The Rise of Algorithmic Trading: Implications for Price Elasticity and Market Competitiveness [Link]
Job Market Paper
Financial markets’ demand elasticity has decreased substantially over the last 20 years, leading to increased volatility and undermining price stability. This paper investigates the impact of the use of AI traders that make unsupervised trading decisions on aggregate stock elasticity. Building on Haddad et al. (2021), I estimate investors’ demand and then I simulate a fictitious financial market populated by artificial intelligence (AI) traders, whose investment decisions are governed by a neural network-based reinforcement learning algorithm. The introduction of AI traders has non-trivial effects on elasticity: on the one hand, because of their high individual-specific elasticity they increase aggregate elasticity compared to a market that is not populated by algorithmic traders. On the other hand, when these agents reduce their exposure on the risky asset, aggregate elasticity drops by around 2.5%. The last is larger when other investors demand is more sensitive to aggregate market elasticity.
Artificial Intelligence, Algorithmic Trading, and Market Stability (with E. Tarantino and F. Sangiorgi) [SSRN]
We investigate how AI-driven investors, modeled via deep reinforcement learning, operate in a calibrated financial market with realistic return predictability and endogenous price impact. We examine whether these agents can learn to detect and exploit return predictability from public signals, decode prices to infer latent demand, and adjust for price impact. To evaluate performance, we compare AI traders to a rational benchmark representing the optimal policy under full knowledge of the data generating process. In simulations, AI traders qualitatively match the benchmark. Quantitatively, however, they fall short when many interact. The presence of other AI traders injects noise into the price process through their exploration, distorting the portfolio-return signals each agent learns from and thus impairing learning. This negative learning externality reduces trading profits, lowering market efficiency and liquidity relative to the benchmark. Our findings suggest caution when extrapolating from partiale-quilibrium analyses of AI trading’s profitability and impact on market quality.
Robust Identification in Repeated Games: An Empirical Approach to Algorithmic Competition (with A. Cozzolino, C. Gualdani, N. Lomys, and L. Magnolfi) [Link]
We develop an econometric framework for recovering structural primitives---such as marginal costs---from price or quantity data generated by firms whose decisions are governed by reinforcement-learning algorithms. Guided by recent theory and simulations showing that such algorithms can learn to approximate repeated-game equilibria, we impose only the minimal optimality conditions implied by equilibrium, while remaining agnostic about the algorithms’ hidden design choices and the resulting conduct---competitive, collusive, or anywhere in between. These weak restrictions yield set identification of the primitives; we characterise the resulting sets and construct estimators with valid confidence regions. Monte~Carlo simulations confirm that our bounds contain the true parameters across a wide range of algorithm specifications, and that the sets tighten substantially when exogenous demand variation across markets is exploited. The framework thus offers a practical tool for empirical analysis and regulatory assessment of algorithmic behaviour.
Passive investing, diversification risk and financial stability [SSRN] NEW!
This paper examines whether the rise of passive investing affects the equity market's capacity to distribute aggregate risk. Using a valuation-adjusted Bartik instrument, I establish that a one-standard-deviation increase in passive ownership raises average pairwise stock return correlations by 9 percentage points. Through a structural demand-system framework, I decompose portfolio risk into correlation and idiosyncraticvolatility components and uncover two opposing forces: passive investing synchronizes asset returns through common, price-insensitive demand shocks, while simultaneously dampening firm-specific volatility. The correlation effect dominates. Relative to a counterfactual economy without passive investors, diversification benefits in the global minimum variance portfolio are 3% lower and its minimum achievable volatility is 19% higher; for the risk parity portfolio the losses are larger, at 9.5% and 25% respectively. Welfare costs concentrate among large, well-diversified institutions and are amplified during recessions-precisely when effective diversification matters most.