La complexité cachée des prix : exploration des mécanismes microstructuraux des marchés financiers

Financial markets are a cornerstone of modern economies. They facilitate exchanges between agents, enable institutional financing, and assign a price—a value—to traded securities. While traditional economic theory posits that these markets are efficient—meaning they instantaneously and perfectly incorporate all available information—reality often tells a different story, as evidenced by the many financial crises, speculative bubbles, abrupt price jumps, and excessive volatility that frequently occur yet remain poorly understood.The objective of this thesis is to improve our understanding of the price formation process by adopting a microstructural perspective—that is, by studying the most elementary mechanisms at work within electronic order books. We begin by analyzing a unique dataset from the Tokyo Stock Exchange, which includes trader identifiers. This empirical study reveals new stylized facts about market impact—that is, the mechanical effect of buy and sell orders on prices.Following this empirical analysis, we propose a unified theoretical framework based on minimal assumptions. This model successfully reconciles the statistical properties of order flow, those of price dynamics, and the nontrivial relationship between the two. It also yields testable predictions that we validate using real market data. Moreover, we develop several algorithms capable of generating realistic simulated markets that are statistically indistinguishable from actual ones, without requiring any exogenous economic information. These tools not only support our theoretical claims through their surprising effectiveness, but also open new avenues for academic and industrial research, which is often hindered by the inaccessibility of privately owned market data.We then introduce a Markovian liquidity model that reveals the existence of “microstructural modes.” These modes enable the study of market stability as well as the prediction of future order flows and price movements. They highlight the marginal stability of markets and show that crises and abnormal price jumps are already foreshadowed at the microscopic scale, in a purely endogenous manner. To better understand these instabilities, we conclude by developing an agent-based model inspired by the quadratic Santa Fe model, enriched with more realistic feedback mechanisms. We numerically demonstrate the emergence of phase transitions within the order book, estimating their critical exponents through finite-size scaling analysis and deriving their stability boundaries analytically.