spb/uqo-research-db

1% ============================================================================2\section{Robustness and Identification}3\label{sec:robustness}4% ============================================================================56The structural break is economically large; this section asks how much we should believe it. We7subject it to a placebo across non-AI factors, an event study, a difference-in-differences design, and8a battery of construction and universe robustness checks.910\subsection{Placebo: is the effect specific to AI? (H3)}11If the post-2022 spread were a mechanical artifact of our pipeline---or a generic ``high-disclosure12firms did well'' phenomenon---it should appear for any semantic factor. Table~\ref{tab:placebo} and13Figure~\ref{fig:placebo} re-run the entire sub-period and interaction design on all eight factors. The14result strongly supports AI-specificity. The three AI factors show large positive post-2022 spreads15($+13.6$ to $+14.3$ points). Macro-risk language shows essentially no spread ($-0.5$, $t=-0.1$). And,16strikingly, ESG-commitment and climate-risk disclosure earn \emph{negative} post-2022 spreads17($-8.8$ and $-5.0$ points, both significant)---high-ESG firms \emph{underperformed} after 2022, in18line with the post-2021 reversal in the sustainable-investing literature \citep{pastor2022dissecting}.19The one non-AI factor that does show a large positive post-2022 spread is cybersecurity ($+13.5$); we20read this not as a violation but as confirmation, since cybersecurity disclosure is tightly bundled21with technology exposure and is itself increasingly AI-related. The cleanest contrast---AI rewarded,22ESG and climate penalized---makes a powerful case that the post-ChatGPT re-rating reflects a thematic23rotation \emph{toward AI}, not a generic disclosure effect.2425\input{tables/tab_placebo}2627\begin{figure}[htbp]\centering28\includegraphics[width=0.92\textwidth]{fig7_placebo.png}29\caption{Placebo test: post-ChatGPT high-minus-low return spread for each of the eight semantic30factors. AI-related factors (red) cluster at the top; ESG and climate disclosure earn negative31spreads.}\label{fig:placebo}32\end{figure}3334\subsection{Event study around ChatGPT}35The portfolio results use annual formation and forward returns; the event study provides a36higher-frequency, cleanly identified complement. Table~\ref{tab:event} and Figure~\ref{fig:event}37report market-model cumulative abnormal returns by pre-event AI tercile. The ordering is monotone at38every horizon and widens steadily over time. The High-minus-Low CAR difference is $+2.9$ points at 6039trading days \tstat{1.77}, $+10.1$ points at one year \tstat{1.97}, and $+14.6$ points at eighteen40months \tstat{1.89}. Because these are abnormal returns---purged of each firm's market beta---the41spread cannot be attributed to high-AI firms simply being higher-beta in a recovering market. A42cross-sectional regression of one-year CAR on pre-event AI exposure yields $+3.1$ points per standard43deviation \tstat{1.39}; adding sector fixed effects halves it to $+2.1$ points \tstat{0.99}, again44locating much---but not all---of the effect in a sector channel. Figure~\ref{fig:eventbin} shows the45relation is smooth across the exposure distribution, not driven by a few outliers.4647\input{tables/tab_event}4849\begin{figure}[htbp]\centering50\includegraphics[width=\textwidth]{fig8_event_car.png}51\caption{ChatGPT event study. Left: mean cumulative abnormal return (market model) by pre-event AI52tercile. Right: the High$-$Low CAR spread.}\label{fig:event}53\end{figure}5455\begin{figure}[htbp]\centering56\includegraphics[width=0.66\textwidth]{fig13_event_binscatter.png}57\caption{One-year cumulative abnormal return after ChatGPT versus pre-event AI exposure (binned58scatter with fitted line).}\label{fig:eventbin}59\end{figure}6061\subsection{Difference-in-differences and the cross-sectional nature of the effect (H4)}62Table~\ref{tab:did} reports difference-in-differences estimates in which treatment is the firm's63pre-determined (2019--2021) AI exposure interacted with the post-2022 indicator, within firm and year64fixed effects. The continuous-exposure DiD coefficient is $+2.7$ percentage points per standard65deviation \tstat{1.51}; the binary version is $+1.7$ points \tstat{0.78}. These are positive and66economically meaningful but weaker than the cross-sectional estimates---exactly the pattern predicted67by H4. Once firm fixed effects absorb the permanent component of AI exposure, only the residual68within-firm variation identifies the effect, and that variation is small. The event-time profile in69Figure~\ref{fig:eventtime} reinforces this reading: the return premium on a firm's \emph{fixed}70pre-event AI exposure is flat and near zero through 2021 and jumps after 2022, peaking in 2024. The71market did not reward firms for adding AI sentences to their filings; it repriced the firms whose72pre-existing, persistent exposure made them AI winners once the narrative turned.7374\input{tables/tab_did}7576\begin{figure}[htbp]\centering77\includegraphics[width=0.82\textwidth]{fig11_eventtime.png}78\caption{Event-study coefficients: the return premium on pre-determined (2019--2021) AI exposure by79year (95\% confidence intervals), relative to 2021. The premium activates after ChatGPT.}80\label{fig:eventtime}81\end{figure}8283\subsection{Construction, horizon, and universe robustness}84Table~\ref{tab:robust} collects the remaining checks. The post-2022 spread is essentially unchanged85whether the signal is the section-weighted mean (base, $+14.6$), the simple mean ($+13.4$), or the86top-5\% tail ($+14.9$). It holds at six-, twelve-, and twenty-four-month horizons (growing with the87horizon, as a persistent premium should). It survives replacing the three AI signals with their first88principal component---which alone explains 86.8\% of their common variance, confirming the ``single89latent AI construct'' interpretation---yielding a post-2022 PC1 spread of $+13.4$ points \tstat{1.97}.90Most importantly for the mega-cap concern, the spread is $+14.1$ points \tstat{1.92} when the91Magnificent~7 are excluded and $+14.5$ points \tstat{1.93} when the ten largest firms are dropped92(Figure~\ref{fig:mag7}). The AI re-rating is a broad-based feature of the index, not the mechanical93footprint of a handful of trillion-dollar winners. Finally, the sector-neutral composite---which94compares firms only to others \emph{within} their GICS sector---produces a much smaller and95insignificant post-2022 spread ($+3.8$ points, $t=0.84$). This is not a failure of the result but a96sharpening of it: a large share of the AI premium operates through a \emph{between-sector} channel, as97the market re-rated AI-exposed sectors (technology above all) relative to others, with a smaller98within-sector residual.99100\input{tables/tab_robust}101102\begin{figure}[htbp]\centering103\includegraphics[width=0.78\textwidth]{fig10_mag7.png}104\caption{Robustness of the AI return spread to excluding mega-cap winners. The post-2022 composite105spread is stable when the Magnificent~7 or the ten largest firms are removed.}\label{fig:mag7}106\end{figure}107108\subsection{An alternative performance outcome: risk}109Performance is not only about mean returns. We also relate AI disclosure to subsequent realized110volatility. High-AI firms exhibit modestly higher forward volatility, and the gap widens after 2022 (a111Q5$-$Q1 difference of about 2.9 annualized volatility points, $t=3.1$). The post-2022 AI premium112therefore came bundled with somewhat higher risk---part, though by no means all, of the elevated113returns is compensation for the greater volatility of AI-exposed firms during a period of rapid114technological repricing.115