intermediation, this research advances the understanding of platform pricing by examining data's unique role as a strategic asset. Formalizing non-rivalrous traits and near-zero re-production costs, this study reveals platforms' strategic trade-offs between differentiated/consistent data strategies, offering insights for leveraging data assets to enhance competitiveness and support sustainable platform growth.
- 13 November 2025
- Published by Emerald Publishing in Industrial Management & Data Systems
- Vol. 126 (9) , 1-30
Optimal data pricing decisions of competitive two-sided platforms with heterogeneous data costs
Purpose: In the digital economy, two-sided platforms increasingly operate as data intermediaries, establishing vital revenue streams and competitive differentiation. This study aims to explore how competing two-sided platforms, acting as data intermediaries, strategically optimize their collection and pricing of user-generated data under cost heterogeneity and varying affiliation behaviors. Design/methodology/approach: This study uses the Hotelling model to construct a multi-stage game framework, analyzing data collection and the related pricing decisions for competitive platforms. Through equilibrium analysis and practical validation, we reveal how duopolistic platforms strategically select full/partial data collection regimes while optimizing pricing, explicitly incorporating per-unit collection costs and cross-sided network effects. Findings: A significant cost advantage prompts differentiated collection strategies, whereas smaller cost gaps favor consistent partial collection by both. Under partial collection, the cost-disadvantaged platform gains disproportionately from increased data intake, driving more aggressive price adjustments. Rising marginal value of data usage can lead platforms to lower price to expand their user base and leverage higher per-unit data utility for greater profitability. Consumer affiliation predominantly shifts optimal pricing, while provider affiliation significantly alters collection decision. Originality/value: This study innovatively analyzes platform pricing from the perspective of data intermediation. Departing from the most existing literature focused on product/service intermediation, this research advances the understanding of platform pricing by examining data's unique role as a strategic asset. Formalizing non-rivalrous traits and near-zero re-production costs, this study reveals platforms' strategic trade-offs between differentiated/consistent data strategies, offering insights for leveraging data assets to enhance competitiveness and support sustainable platform growth.
- Open Access
- 1 January 2024
- Published by Elsevier in SSRN Electronic Journal
Bubbly Data
Data are emerging as an important asset with which firms can apply for credit. We develop a dynamic general equilibrium model with a vertical production chain to characterize the role of rational bubbles in the data economy. On the investment side, we characterize nonrivalry in the process of data capital accumulation; on the financing side, we highlight the contextual or strong asset-specific nature of data capital. Moreover, large variance in depreciation rates for different data assets is also considered. Our analysis shows that both the labor reallocation effect and the investment efficiency effect that occur in the bubbly equilibrium affect data capital accumulation. Bubbles that emerge in the data product sector can provide additional liquidity to loosen firms' financing constraints, but in an environment where data investment is highly nonrivalrous, bubbles can be welfare-reducing. Bubbles are more likely to emerge when data depreciation rates or asset specificity are high. Furthermore, our dynamic analysis implies that there could be a data-driven cycle along the transitional path when a nonrivalry in data investment exists.
- 1 January 2024
- Published by Elsevier in SSRN Electronic Journal
Open AccessBubbly Data
Data are emerging as an important asset with which firms can apply for credit. We develop a dynamic general equilibrium model with a vertical production chain to characterize the role of rational bubbles in the data economy. On the investment side, we characterize nonrivalry in the process of data capital accumulation; on the financing side, we highlight the contextual or strong asset-specific nature of data capital. Moreover, large variance in depreciation rates for different data assets is also considered. Our analysis shows that both the labor reallocation effect and the investment efficiency effect that occur in the bubbly equilibrium affect data capital accumulation. Bubbles that emerge in the data product sector can provide additional liquidity to loosen firms' financing constraints, but in an environment where data investment is highly nonrivalrous, bubbles can be welfare-reducing. Bubbles are more likely to emerge when data depreciation rates or asset specificity are high. Furthermore, our dynamic analysis implies that there could be a data-driven cycle along the transitional path when a nonrivalry in data investment exists.
- Open Access
- 1 January 2023
- Published by Elsevier in SSRN Electronic Journal
Consumer Data Collection and Pricing for Two-sided Market Platforms
In addition to intermediating transactions of goods or services, two-sided platforms also play the role of data intermediaries in the era of the digital economy. This new role brings additional profit to the platforms, and in reality platforms are incentivized to collect and sell consumer data. The main distinction between the sale of data and the sale of services is that data are by-products of consumer behavior and are replicable, implying a difference in the platform's profit model. In a duopoly bilateral market, data collection and sales behaviors of platforms have yet to be characterized. In this paper, we build a two-period model to study data collection and sales of competitive platforms considering multihoming users and service providers. Our main findings are as follows. First, due to differences in costs, platforms may adopt differentiated or consistent data collection strategies in the following three ways: one platform fully collects while the other partially collects; both platforms fully collect; both platforms partially collect. In particular, when both platforms partially collect data, the cost-disadvantaged platform may invest more than its cost-advantaged counterpart. Second, when both platforms partially collect data, they should set different prices for the data, closely linked to their respective costs. In such scenarios, as data usage revenue increases, it may be more beneficial for both platforms to lower their data pricing. When both platforms fully collect, their data pricing is identical, irrelevant to investment costs. Third, when one platform fully collects data while the other opts for partial collection, the data pricing of the platform with full collection is mainly influenced by its competitor's cost and strategies, irrespective of its own investment cost. For the platform with partial collection, its data pricing is primarily affected by its own costs.
- 1 January 2023
- Published by Elsevier in SSRN Electronic Journal
Open AccessConsumer Data Collection and Pricing for Two-sided Market Platforms
In addition to intermediating transactions of goods or services, two-sided platforms also play the role of data intermediaries in the era of the digital economy. This new role brings additional profit to the platforms, and in reality platforms are incentivized to collect and sell consumer data. The main distinction between the sale of data and the sale of services is that data are by-products of consumer behavior and are replicable, implying a difference in the platform's profit model. In a duopoly bilateral market, data collection and sales behaviors of platforms have yet to be characterized. In this paper, we build a two-period model to study data collection and sales of competitive platforms considering multihoming users and service providers. Our main findings are as follows. First, due to differences in costs, platforms may adopt differentiated or consistent data collection strategies in the following three ways: one platform fully collects while the other partially collects; both platforms fully collect; both platforms partially collect. In particular, when both platforms partially collect data, the cost-disadvantaged platform may invest more than its cost-advantaged counterpart. Second, when both platforms partially collect data, they should set different prices for the data, closely linked to their respective costs. In such scenarios, as data usage revenue increases, it may be more beneficial for both platforms to lower their data pricing. When both platforms fully collect, their data pricing is identical, irrelevant to investment costs. Third, when one platform fully collects data while the other opts for partial collection, the data pricing of the platform with full collection is mainly influenced by its competitor's cost and strategies, irrespective of its own investment cost. For the platform with partial collection, its data pricing is primarily affected by its own costs.