1\subsection{The classical economy: stylized facts and mechanism checks}2\label{sec:results_classical}34Before any language model enters the market, the classical economy must (i) reproduce the stylized facts5it was calibrated to, and (ii) respond to the mechanisms of Section~\ref{sec:theory} in the predicted6direction. Table~\ref{tab:stylized_facts} summarizes the 30 control seeds of the classical design7(seeds 101--130): first-order autocorrelation of tick returns is indistinguishable from zero8($-0.007$, seed SD 0.15), the variance ratio at five ticks is 1.18, excess kurtosis averages 20 with a9minimum of 1.8 across seeds, the first ten autocorrelations of absolute returns sum to $2.0$, the quoted10spread is 21~bp (effective 30~bp), and the mid tracks the latent value with a mean absolute deviation11of 0.63\,\%. The market absorbs the latent $+2.2\,\%$ shock of tick~300 in 25~ticks on average, i.e.\12before the dividend of tick~320 reveals it, through the informed fundamentalists and the news flow; the1320-tick response ratio of 0.28 says that most of the adjustment happens in the last ticks before the14announcement. No control seed produces a bubble episode (5\,\% for ten ticks) or a crash (10\,\% in15twenty ticks), which is the floor effect anticipated in Section~\ref{sec:identification}: in the16classical economy instability can only be created, not reduced.1718Table~\ref{tab:classical_cells} reports cell means and Table~\ref{tab:classical_effects} the paired19effects against the control for the five primary outcomes, with Holm-adjusted $p$-values within each20cell. Four results matter for what follows.2122\paragraph{Feedback and leverage create instability (Propositions~\ref{prop:feedback} and \ref{prop:spiral}).}23Raising the trend-follower share from 6 to 16\,\% raises the tracking error by 0.86 percentage points24($d = 1.26$), the spread by 14~bp and the maximum drawdown by 2.8 points, and creates bubble episodes25in 13\,\% of seeds where the control has none. Raising the leveraged share from 4 to 19\,\% has larger26effects on every outcome: tracking error $+2.1$ points, time to absorb $+28$ ticks, spread $+16$~bp,270.9 bubble episodes per run and a maximum drawdown of 7.8\,\% against 3.3\,\%; adding a funding shock28(a 50\,\% cut of the leverage cap twenty ticks after a negative latent shock) pushes the bubble count to291.4 and the drawdown to 9.4\,\%, and produces crashes in 30\,\% of seeds. Figure~\ref{fig:baseline_paths}30shows a representative seed of each economy. Leveraged traders lose money in both cells ($-4.8$ and31$-6.0$\,\% of initial wealth against $+0.7$\,\% in the control), and the losses are transferred to the32fundamentalists and market makers who absorb the forced sales.3334\paragraph{Information quality and misleading news.}35Lowering the precision of private signals from 2 to 0.5 has no detectable effect on any primary outcome,36and raising it to 8 tightens the spread by 3~bp and the drawdown by half a point without changing the37tracking error. The small effects are a property of this economy rather than a failure of the38information channel: the latent shock is revealed by the next dividend within at most twenty ticks and39by public news in the meantime, so private signals shorten the absorption of a shock that the market40would absorb anyway. Misleading news is different: when half of the public items carry a hidden4110-cent bias, the tracking error rises by 0.08 points ($d = 0.74$), the spread by 3~bp and the drawdown42by half a point, because the public value itself is pulled away from the latent value and the43fundamentalists trade on it. This cell is the classical counterpart of a shared, biased model: a common44error in the public signal degrades price discovery even when every trader is otherwise rational.4546\paragraph{Liquidity withdrawal.}47When the five market makers stop quoting for sixty ticks at the moment of the latent shock, the shock48takes 8~ticks longer to absorb ($p = 0.05$ unadjusted), but the tracking error, spread and drawdown49measured over the whole post-warmup window are unchanged: the remaining limit-order traders supply50enough liquidity for the price to find its level once the market makers return.5152\paragraph{Safeguards and market design.}53The 5\,\% circuit breaker almost never fires in the classical economy (0.13 halts per run) and has no54effect; it matters only where prices move by 5\,\% in ten ticks, which the LLM arms below test. The55position limit of 300 lots is actively harmful here: it caps exactly the traders who absorb order56imbalances, so the tracking error rises by 1.3 points, the time to absorb by 28 ticks and the spread by5718~bp, and 0.6 bubble episodes per run appear. The periodic call auction (clearing every five ticks with58a $\pm 2\,\%$ collar) raises the tracking error by 0.7 points and the drawdown by 2.3 points relative59to continuous trading at the same shock, which is the expected cost of batching with a mechanical60collar; it is reported as a mechanism-robustness check in Section~\ref{sec:robustness}, not as a61recommendation.6263\paragraph{Placebo and negative controls.}64Replacing the latent shock by two placebo events leaves every outcome within its control range except65the tracking error, which falls by 0.1 points because there is no shock to absorb; no bubble, crash or66halt appears. The two cells in which 25 or 50\,\% of the slots run the \emph{mock} policy, a seeded67value-plus-noise rule with independent errors, are the negative control for the LLM treatment: with68independent errors, replacing a quarter or half of the traders changes neither the tracking error nor the69time to absorb, bubbles or drawdowns, and tightens the spread (by 4 and 7~bp) because the mock rule70rests limit orders near the mid. Whatever the live LLM arms do beyond this is attributable to the71language model's decisions, not to the act of replacing slots.7273\paragraph{Two assets.}74With two assets whose dividend innovations load 0.7 on a common factor, the fundamental return75correlation across dividend periods is 0.38 while tick-return correlation between the two mids is76$-0.01$: microstructure noise dominates at the tick frequency and the two books are priced almost77independently, so excess comovement is negative in the control ($-0.39$). Adding leveraged traders and a78funding shock to asset~0 produces crashes in asset~0 in half of the seeds and raises the maximum drawdown79of asset~1 from 3.6 to 11.6\,\% although its fundamentals are untouched, and the tracking error of80asset~1 from 0.6 to 7.4\,\%: contagion through common funding, not through correlated fundamentals. The81Diebold--Yilmaz spillover index stays near 1.2\,\% in both cells, because the variance decomposition of a82VAR on tick returns does not pick up the forced-sale episodes (Table~\ref{tab:contagion}).8384\begin{table}[htbp]85\centering86\caption{Stylized facts of the classical control economy (30 seeds, post-warm-up window).}87\label{tab:stylized_facts}88\small89\input{tables/tab_stylized_facts}90\end{table}9192\begin{table}[htbp]93\centering94\caption{Classical mechanism checks: cell means (30 matched seeds per cell). Stars: paired $t$-test95against the control, Holm-adjusted within the cell ($^{*}$ 10\,\%, $^{**}$ 5\,\%, $^{***}$ 1\,\%).}96\label{tab:classical_cells}97\scriptsize98\resizebox{\textwidth}{!}{\input{tables/tab_classical_cells}}99\end{table}100101\begin{table}[htbp]102\centering103\caption{Classical mechanism checks: paired effects against the control with 95\,\% confidence intervals,104Cohen's $d_z$ and Holm-adjusted $p$-values (five primary outcomes per cell).}105\label{tab:classical_effects}106\scriptsize107\resizebox{\textwidth}{!}{\input{tables/tab_classical_effects}}108\end{table}109110\begin{table}[htbp]111\centering112\caption{Two-asset economy: comovement and spillovers (30 seeds per cell).}113\label{tab:contagion}114\small115\resizebox{\textwidth}{!}{\input{tables/tab_contagion}}116\end{table}117118\begin{figure}[htbp]119\centering120\includegraphics[width=\textwidth]{figures/fig_baseline_paths}121\caption{Mid price, latent value and public value in one seed (101) of three classical economies. The122latent $+10$-cent shock of tick~300 is revealed by news and by the dividend of tick~320.}123\label{fig:baseline_paths}124\end{figure}125126\begin{figure}[htbp]127\centering128\begin{subfigure}{0.49\textwidth}\includegraphics[width=\textwidth]{figures/fig_classical_mad_pct}\caption{Tracking error}\end{subfigure}129\begin{subfigure}{0.49\textwidth}\includegraphics[width=\textwidth]{figures/fig_classical_time_to_absorb}\caption{Time to absorb the latent shock}\end{subfigure}130\begin{subfigure}{0.49\textwidth}\includegraphics[width=\textwidth]{figures/fig_classical_max_dd}\caption{Maximum drawdown}\end{subfigure}131\begin{subfigure}{0.49\textwidth}\includegraphics[width=\textwidth]{figures/fig_classical_spread_bp}\caption{Quoted spread}\end{subfigure}132\caption{Classical mechanism checks: cell means with 95\,\% confidence intervals over 30 seeds (control in grey).}133\label{fig:classical}134\end{figure}135