spb/phd_thesis Public

1\section*{Figures}2\addcontentsline{toc}{section}{Figures}34% Figure 1: Intraday Volatility Pattern5\begin{figure}[H]6\centering7\includegraphics[width=0.85\textwidth]{figures/figD3_intraday_vol_pattern.pdf}8\caption{Intraday volatility pattern on FOMC days. Red dashed line marks the 14:00~ET announcement. This figure documents the characteristic volatility spike at announcement time, motivating our high-frequency identification strategy. The pre-announcement period shows relatively stable volatility, while the post-announcement spike and gradual decay are consistent with information processing models.}9\label{fig:D3}10\end{figure}1112% Figure 2: Semantic Measures Scatter13\begin{figure}[H]14\centering15\includegraphics[width=0.75\textwidth]{figures/figD6_stance_novelty_scatter.pdf}16\caption{FOMC semantic measures: stance vs.\ novelty (z-scored). Points colored by stance tercile. The low correlation between stance and novelty ($r = 0.12$) supports the conditional independence assumption (Assumption~\ref{ass:indep}) and motivates the separate estimation of tone and novelty effects in our decomposition analysis.}17\label{fig:D6}18\end{figure}1920% Figure 3: Semantic Time Series21\begin{figure}[H]22\centering23\includegraphics[width=0.9\textwidth]{figures/figD7_semantic_timeseries.pdf}24\caption{Evolution of policy stance and novelty scores over time. Novelty peaks during the 2008 financial crisis and 2020 pandemic correspond to major policy regime changes. Stance shifts from dovish (2008--2015) to hawkish (2017--2019, 2022--2025) track well-known monetary policy cycles.}25\label{fig:D7}26\end{figure}2728% Figure 4: Heatmap of Event-Level Coefficients29\begin{figure}[H]30\centering31\includegraphics[width=0.85\textwidth]{figures/figR6a_rv_reg_heatmap_30min.pdf}32\caption{Heatmap of event-level $\Delta$RV regression coefficients by ticker. Darker shading indicates stronger effects. Stance effects are negative and significant across most contracts, while novelty effects are positive. The stance$\times$novelty interaction is concentrated in VIX and Treasury securities, suggesting that the joint impact of tone and information content operates primarily through the uncertainty channel.}33\label{fig:R6a}34\end{figure}3536% Figure 5: Cumulative Drift by Stance37\begin{figure}[H]38\centering39\includegraphics[width=0.85\textwidth]{figures/figR8_drift_by_stance.pdf}40\caption{Cumulative return drift by stance tercile. Dovish announcements generate positive equity drift that strengthens over 45 minutes, while hawkish announcements produce symmetric negative drift. The monotonic separation between terciles and gradual strengthening over time support H1c (tone effects increasing in magnitude) and the fundamental repricing interpretation.}41\label{fig:R8}42\end{figure}4344% Figure 6: IRF - Stance on Cumulative Returns45\begin{figure}[H]46\centering47\includegraphics[width=0.85\textwidth]{figures/figR10_irf_ret_h_stance.pdf}48\caption{Impulse response: cumulative return to policy stance (post-announcement, Newey--West SE). The equity (ES) response builds gradually over the 120-minute window, consistent with slow fundamental repricing by heterogeneous investors (H1c). The VIX response is negative and peaks within 30 minutes, consistent with directionally clear statements resolving uncertainty quickly.}49\label{fig:R10_stance}50\end{figure}5152% Figure 7: IRF - Abnormal Return to Stance x Novelty53\begin{figure}[H]54\centering55\includegraphics[width=0.85\textwidth]{figures/figR10_irf_abn_h_stance_x_novelty.pdf}56\caption{Impulse response: abnormal cumulative return to stance $\times$ novelty interaction. VIX shows persistent, significant effects from 5 to 120 minutes ($t = -5.06$ at peak), representing the most robust finding in our analysis. This confirms that the joint impact of tone and novelty on uncertainty is both economically large and statistically robust across the full post-announcement window.}57\label{fig:R10_abn_interaction}58\end{figure}5960% Figure 8: Placebo Tests61\begin{figure}[H]62\centering63\includegraphics[width=0.85\textwidth]{figures/figR10b_irf_placebo_stance.pdf}64\caption{Pre-announcement placebo: IRF for stance. All coefficients are near zero and statistically insignificant across all tickers and horizons, validating our event study design. The absence of pre-announcement effects rules out information leakage and confirms that our identification strategy successfully isolates the causal impact of FOMC statement content.}65\label{fig:placebo_stance}66\end{figure}6768% Figure 9: Multi-Method Robustness Heatmap69\begin{figure}[H]70\centering71\includegraphics[width=0.85\textwidth]{figures/figRob2_multimethod_heatmap.pdf}72\caption{Multi-method robustness heatmap: number of methods (out of 5) yielding $p < 0.10$ for each ticker--variable combination. Darker cells indicate more robust results. Stance effects on $\Delta$RV and $\Delta\log$RV are the most robust, with 3--4 methods confirming significance for ES, CL, and Treasury contracts. The interaction term is most robust for VIX, consistent with the uncertainty channel interpretation.}73\label{fig:rob_heatmap}74\end{figure}75