Giovanni RIZZI's PhD Thesis, 09 October 2026

Essays in Industrial Economics

Supervisors

Professor Patrick FEVE

To participate, please contact the doctoral school

Memberships

• Patrick REY: Professor in Economics,TSE-R, University Toulouse Capitole Supervisor
• Ozlem BEDRE-DEFOLIE: Professor in Economics, European University Institute Rapporteure
• Andrei HAGIU : Professor in Economics, Questrom School of Business, Boston University Rapporteur
• Alexandre DE CORNIERE : Professor in Economics, TSE-R, University Toulouse Capitole Examinateur
• Emilio CALVANO : Professor in Economics,Professor in Economics, Examinateur

Abstract

The three chapters that make up this thesis study how the underlying microeconomic structure of technology and strategic 
decisions determines the outcome of competition. More specifically, the first chapter identifies a property of Bayesian learning 
which generates data-driven network effects, which the second chapter shows can lead to pervasive market failure in the 
market for digital applications. The third chapter connects the incentives of firms engaging in collusion to explain the unexplored 
link between macroeconomic stability and market power.
Chapter 1, A Theory of the Value of Data, develops a theory of the value of data for prediction. A firm predicts a target variable 
for a target individual and designs its dataset along three margins: the number of training observations, the covariates recorded 
in the training sample, and the covariates recorded for the target. A Bayesian model with a random-matrix (Marchenko–Pastur) 
limit yields a closed-form value of data, the expected fall in out-of-sample error. The value factors into two pieces: how much 
the training data teach the firm about the covariate-to-outcome mapping, and how many target covariates the firm can apply 
that knowledge to. Two results follow. Covariates exhibit economies of scope: one covariate grows more valuable as the firm 
observes others, because broader scope sharpens learning about every weight. And observations and covariates complement 
each other in small datasets yet turn into substitutes in large ones, so the largest firms gain least from adding users. These 
curvature properties speak directly to merger review: pooling covariates across the same users creates synergies, while pooling 
users who share the same covariates runs into diminishing returns.
Chapter 2, Business-Model Competition and Privacy-preserving Apps, which is joint with Doh-Shin Jeon and Shota Ichihashi, asks 
why paid, privacy-preserving apps make up only 3 to 5 percent of major app stores. The answer is a market failure. An ad-tech 
combines data across all ad-funded apps, so those apps enjoy data-driven network effects: more consumers served by any adfunded app raises targeting accuracy, and hence ad revenue per user, for every ad-funded app. A continuum of app categories, 
endogenous entry, and an equilibrium data stock fixed point produce a data multiplier that amplifies any exogenous rise in data. 
Better targeting competes on the intensive margin but deters privacy-preserving entry on the extensive margin. With any 
positive entry cost, consumer surplus is single-peaked in targeting accuracy. When ad-funded apps dominate, consumers gain 
from limits on data combination and from digital-advertising taxes, and the chapter characterizes when consent mandates and 
an ad valorem tax implement the consumer-preferred market structure.

Chapter 3, The Great Moderation and the Rise of Markups, which is joint with Friedrich Lucke and Giovanni Morzenti, identifies 
a cost of business-cycle stability: stable demand makes tacit collusion easier to sustain. In a repeated oligopoly, the temptation 
to undercut peaks when demand peaks, so lower volatility raises the highest sustainable collusive markup, most sharply in 
concentrated markets. A US state–sector panel spanning the Great Moderation tests the mechanism, instrumenting volatility 
with the staggered deregulation of interstate banking. A 1 percent fall in volatility raises markups by roughly 0.2 percent; the 
effect concentrates in the most concentrated quartile and vanishes in the least. Under CES demand the estimates imply a 
monopoly markup of 1.317, close to the data, and the decline in volatility accounts for up to two-thirds of the markup rise 
between 1980 and 1997. The finding exposes a trade-off between macroeconomic stabilization and competition: merger 
scrutiny should tighten as demand stabilizes.