Abstract:  In settings with instrumental variables, the TSLS estimator is the most popular way of summarizing causal evidence. Yet in many settings, the instrument monotonicity assumption needed for its causal interpretation is refuted. A prominent example are designs using the (quasi-)random assignment of defendants to judges as an instrument for incarceration. But ultimately, we may not be interested in the TSLS estimand itself, but rather in the impact of some counterfactual policy intervention (e.g. an encouragement to release more defendants). In this paper, we derive tractable sharp bounds on the impact of such counterfactual policies under reasonable sets of assumptions. We show that for a variety of common policy exercises, the bounds do not depend on whether one imposes instrument monotonicity, and thus one can drop this often-tenuous assumption without loss of information. We explore other restrictions that can help to tighten the bounds, including the policy invariance assumption commonly used in applications of the marginal treatment effects framework and its relaxations. We illustrate the usefulness of this approach in an application involving the quasi-random assignment of prosecutors to defendants in Massachusetts.

Abstract:  We analyze a dynamic model of a monopolistic informed investor who receives private information on an ongoing basis and faces a post-trading disclosure requirement. Our main result is that, under certain conditions, characterizing the equilibrium of the trading game between the investor and a market maker can be reduced into solving the investor's information allocation problem, which is mathematically equivalent to a fictitious consumption-saving model with a borrowing constraint. Hence, insights from the well-established consumption-saving literature, such as the permanent income hypothesis, can be adapted to shed light on the informed investors’ trading strategy and the equilibrium asset prices and market liquidity. What is behind this unexpected mathematical equivalence? We show that if the investor's commitment value is zero, one can reduce the two-player trading game between the investor and a market maker into a single-player information allocation game, which can be turned into a consumption-saving model simply by relabeling.

Abstract:  This paper studies whether working from home (WFH) affects workers’ performance in public sector jobs. Studying public sector initiatives allows us to establish baseline estimates on the impact of WFH in settings where incentives to perform are weak. Exploiting novel administrative data and plausibly exogenous variation in work location, we find that WFH increases productivity by 12%. These productivity gains are primarily driven by reduced distractions. They are not explained by differences in quality, hours worked, absenteeism, characteristics of reported cases, training, administrative duties, or task allocation. Importantly, productivity gains nearly double when tasks are assigned by the supervisor.

Abstract:  

Abstract:  We develop a statistical framework to learn the high-dimensional stochastic discount factor (SDF) from a large set of characteristic-based portfolios. Specifically, we provide statistical support to use the MAXSER method proposed in Ao Li Zheng (2019) to screen for potentially useful factors and develop a statistical inference theory for further factor selection to construct the SDF portfolio. Applying our approach to a large number of characteristic-based portfolios, we find that our SDF estimator performs well in achieving a high Sharpe ratio and explaining the cross-section of expected returns of various portfolios.

Abstract:  This paper investigates how the relationship between observed and counterfactual distributions can serve as an identifying condition for treatment effects when the treatment is endogenous. We show that this condition holds in a broad class of nonparametric treatment effect models. To that end, we first provide a novel characterization of the widely used rank similarity assumption. Our characterization highlights the strength of this assumption and motivates a relaxation that remains economically meaningful, resulting in our proposed identifying condition. It also supports the use of richer exogenous variation in the data (e.g., multi-valued or multiple instruments) in exchange for weaker assumptions. The primary goal is to offer empirical researchers tools that are robust, easy to implement, and capable of yielding informative policy evaluations.

Abstract:  In efficient markets, asset returns show no predictability over short periods, such as daily or intraday intervals, where expected returns are nearly zero. However, the presence of pricing errors---transitory components of asset prices reflecting various market frictions or biases---induces return reversals. These reversals provide evidence against market efficiency, particularly in terms of liquidity provision. We propose to test market efficiency based on the joint inference of the covariances of observed returns over multiple horizons. We show that analyzing a small set of horizons is sufficient to detect inefficiencies caused by a wide range of transitory pricing errors, both theoretically and practically. Extensive simulations highlight the superiority of our multi-horizon approach over traditional tests, especially when pricing errors are weak and exhibit complex serial dependencies. Moreover, our test statistic serves as a natural liquidity measure that can effectively identify financial crisis with significant liquidity drains.

Abstract:  This study uses a randomized controlled trial in rural Thailand to evaluate the impact of intensive and hands-on on-site training for preschool teachers on cognitive and non-cognitive skills. The main finding is that the intervention significantly improved children's cognitive skills, namely mathematics and literacy, by 0.13 and 0.18 standard deviation, respectively. Parents observed a slight improvement in non-cognitive skills, especially externalizing skills, but the results were insignificant. The training significantly improves teaching quality, especially regarding the plan-do-review process and classroom environment. The intervention is more beneficial to children with lower baseline skills, who attended a larger classroom, and who have more books at home.

Abstract:  Creating AI systems that augment human capabilities and promote personal and societal flourishing demands expertise in multiple research fields. My research takes an interdisciplinary, human-centered approach to the development of personal AI systems and the understanding of the complexity of human-AI interaction. More specifically, my research: (1) Creates novel AI prototypes through personalized multi-modal systems that support human flourishing by facilitating learning and enhancing well-being; (2) Examining the science of human augmentation by AI systems through large-scale experimental studies that reveal the effects of AI on human decision-making, sense-making, behavior, beliefs, sense of self, and other critical aspects; (3) Proposes new techniques including novel platforms and tools that others can use to implement human augmentation systems; and (4) Develops novel research methods that combine qualitative and quantitative analyses of human-AI interaction. The goal of my work is to establish a new discipline that focuses on the science of human-AI interaction for human augmentation and empowers AI developers with a more informed understanding of the implications of design choices for AI systems that interact with humans. My hope is to catalyze a new intellectual renaissance and contribute to the advancement of AI that benefits the human experience.

Abstract:  The recent decades have witnessed a remarkable surge in machinery imports from China to African economies. Our study aims to examine the causes and local implications of these imports. Notably, we find that China's domestic decommissioning policies have played a significant role. City industries that face urgent capacity elimination requirements are more inclined to export machinery equipment to emerging markets, particularly African countries with weaker environmental regulations and abundant resources. By using instrumental variables based on China's annual lists of eliminated capacity, we discover mixed effects of machinery imports on African economies. While they contribute to increased output and value-added in recipient industries that use these machines for production, the impact on employment and sectoral linkages is limited. Treated regions experience a shift in employment from agriculture to other sectors, primarily services. Meanwhile, these regions also experience a rise in pollution levels. This serves as a cautionary tale for the role of massive machinery transfer in African industrialization, underscoring the need to pay attention to job creation and sustainable development.

Abstract:  This paper explores a potential "dark side" of patent trade in reinforcing monopolists' market power. We analyze how the patent trade boom affects new innovations and industry dynamics. While easier patent trade encourages more patent creation, these new patents disproportionately benefit existing monopolists, who are more likely to acquire them. Using an endogenous growth model with patent trade, we show that subsidizing patent transactions may shift investors' research focus toward monopolists, further entrench their market power, and ultimately reduce social welfare. An optimal subsidy policy should account for the initial misallocation in the invention market.

Abstract:  Does news media coverage of autocracies hinge on their relation- ship with those regimes? Exploiting a large-scale media crackdown in May 2019 in China, in which multiple influential UK- and US-based news sites were blocked, we find that news outlets adopted a more negative tone in their coverage of China and reported more frequently on sensitive topics such as human rights, after being blocked, compared to those with no access change. Such effects are absent in news on economic topics and opinion articles. We investigate several mechanisms underlying these findings, including reduced self-censorship, diminished journalistic resources, and changes in readership composition after losing access. 

Abstract:  We explore motivated self-control along with imperfect memory as mechanisms to enhance future self-confidence over one’s self-control ability. Through a field experiment involving university students, we conducted two rounds of self-control-dependent tasks. The results show that pre-announced reminders of past performance before the second-round task significantly increase first-round completion compared to sudden ex-post reminders, thereby boosting self-confidence and willingness to participate in the second-round task. The students exhibit optimistic memory biases in the absence of reminders, which also enhance self-confidence and second-round participation, while to a lesser extent. We develop an intra-person multiple-self model in which a present-biased agent chooses to persist in tasks to signal self-control ability to their future selves to resolve time-inconsistency, aligning with our experimental findings.

Abstract:  This paper analyzes the practice of tying ancillary services (e.g., order fulfillment, payment processing, or pre-installed applications) to the core intermediation function offered by dominant digital platforms such as online marketplaces or smartphone operating systems. We study when tying is profitable for platforms, and its effects on competition. Tying induces an inefficiency by forcing the ancillary service on consumers who don't value it very much. But it also reduces sellers' market power by preventing them from using their service adoption decisions to differentiate. The second effect is so strong that the platform prefers to tie the service to increase consumers' value for the platform. Consumers benefit overall from tying. Thus, when we consider policies that ban tying or break-up the platform, the net effect is to harm consumers.

Abstract:  While it is widely recognised that linear (structural) VARs may fail to capture important aspects of economic time series, the use of nonlinear SVARs has to date been almost entirely confined to the modelling of stationary time series, because of a lack of understanding as to how common stochastic trends may be accommodated within nonlinear models. This has unfortunately circumscribed the range of series to which such models can be applied -- and/or required that these series be first transformed to stationarity, a potential source of misspecification -- and prevented the use of long-run identifying restrictions in these models. To address these problems, we develop a flexible class of additively time-separable nonlinear SVARs, which subsume models with threshold-type endogenous regime switching, both of the piecewise linear and smooth transition varieties. We extend the Granger--Johansen representation theorem to this class of models, obtaining conditions that specialise exactly to the usual ones when the model is linear. We further show that, as a corollary, these models are capable of supporting the same kinds of long-run identifying restrictions as are available in linearly cointegrated SVARs.

Abstract:  This article examines the impact of the Fed's balance sheet policies on economic activity,  which has become integral to monetary policy since 2009. I extract the surprise component of ten-year Treasury-note futures around a thirty-minute window bracketing the closure of auctions for ten-year bonds, exploit the composition of Large Scale Asset Purchase (LSAP) programs, and construct shocks to long-term rates that mimic LSAP shocks. Utilizing high-frequency identification and data on this alternative shock series, I estimate that balance sheet expansion by the Fed increases economic activity and reduces credit spreads - when the economy is stable, normalization generates a slowdown in economic activity without disrupting financial conditions – demonstrating the link between balance sheet policies, economic activity, and interest rates, which previous work examining the impact of LSAPs has been unable to show empirically.

Abstract:  This paper develops an empirical Bayes estimator for a panel data model with two-way fixed effects. The hyperparameters that control the variance (degree of shrinkage) and the location of the prior are determined by minimizing an unbiased risk estimate. We established optimality of the proposed estimator by showing that it asymptotically attains the same loss as an oracle estimator with a hyperparameter that is chosen based on the knowledge of the fixed effects. In a Monte Carlo study we show that the proposed estimator outperforms a number of competitors, including the least squares estimator. The method will be applied to the estimation of teacher values-added from a linked student-teacher data set obtained from the North Carolina Education Research Data Center.

Abstract:  A worker interacts with a sequence of clients under a manager’s supervision. I highlight a novel role of this manager’s mediation in addressing the worker’s moral hazard, namely to intertemporally reduce suspensions of the worker’s service that are surplus-depleting but crucially serve as punishments to motivate her costly effort. I show that, to best address moral hazard, the manager at times secretly asks a high-performing worker to scale down her effort against a current client and implements dynamic correlation by telling the worker that current underperformance will not be punished. These occasions are frequent in the short run and eventually disappear.

Abstract:  This paper studies a grand mechanism problem in which an unknown common state is a number of agents. The principal designs both a protocol structure in terms of how the information is disclosed and a collection of sub-mechanisms. The former is a mapping from numbers of agents to probability distributions over disclosure rules, and the latter’s different sub-mechanisms may each contain unique allocation and payment functions depending on the number of agents as well as a disclosure rule. Since information design on the common state affects a feasible set of mechanisms the principal can choose, choosing a protocol structure and such a collection are interwoven. First, we establish the existence of the optimal grand mechanism resolving the problem. We then show that there is a single threshold for the optimal grand mechanism if a sub-mechanism cannot depend on the number of agents. Interestingly, the main result shows that if a sub-mechanism can also depend on it, the optimal grand mechanism is characterized by double thresholds such that the principal does not announce the number if it is in the middle range.

Abstract:  This paper studies semiparametric Bayesian inference for the average treatment effect on the treated (ATT) within the difference-in-differences research design. We propose two new Bayesian methods with frequentist validity. The first one places a standard Gaussian process prior on the conditional mean function of the control group. The second method is a double robust Bayesian procedure that adjusts the prior distribution of the conditional mean function and subsequently corrects the posterior distribution of the resulting ATT. The double robustness implies that the lack of smoothness of conditional mean functions can be compensated by high regularity of the propensity score, and vice versa. We prove semiparametric Bernstein-von Mises (BvM) theorems for both proposals. Monte Carlo simulations and an empirical application demonstrate that the proposed Bayesian DiD methods exhibit strong finite-sample performance compared to existing frequentist methods.

Abstract:  We examine the effect of management practices on the performance of entrepreneurs, micro and small firms. Based on a novel data set from the German Socio-Economic Panel (SOEP) that links individual to firm level data, we construct for the first time a management score that maps management activities among micro and small firms. Our empirical analysis shows a significantly positive relationship between management scores and firm performance measured by firm productivity while other well-established factors that have been shown to impact entrepreneurial activities, are not associated with productivity. Our findings highlight the critical role of effective management in improving the resilience and growth trajectories of micro and small businesses.

Abstract:  We compare the performance of carbon taxes and cap-and-trade schemes in an environmental dynamic stochastic general equilibrium model featuring financial frictions. Our findings reveal a critical impact of financial frictions on the propagation of business cycle shocks under different climate policies. Cap-and-trade schemes significantly reduce the welfare cost of business cycles compared to carbon taxes. The welfare cost gap between the two policies narrows when macroprudential rules that mitigate financial accelerator effects are in place. These rules can go a long way in smoothing business cycle fluctuations and can be optimally designed to simultaneously promote financial stability while aligning and reducing the welfare costs of business cycles under varying climate policies.

Abstract:  Causal Reduced Form (CRF) is an "outcome (Y) representation linear in treatment D", with a causal parameter of interest as the slope of D. CRF is (almost) model-free, and holds for any Y (binary, count, continuous,...). Diverse CRF's appeared for various types of D: binary exogenous, multiple exogenous, binary endogenous, network treatment, mediator, DiD, etc. CRF has three uses. Firstly, it reveals the restrictions embedded in commonly used structural forms. Secondly, being linear in D, it allows estimating the causal parameter with OLS, IVE or GMM. Thirdly, substituting the CRF into an estimator formula shows what the estimator actually estimates.  As it turns out, minor variants of many OLS and IVE controlling D and covariates X are consistent to an "overlap-weighted" average of X-heterogeneous effects of D for any form of Y.

Abstract:  Economists have increasingly recognized the role of genetics in economic outcomes. Epigenetics, which studies changes in gene expression without altering DNA sequences, provides insights into how socioeconomic environments affect human biology and influence economic behaviors and outcomes. This paper explores the application of “epigenetic clocks”, which aggregate epigenetic data to predict biological aging and health risks, in economics research. We propose a novel integrated measure of epigenetic aging, the Multi EpiGenetic Age (MEGA) clock: several epigenetic clocks are combined to reduce measurement error and improve efficiency. Using data from the Avon Longitudinal Study of Parents and Children (ALSPAC), we apply the MEGA clock in two empirical settings: first, we examine the association between the longitudinal exposure to child abuse and epigenetic age acceleration in adolescence; second, we test the association between epigenetic age acceleration and early-adulthood cognitive and socioemotional outcomes. Our findings reveal that (i) exposure to child maltreatment before adolescence is associated with half a year of accelerated epigenetic aging and that (ii) epigenetic aging predicts moderately worse cognitive and socioemotional outcomes in early adulthood. These results highlight the usefulness of epigenetic aging as a metric for understanding the long-term effects of early-life adversity and inform economic policies targeting public health and productivity.

Abstract:  The feasibility of cardinally measuring ‘wellbeing’ through surveys has become a contentious topic in recent debates. Several influential papers argue that treating survey data as purely ordinal and accounting for heterogeneity within response categories can overturn many established findings in the literature. To evaluate this claim, we conduct a systematic replication of the entire body of wellbeing research published in leading economics journals since 2010. Our analysis covers more than 50 studies, comprising approximately 50,000 regression coefficients. We examine whether the signs of these coefficients remain stable under all positive monotonic transformations of the wellbeing scale and when plausible heterogeneity within response categories is introduced. Our findings suggest that the risk of sign reversal is relatively low and is consistently linked to specific aspects of research design. Additionally, we explore the sensitivity of coefficients in terms of significance and magnitude when cardinality assumptions are relaxed. Our results indicate that while coefficients are generally robust, some exhibit considerable sensitivity under these adjustments.

Abstract: We measure the strength of competitive investment in human capital with the size of “good-school premium” in housing prices in China. We find it to be around 18% of the housing value on average.  We conjecture that an important driver for competitive investment in human capital is the pressure from the marriage market. Empirically we find the local relative shortage of brides is a significant predictor of the size of the good-school premium.

Abstract:  We study a mechanism design problem of a monopoly platform that matches content of varying quality, ads with different ad revenues, and consumers with heterogeneous tastes for content quality. The optimal mechanism balances revenue from advertising and revenue from selling access to content: Increasing advertising revenue requires serving content to more consumers, which may reduce access revenue. Contrary to the standard monopolistic screening, the platform may serve content to consumers with negative virtual values while, to reduce information rents, limiting their access to higher-quality content. Then, an increase in ad profitability reduces its incentive to invest in content quality.

Abstract:  We study optimal liability for AI-powered products with semi-autonomous capabilities (e.g., self-driving vehicles). Both AI and human user errors can cause product failures that harm third parties. While AI introduces the extreme risk of large-scale social harm that renders full liability impractical, a well-designed liability rule for ordinary loss can implement efficient levels of autonomy and R&D investment in AI safety under a "balanced" R&D condition. However, full efficiency cannot be achieved when R&D efforts are specifically targeted at mitigating extreme risks. Notably, optimal liabilities for AI and human users differ even when the losses incurred are identical.

Abstract:  Risk averse sellers can justify the common observation that reserve prices appear lower than the theory with a risk-neutral seller would predict. We provide weak conditions to identify the CRRA utility function for the seller and propose an estimator for it. We model the bidders' valuations flexibly without incurring the curse of dimensionality. Our estimator is akin to a semiparametric M-estimator where the nuisance parameters are the quantile function and its derivative. We provide conditions under which our estimator has an asymptotic normal distribution and converges at the parametric rate. We present results in the context of an ascending auction, although the same idea can be applied to study a seller's risk aversion in a first-price auction.

Abstract:  We study the problem of distribution-free dependence detection and modeling through the new framework of binary expansion statistics (BEStat). The binary expansion testing (BET) avoids the problem of non-uniform consistency and improves upon a wide class of commonly used methods (a) by achieving the minimax rate in sample size requirement for reliable power and (b) by providing clear interpretations of global relationships upon rejection of independence. The binary expansion approach also connects the symmetry statistics with the current computing system to facilitate efficient bitwise implementation. Modeling with the binary expansion linear effect (BELIEF) is motivated by the fact that two linearly uncorrelated binary variables must be also independent. Inferences from BELIEF are easily interpretable because they describe the association of binary variables in the language of linear models, yielding convenient theoretical insight and striking parallels with the Gaussian world. With BELIEF, one may study generalized linear models (GLM) through transparent linear models, providing insight into how modeling is affected by the choice of link. We explore these phenomena and provide a host of related theoretical results. This is joint work with Benjamin Brown and Xiao-Li Meng.

Abstract:  We analyze the identification of the discount factor in standard stationary infinite horizon dynamic discrete choice models. In single-agent models, the discount factor is identified up to a finite number of points under an exclusion restriction. The cardinality of the identified set is no larger than the cardinality of the state space. Further, the commonly used functional form assumptions on period utility function, such as additivity in a dummy variable or linearity in a state variable, provide identifying restrictions even when no exclusion restrictions are available. In multiple-agent models, the discount factor is identified up to a finite number of points under assumptions on the period profit function, including normalization at one choice, irrelevance of other firms' lagged actions, and exchangeability of other firms' actions.

Abstract:  Using a novel dataset that integrates inflation expectations with information on social network connections, we show that inflation expectations within one's social network have a positive, causal relationship with individual inflation expectations. This relationship is stronger for groups that share demographic characteristics, such as gender, income, or political affiliation. In a monetary-union New-Keynesian model, socially determined inflation expectations induce imperfect risk-sharing, and can affect the inflation and real output propagation of local and aggregate shocks. To reduce welfare losses due to socially determined expectations, monetary policy should optimally put more weight on the inflation rate of socially more connected regions.

Abstract:  This paper studies consumers' privacy choices when firms can use their data to make personalized offers. We first introduce a general framework of personalization and privacy choice, and then apply it to personalized recommendations, personalized prices, and personalized product design. We argue that due to firms' reaction in the product market, consumers who share their data often impose a negative externality on other consumers. Due to this privacy-choice externality, too many consumers share their data relative to the consumer optimum; moreover, more competition, or improvements in data security, can lower consumer surplus by encouraging more data sharing.

Abstract:  The recent influx of financial traders into carbon allowance markets has raised concerns about its distortive effects on carbon allowance prices and its repercussions for firms that rely on these price signals to make emissions decisions. This paper studies how financial carbon trading affects the allocative efficiency of carbon allowance markets and highlights the importance of facilitating financial arbitrages rather than imposing restrictions. Exploiting allowance transaction data in the European carbon market and using carbon policy shocks as supply shifters, I identify a price-inelastic carbon demand by large financial traders. The lack of elastic arbitrage capital is associated with a decline in the carbon price informativeness and contributes to the carbon market crash during the Russia-Ukraine war. The decreased informativeness has real effects: I find that firms with inferior private information reduce their emissions less efficiently when the carbon price is less informative, and the cross-sectional dispersion of carbon intensity increases with the informational inefficiency. I develop a macro-finance model with managerial learning from carbon prices that rationalizes these novel empirical findings.

Abstract:  Analyzing data from an automobile insurance company, we find that policyholders’ past accident experience reduces their accident probabilities in the current period. Our explanation is that accidents make policyholders update their beliefs about their innate risk types upward and hence exert more caution. We rule out alternative explanations, including changes in financial incentives to prevent future accidents, underreporting of minor accidents, more limited funds after an accident, etc. We find that the learning effect is stronger for female drivers than for male drivers because female drivers are more risk-averse.

Moreover, we find that the magnitude of the negative effect of a past accident is smaller if the accident occurred longer ago, suggesting a forgetting effect in the learning process. Structural estimations show that the monthly forgetting rate is 0.1385.

One managerial implication is that current insurers’ pricing schedules are inconsistent with policyholders’ actual risk dynamics. After an accident, the policyholder’s premium will reach the highest level in the next contract year and then decrease thereafter. However, because of learning with forgetting, her risk should reach its lowest level in the next contract year and then increase thereafter.

Abstract:  This paper investigates the effect of a credit supply shock on housing prices using province-level panel data covering all regions in China for the period 1999–2021. The credit shock used as an instrument is an event of bank branching deregulation in 2009, whose impact differs across provinces. The results indicate that the bank branching deregulation causes an increase in bank credit. We find that an increase in bank credit, instrumented by the measure of deregulation, leads to a sizable increase in the house price growth by 1 percentage point after the shock. The cumulative effect increases over time. Four years after the shock, house price growth is 2 percentage points higher. The pre-event trend and the placebo test suggest that the identified shock is exogenous and not related to demand shocks.

Abstract:  A seller can offer an experience good directly to consumers and indirectly through an intermediary. When selling indirectly, the intermediary provides recommendations based on the consumer’s match value and the prices at which the product is sold. The intermediary faces the trade-off between extracting rents from consumers who strongly care about the match value versus providing less informative recommendations but also serving consumers who do not. We analyze the allocative and welfare effects of prohibiting price parity clauses and/or regulating the intermediary’s recommender system. Prohibiting price parity clauses is always welfare decreasing in our model.

Abstract:  We examine the impact of liquidity shocks on commission-based financial advisors through a quasi-natural experiment. Following a regulatory change in Singapore, financial advisors now face a capped commission limit for the first year of regular premium life insurance policies, with the remaining income distributed over subsequent years. Our findings show that a decrease in liquidity from an income source causes financial advisors to adjust their supply of financial advice. Within the life insurance market, a 1 percent decrease in the commission rate for the first year of regular premium life insurance results in a 1.8 percent reduction in policy sales relative to single premium life insurance, which retains an unchanged commission structure. Additionally, financial advisors who also work as stockbrokers increased their client base. There is no evidence that they are more risk-seeking. Consistent with the buffer stock model, financial advisors reduced their personal stock market investments. Our results inform optimal regulatory design in consumer finance markets with intermediary payment structures.

Abstract:  We provide necessary and sufficient conditions for an (Unbiased) Block estimator to have Uniformly Minimum Variance. Our theory parallels the theory of UMVU estimation, the main novel insight being the focus on the covariance among blocks. We use this theory to derive lower variance bounds for block estimators of functionals of high-frequency volatility when the block size is fixed. We further show the relevance of the new theory for the classical problem of estimation of homoskedastic nonparametric regressions with varying mean. Finally, we introduce a new test for the presence of drift in financial data which exploits the precision of BUMVU estimators. The test shows abundant presence of drift in financial data.

Abstract:  We examine the choices and consequences of digital technology adoption at the firm-level, by modelling the adoption as the result of a firm’s optimal decision. Using the endogenous treatment effect framework, we address simultaneously which firm characteristics are more likely to lead to adoption and what is the outcome of the adoption. We then apply the model to a panel of Chinese public listed firms. Different from existing literature on robots/automation and using firm-level data from advanced economies, we find evidence for a negative selection – on average it is those otherwise less profitable Chinese firms that adopted digital technology. The average treatment effect on the treated suggests digital technology has led to a 4% increase in profit for those adopted firms. They also gain market share, reduce production cost, enhance productivity, and substitute capital for labour. More interestingly, the average treatment effect on the non-treated predicts a potentially large gain for those non-adopted firms, suggesting on average the annualized cost of adoption for such firms should be at least 5% of their annual profit.

Abstract:  Artificial Intelligence (AI) is increasingly utilized to assist creative professionals in their work. With the emergence of General AI, more nonroutine and creative tasks previously reserved for human workers can now be automated. While existing research has focused on the effects on the final outcomes, there is a gap in understanding the impact on the creative process itself. To address this gap, we conducted an experimental study involving over 200 professional artists, designers, and art school students. Participants were tasked with creating two illustrations: one without AI assistance and the other with access to text-to-image AI technology. By employing a within-subject design, we were able to examine how AI influences the creative workflow, how creators leverage the technology differently, and how it affects the final output. Our results indicate that AI significantly reshaped the creative workflow, accelerating the process without compromising the quality of the output. Our findings have implications for the job design and worker skills in the creative industries.

Abstract:  Specialization enhances productivity, but it also introduces costs that can limit its overall benefits. One such cost, often overlooked, is the underappreciation of other specializations. In this laboratory experiment, we investigate how this underappreciation leads to inefficiencies and inequities in real-world interactions. Participants are randomly assigned to one of two distinct specializations. Different specialists do distinct tasks, with varying difficulties within specializations. After allocation, they are paired with someone from the other specialization to engage in joint production, followed by bargaining over their joint output. We manipulate two factors: whether participants can experience their partner's task (Exposure treatment) and whether the expected output of each task is revealed (Information treatment).

We find that the final allocation is more favourable for participants with easier tasks, but this effect is diminished with exposure. Exposure leads those with easier tasks to propose a smaller share initially, while the effects of exposure and information on those with harder tasks depend on their partner’s performance. Information increases the number of bargaining made before agreement, while exposure reduces them. Over time with exposure, productivity modestly improves. Following the spirit of Cappalen et al., (2010), in a FMM model, exposure and information reduce the prevalence of Strict and Liberal Egalitarian types, shifting preferences toward Weighted Liberal Egalitarianism, where both actual and expected output are considered in allocation decisions.

Abstract:  TBC

Abstract:  We study minimax regret treatment rules in finite samples under matched treatment assignment in a setup where a policymaker, informed by a sample, needs to decide between T different treatments for a T≥2. Randomized rules are allowed for. We show that the generalization of the minimax regret rule derived in Stoye (2009) for the case T = 2 is minimax regret for general finite T > 2. We also show by example, that in the case of random assignment the generalization of the minimax rule in Stoye (2009) to the case T > 2 is not necessarily minimax regret and derive minimax regret rules for a few small sample cases, e.g. for N = 2 when T = 3. 

We also discuss numerical approaches to approximate minimax regret rules for unbalanced panels. We then study minimax regret treatment rules in finite samples when a specific quantile (rather than expected outcome) is the object of interest. We establish that all treatment rules are minimax regret under "matched" and "random sampling" schemes while under "testing an innovation" no-data rules are shown to be minimax regret.

Abstract:  This paper studies the long-term effects of technology and know-how transfers on structural transformations. In the 1950s, the Soviet Union supported the construction of the 156 Projects, which were large-scale, capital-intensive industrial clusters in China. These projects included a technology transfer, consisting of state-of-the-art Soviet machinery and equipment, and a know-how transfer, via the training of Chinese engineers, production supervisors, and high-skilled technicians by Soviet experts. We use newly assembled data that follow steel plants for over four decades, and we exploit natural variation in the transfers they eventually received. We find that, while production advantages stemming from Soviet technology faded away if not complemented with training, the know-how transfer had a long-lasting impact on plant performance, stimulated technology upgrade when China was a closed economy, and increased exports to the Western world when China engaged in international trade. The know-how transfer also generated productivity and technology spillovers onto complementary establishments.

Abstract:  Asian males are often perceived through the "model minority" myth, which can obscure the discrimination they face. While research has documented discrimination against Asians, little is known about how a setting’s diversity affects perceptions of their status. This study examines how diversity and stereotype-challenging information influence the perception of discrimination in hiring contexts. Participants are assigned to either an information treatment group, which receives an article on the disadvantaged status of Asians, or a control group. They then evaluate multiple hiring outcomes featuring varied ethnic and gender diversity.

We hypothesize that in the control group, increased diversity will result in lower perceived discrimination against Asian males. This difference is attributed to: 1) attention dilution, 2) biases favoring Asians due to the "model minority" stereotype, and 3) an assumption that diversity equals to fairness. In the treatment group, we expect heightened awareness of Asian males' status, leading to higher perceived discrimination against them, thus narrowing the gap with other minority groups.

By incorporating eye-tracking technology, we assess how attention mediates perceptions of discrimination. Our findings will provide crucial insights into the implicit discrimination faced by Asian males in the job search process, a topic of particular relevance in contemporary discussions about affirmative action and diversity policies.

Abstract:  There is a common presumption that network effects lead firms to race to enter first to gain a network effect advantage. We explore this possibility in a model of two vertically differentiated firms deciding whether and when to enter a new market in an infinite horizon setting. The firms face a trade-off between delaying entry to wait for market demand to grow and entering earlier to preempt their rival. Without network effects, the high-quality firm is the only one to enter, and it enters too early due to the threat of preemption from the low-quality firm. Introducing network effects can either exacerbate or alleviate this inefficiency, depending on consumers' beliefs about equilibrium selection in the off-equilibrium subgame where both firms enter, as well as other model primitives. We show that this is true when the beliefs favor the firm that enters first. However, when beliefs always favor the high-quality firm, entry occurs later (more efficiently) as there is less concern about preemption by the low-quality firm. In the case of adaptive beliefs, where the firm that won in the previous period is favored, surprisingly results can go in either direction. If firms are patient enough, it can result in the most efficient (later) entry for the high-quality firm. Thus, we show whether network effects really intensify the race to enter a market depends on the nature of beliefs, the discount factor, the asymmetry of the firm’s quality and the strength of network effects.

Abstract:  Many structural econometric models include latent variables on whose probability distributions one may wish to place minimal restrictions. Leading examples in panel data models are individual-specific variables sometimes treated as “fixed effects” and, in dynamic models, initial conditions. This paper presents a generally applicable method for characterizing sharp identified sets when models place no restrictions on the probability distribution of certain latent variables and no restrictions on their covariation with other variables. In our analysis latent variables on which restrictions are undesirable are removed, leading to econometric analysis robust to misspecification of restrictions on their distributions which are commonplace in the applied panel data literature. Endogenous explanatory variables are easily accommodated. Examples of application to some static and dynamic binary, ordered and multiple discrete choice and censored panel data models are presented.

Abstract:  We study the optimal inflation target in an open economy with a zero lower bound (ZLB) on nominal interest rates calibrated to the Euro area. When uncovered interest rate parity (UIP) holds, the optimal inflation target is smaller than in a closed economy. The key to this result is that real interest rates at the ZLB increase less as trade openness increases. This less pronounced increase in real interest rates in an open economy mitigates the contraction in aggregate demand, thus reducing the cost of ZLB. Additionally, the optimal inflation target in a monetary union is lower than in a flexible exchange rate regime: forming a monetary union results in a decrease in the optimal inflation rate by 0.24\%, thereby increasing per-period welfare by 0.07\%. When departures from UIP are significant, the optimal inflation target may be higher than in a closed economy, and the desirability of a monetary union increases further. 

Abstract:  We fully characterise (strong) implementability with transfer and without invoking quasilinearity, via a novel cyclical monotonicity condition that extends Rochet (1987). We then apply it to crack the problem of implementing monotone allocations for the general case under the assumption of the possibility of compensation: (i) the single crossing condition, when type space is totally ordered and outcome space is partially ordered, is sufficient and necessary for strongly implementing all monotone allocation; and (ii) with the strict single-crossing condition and totally ordered type space and outcome space, an allocation is implementable if and only if it is monotone. No additional structure or regularity conditions are needed.

Abstract:  1. We find that positive news extracted by ChatGPT from the front pages of the Wall Street Journal is related to macroeconomic conditions and can predict monthly stock market returns. Consistent with existing theories, investors tend to underreact to positive news, especially during periods of economic downturns, high information uncertainty, and high news novelty. However, negative news is negatively associated with contemporaneous returns and has no predictive power. We find further that traditional methods, such as word lists and BERT, fail to have comparable predictability, and ChatGPT appears at present the best in capturing economic news about the market risk premium.

2. We introduce a general approach for analyzing large-scale text-based data, combining the strengths of neural network language processing and generative statistical modeling to create a factor structure of unstructured data for downstream regressions typically used in social sciences. We generate textual factors by (i) rep- resenting texts using vector word embedding, (ii) clustering the vectors using Locality-Sensitive Hashing to generate supports of topics, and (iii) identifying relatively interpretable spanning clusters (i.e., textual factors) through topic modeling. Our data-driven approach captures complex linguistic structures while ensuring computational scalability and economic interpretability, plausibly attaining certain advantages over and complementing other unstructured data analytics used by researchers, including emergent large language models. We conduct initial validation tests of the framework and discuss three types of its applications: (i) enhancing prediction and inference with texts, (ii) interpreting (non-text-based) models, and (iii) con- structing new text-based metrics and explanatory variables. We illustrate each of these applications using examples in finance and economics such as macroeconomic forecasting from news articles, interpreting multi- factor asset pricing models from corporate filings, and measuring theme-based technology breakthroughs from patents. Finally, we provide a flexible statistical package of textual factors for online distribution to facilitate future research and applications.

Abstract:  Food choices are crucial for health, and yet have been found easily influenced by choice architecture strategies. We conduct a randomized controlled experiment to evaluate the effects of different strategies in food choices, and employ a machine learning method to assess the individual-level heterogeneity of these effects. Participants select ten out of twenty food items and are randomly assigned to one of eight conditions: two benchmarks (control and financial incentives) and six choice architectures (portion flexibility, information, defaults, recommendations, bundling, and categorization). We find that providing calorie information and setting healthy defaults significantly improve the healthiness of choices, comparable to providing 20-30% of the choice value as financial incentives for healthy choices. Allowing portion flexibility significantly reduces calorie and fat intake, with 69% of selections including at least one half-portion item at the same price. The varying effect heterogeneity of treatments provides insights for policy targeting and highlights fairness concerns in choice architecture implementation.

Abstract:  Recent studies have indicated that including the declining trends in nominal yields (falling stars) in Gaussian affine term structure models leads to more accurate esti- mates of the term premium. Considering how the behavior of short-term yields varies significantly during zero lower bound (ZLB) periods, there is reason to believe that term premium dynamics might undergo structural changes when interest rates are at or near zero. To explore this possibility, we introduce and estimate a novel arbitrage-free affine term structure model that incorporates falling stars, along with regime shifts between ZLB and non-ZLB states. Our model comparisons reveal that the new model outperforms the model that only incorporates falling stars. More importantly, our findings suggest that the model excluding regime shifts does not adequately capture the temporary yet significant fluctuations in the term premium observed during ZLB periods. This is be- cause the influence of factors representing the yield spread remains understated if the compression of yields during ZLB periods is not considered. Consequently, both falling stars and the ZLB seem essential elements of the U.S. yield curve that must be included to accurately estimate the term premium.

Abstract:  This talk explores the intriguing idea of "Intelligence Darwinism" where decision quality built on intelligence capacity enabled by information capacity derived from brain (synaptic) plasticity shapes natural selection through inter- and intra-species competition. Through engaging examples and simple explanations, we seek to unravel the ways in which intelligent decision making delivers biodiversity through survival of the smartest from the lowly worm, C. Elegans, whose brain has 302 neurons to us homo sapiens endowed with 86 billion neurons. Join us to explore and discover the fascinating interplay between information and intelligence capacities in driving evolutionary progress. It has also not escaped attention that our research has natural implications for what artificial intelligence may portend for the future of our species.

Abstract:  This paper addresses two questions: What is the economic nature of consumer data, and how should consumers be compensated for their data? We develop a theoretical model to analyze compensation mechanisms for consumer data collection. Digital platforms provide essential digital services and subsequently collect consumer data for monetization. Our analysis sheds light on the interdependence between data collection and digital service provision, and explores the nature of cross-subsidization in the design of compensation mechanisms. We examine the optimal compensation schemes by a monopolistic digital platform and provide policy implications for digital privacy regulations.