Latest Research Publications

We investigate a data-driven multiperiod inventory replenishment problem with uncertain demand and vendor lead time (VLT) with accessibility to a large quantity of historical data. Different from the traditional two-step predict-then-optimize (PTO) solution framework, we propose a one-step end-to-end (E2E) framework that uses deep learning models to output the suggested replenishment amount directly from input features without any intermediate step. The E2E model is trained to capture the behavior of the optimal dynamic programming solution under historical observations without any prior assumptions on the distributions of the demand and the VLT. By conducting a series of thorough numerical experiments using real data from one of the leading e-commerce companies, we demonstrate the advantages of the proposed E2E model over conventional PTO frameworks. We also conduct a field experiment with JD.com, and the results show that our new algorithm reduces holding cost, stockout cost, total inventory cost, and turnover rate substantially compared with JD’s current practice. For the supply chain management industry, our E2E model shortens the decision process and provides an automatic inventory management solution with the possibility to generalize and scale. The concept of E2E, which uses the input information directly for the ultimate goal, can also be useful in practice for other supply chain management circumstances.

Estimation of the covariance matrix of asset returns is crucial to portfolio construction. As suggested by economic theories, the correlation structure among assets differs between emerging markets and developed countries. It is therefore imperative to make rigorous statistical inference on correlation matrix equality between the two groups of countries. However, if the traditional vector-valued approach is undertaken, such inference is either infeasible due to limited number of countries comparing to the relatively abundant assets, or invalid due to the violations of temporal independence assumption. This highlights the necessity of treating the observations as matrix-valued rather than vector-valued. With matrix-valued observations, our problem of interest can be formulated as statistical inference on covariance structures under sub-Gaussian distributions, i.e., testing non-correlation and correlation equality, as well as the corresponding support estimations. We develop procedures that are asymptotically optimal under some regularity conditions. Simulation results demonstrate the computational and statistical advantages of our procedures over certain existing state-of-the-art methods for both normal and non-normal distributions. Application of our procedures to stock market data reveals interesting patterns and validates several economic propositions via rigorous statistical testing.

A university uses both early-stage selection outcome (high-school affiliation) and late-stage admission test outcome (standardized test scores) to select students. We use this model to study policies that have been proposed to combat inefficient gaming in college admissions. Increasing university enrollment size can exacerbate gaming and worsen the selection outcome. Abolishing standardized tests for university admissions increases gaming targeting high-school admissions and worsens the selection outcome, whereas eliminating high-school ability sorting may improve the university selection outcome under some cost conditions of gaming. Committing to a lower-powered selection scheme can improve the selection outcome by reducing gaming behaviors.

New products are highly valued by manufacturers and retailers due to their vital role in revenue generation. Product life cycle (PLC) curves often vary by their shapes and are complicated by promotional activities that induce spiky and irregular behaviors. We collaborate with JD.com to develop a flexible PLC curve forecasting framework based on Bayesian functional regression that accounts for useful covariate information, including product attributes and promotion. The functional model treats PLC curves as target variables and includes both scalar and functional predictors, capturing time-varying promotional activities. Harnessing the power of basis function transformation, the developed model can effectively characterize the local features and temporal evolution of sales curves. Our Bayesian framework can generate initial curve forecasts before the product launch and update the forecasts dynamically as new sales data are collected. We validate the superior performance of our method through extensive numerical experiments using three real-world data sets. Our forecasting framework reduces the forecasting error by 5.35%–30.76% over JD.com's current model and outperforms alternative models significantly. Furthermore, the estimated promotion effect function provides useful insights into how promotional activities interact with sales curves.

We propose a novel framework for analyzing linear asset pricing models: simple, robust, and applicable to high-dimensional problems. For a (potentially misspecified) stand-alone model, it provides reliable price of risk estimates for both tradable and nontradable factors, and detects those weakly identified. For competing factors and (possibly nonnested) models, the method automatically selects the best specification—if a dominant one exists—or provides a Bayesian model averaging–stochastic discount factor (BMA-SDF), if there is no clear winner. We analyze 2.25 quadrillion models generated by a large set of factors and find that the BMA-SDF outperforms existing models in- and out-of-sample.

We study the economic value of market timing in foreign exchange (FX) markets, that is, using information about the conditional Sharpe ratio to adjust the notional value of a conditionally mean–variance efficient currency portfolio. Our strategy trades more (less) aggressively when the conditional risk-return trade-off is more (less) favorable. This leads to a significant improvement in the out-of-sample unconditional Sharpe ratio, skewness, and maximum drawdown per 1% expected excess return. The strategy’s market timing predicts returns, volatility, and skewness in FX markets. Popular currency pricing factors do not explain the strategy’s high average excess returns. Our findings suggest that it is costly to impose leverage or risk (i.e., conditional volatility) limits or other inferior market timing policies when constructing currency trading strategies.

We study the effects of mark-to-market accounting (MTM) for banks following the originate-to-distribute lending model. Banks have expertise in originating loans, but it is costly for them to retain the loans on their books. We study how the accounting measurement of the retained loans affects the banks’ origination and retention decisions. We show that, relative to historic cost accounting (HC), MTM has three consequences. First, it improves the accuracy of loan measurement ex post. Second, it forces banks to retain more risk exposure on their own books. Finally, it can reduce ex ante origination efforts and lower the average quality of loans in the economy.

Although the significance of relational ties with supply chain partners to firms' operations management has been frequently studied, it is unclear how political ties, which represent another form of relational ties that is especially important in emerging economies, relate to operational efficiency. Drawing on the political embeddedness perspective, we propose a negative association between political ties and operational efficiency and examine how this association is moderated by environmental and firm-level factors. Using panel data of listed private firms in China, we show that political ties are negatively correlated with operational efficiency. In addition, this negative relationship is stronger when firms operate in regions with less developed factor markets and in highly competitive industries. However, the negative relationship is weaker for firms with high levels of foreign shareholding and customer concentration. These findings caution against the downsides of political ties for operations management and highlight strategies for reducing their negative effect.

A firm manager is concerned about both the firm value and the market assessments of his abilities. When investing in a project, he has private knowledge of his project-related ability that interacts with the project investment, and his general ability that produces a cash flow independent of the project cash flow. The concerns about the general (project-related) ability assessment create a signaling incentive to decrease (increase) investment. In the presence of underinvestment (overinvestment), higher-quality earnings information reduces (improves) equilibrium efficiency. When the manager issues an earnings forecast as an additional signaling device, the forecast is upwardly biased, and the equilibrium investment is smaller than that without a forecast. The latter is because the signaling incentive to decrease investment is strengthened. When the manager’s concerns about the general ability assessment are relatively large, he is better off by committing to no forecast. Novel empirical predictions about investment and earnings forecast emerge.