« Back to Results

Putting the "Ec" in Tech: Economics at Tech Firms

Paper Session

Friday, Jan. 4, 2019 12:30 PM - 2:15 PM

Atlanta Marriott Marquis, Marquis Ballroom A
Hosted By: National Association for Business Economics
  • Chair: Carolyn Evans, Intel Corp.

How Artificial Intelligence and Machine Learning Can Impact Market Design

Paul Milgrom
,
Stanford University
Steven Tadelis
,
University of California-Berkeley

Abstract

In complex environments, it is challenging to learn enough about the underlying characteristics of transactions so as to design the best institutions to efficiently generate gains from trade. In recent years, Artificial Intelligence has emerged as an important tool that allows market designers to uncover important market fundamentals, and to better predict fluctuations that can cause friction in markets. This paper offers some recent examples of how Artificial Intelligence helps market designers improve the operations of markets, and outlines directions in which it will continue to shape and influence market design.

Economics at Uber

Jonathan Hall
,
Uber

Abstract

Economics is well embedded (but not embedded enough) across Uber's organization, and we have learned the hard way over time some best practices for translating the insights of economics into business success. In this talk, I describe the role of the economist across several business units including marketplace product, corporate strategy, research, and public policy. I then discuss the skills and personality traits that seem to make economists successful and influential in each of these roles and at different levels of seniority. Graduate programs in economics supply some but not all of the experience and skills required to be successful in business, so I conclude with a few ideas for how economists can learn from non-economists to steepen the learning curve.

Digital Disintermediation and Efficiency in the Market for Ideas

Christian Peukert
,
Catholic University of Portugal
Imke Reimers
,
Northeastern University

Abstract

Digital technology has allowed inventors to circumvent traditional intermediaries and directly reach consumers, which may affect licensing outcomes and efficiency in the market for ideas. We study these impacts theoretically and empirically in the book publishing industry, where the number of new books available to consumers has almost doubled after the advent of digital self- publishing platforms. Using data on over 90,000 license deals between authors and publishers from 2002 to 2015, we identify disintermediation-related changes in this market from quasi-experimental variation across product types over time. Consistent with digital self-publishing improving an author’s bargaining position, we find that authors get substantially more favorable license deals. We further show that ex-ante license fees reflect ex-post demand more accurately. This is consistent with additional entry generating more information about a product type’s realized appeal. In markets in which product appeal is difficult to predict, such improvements in the information environment can have large impacts on efficiency and welfare.

A Machine Learned, Real Time Measure of the Rate of Inflation

Patrick Bajari
,
University of Washington
Victor Chernozhukov
,
Massachusetts Institute of Technology
Ramon Huerta
,
University of California-San Diego
Ashish Mishra
,
Amazon
Bernhard Schoelkopf
,
Max Planck Institute

Abstract

It has long been recognized that quality adjustments can help more accurately measure the rate of inflation when the underlying panel of products is unbalanced. However, the agencies lack techniques for constructing quality adjustments for all of the prices in the US economy. Recent developments in natural language process, computer vision and machine learning offer new tools that may assist with the scalable estimation of hedonic prices. We use a unique panel of data from the retailers Amazon which has text and pictures to construct data for hedonic adjustments for clothing prices. These techniques effectively let us use product descriptions, customer reviews and pictures as control variables to predict clothing prices. We also use a variety of ML methods including boosted trees, deep learning and half-sibling regressions to predict prices as a function of these controls. These methods have two main advantages. First, they allow us to automatically construct controls (e.g. feature engineering) for tens of millions of unique items in a scalable model. Second, our machine learned methods have considerably higher predictive accuracy than standard linear regression models used in hedonics because they accommodate high dimensional data, allow for non-linearity and search for optimal model specifications in a high dimensional space.
JEL Classifications
  • L1 - Market Structure, Firm Strategy, and Market Performance
  • D2 - Production and Organizations