Publications

This article develops a method to automatically discover and quantify human-interpretable visual characteristics directly from product image data. The method is generative, and can create new visual designs spanning the space of visual characteristics. It builds on disentanglement methods in deep learning using variational autoencoders, which aim to discover underlying statistically independent and interpretable visual characteristics of an object. The impossibility theorem in the deep learning literature indicates that supervision with ground truth characteristics would be required to obtain unique disentangled representations. However, these are typically unknown in real world applications, and are in fact exactly the characteristics we want to discover. Extant machine learning methods require ground truth labels for each visual characteristic, resulting in a task requiring human evaluation and judgment to both design and operationalize. In contrast, this method postulates the use of readily available product characteristics (such as brand and price) as proxy supervisory signals to enable disentanglement. This method discovers and quantifies human-interpretable and statistically independent characteristics without any specific domain knowledge on the product category. It is applied to a dataset of watches to automatically discover interpretable visual product characteristics, obtain consumer preferences over visual designs, and generate new ideal point designs targeted to specific consumer segments.

This paper studies the impact of capital projects on student learning and the real estate market,
using nationwide data on U.S. school districts and focusing on what investments work and on
whom. We use newly collected data on school capital bonds, test scores, and house prices for 28
U.S. states and a new research design that identifies the causal impact of bond authorizations in
the presence of dynamic and heterogeneous treatment effects. On average, bond authorization
significantly raises test scores and house prices. Yet, there are large differences across bonds and
districts. Spending on infrastructure renovation and upgrades, such as HVAC or roofs, raises
test scores but not house prices; conversely, spending on athletic facilities increases house prices
but not test scores. Bond authorization is most beneficial in districts with more disadvantaged
student populations, in part because these districts prioritize bonds that improve learning. We
find suggestive evidence that capital funding rules drive differences in bond impacts.

After confronting the myths sur- rounding momentum investing1 and value investing,2 we realized two things: 1) We had passed over the first anomaly discovered in academic finance and the one that had been around the longest—size, and 2) despite its longevity and the attention it has received, there is still much confusion and debate surrounding the size anomaly.