Stock Market Responses to Economy-Wide Events

Economy-wide events are shocks that hit the whole market at once: a central-bank rate decision, a national election, a war or geopolitical crisis, a pandemic, or a sweeping regulatory or tax change. The event study is the right tool for measuring their effect on equity prices, but this event type carries a defining methodological twist that distinguishes it from firm-specific applications such as earnings announcements or mergers and acquisitions. Because the shock moves the entire market, the domestic market index you would normally use as the expected-return benchmark is itself contaminated by the event, so a naive market-model abnormal return mechanically nets out the very effect you set out to measure. Credible designs solve this in one of two ways: they isolate the unexpected component of the shock with a high-frequency surprise measure, or they shift from the average market reaction to the cross-section of who reacts more and why, benchmarked against something outside the shock (a cross-country index, a sector portfolio, or factor-mimicking portfolios). This page lays out the literature, the stylized facts with magnitudes, two worked numeric examples (a first-stage surprise and a second-stage cross-section), the event-type-specific method, the common pitfalls, and how to run it with our tools.

The intuition is worth fixing before the formulas. Trying to measure how much one stock was moved by a market-wide event using the home index as the benchmark is like trying to measure how high one swimmer was lifted by a wave by comparing them with the average swimmer in the same wave: everyone rose together, so the comparison tells you almost nothing about who was lifted most. To learn anything you must compare against still water (an unaffected foreign or factor benchmark) or measure the size of the wave itself (the policy surprise). That single choice, surprise versus cross-sectional benchmark, organizes the entire use case.

This is not a niche concern. Savor and Wilson (2013) show that the average US market excess return is about 11.4 basis points on the roughly 13% of trading days that carry a scheduled macro announcement (FOMC, CPI, employment) versus about 1.1 basis points on all other days (1958 to 2009), so more than 60% of the entire equity risk premium is earned on macro-announcement days. The economy-wide event is where the risk premium actually accrues, which is why getting its measurement right matters.

The range of events studied under this heading is wide, from natural disasters and economic collapse to the appointment of key personnel in politics. One long-running strand, the value of political connections in times of crisis or regulatory change, is treated as a sub-case below. The methodological core, shared across monetary policy, elections, geopolitics, pandemics and regulation, is what makes the use case coherent. It is also a refinement of the founding method: the first event study, Fama, Fisher, Jensen and Roll (1969), already benchmarked returns against the market, so the economy-wide problem is exactly what happens when the market benchmark itself is the thing being shocked.

What the research shows

The empirical literature delivers a consistent message: for an economy-wide event the average index move is often the least informative number, and the action is in the surprise component and in the cross-section of firm responses. The headline results below give signs, magnitudes and the canonical sources. The cross-sectional designs (Trump-tax, COVID-leverage, war-proximity) have replicated well; the average magnitudes (the pre-FOMC drift, the Bernanke-Kuttner coefficient) are sample-period-dependent and several have decayed or been reinterpreted, which is exactly the meta-narrative the table below is built to make scannable. The theoretical reason the cross-section dominates the mean is supplied by Pastor and Veronesi (2012, 2013): in their model the average stock-price reaction to a policy change is negative, larger when prior policy uncertainty is high and the economy is weak, and political uncertainty commands a non-diversifiable risk premium because policy hits all firms at once. That is precisely why the informative signal lives in the dispersion of who is exposed, not in the index average.

Stylized facts at a glance

Event / signal	Seminal paper	Sign	Magnitude (labelled)	Window
FOMC rate surprise	Bernanke-Kuttner 2005	+	~+1% per unanticipated 25bp cut; coefficient -4.68 per 1pp surprise, R² 0.17. Modern HF replication: -5.4% S&P 500 per 100bp surprise, t~8 (Bauer-Swanson 2023)1	1 day (or 30 min)
Pre-FOMC drift	Lucca-Moench 2015	+	+49bp (1994-2011); decayed to +9.2bp (2016-2019, Kurov et al. 2021)	24h before announcement
2016 US election tax cross-section	Wagner-Zeckhauser-Ziegler 2018	+	+0.41pp per +1 SD cash effective tax rate (SD 15.4; coeff 0.032)	1 day (Nov 9)
COVID Fever-phase leverage	Ramelli-Wagner 2020	−	~-3.06% per +1 SD leverage (coeff -0.135, SD 22.69pp); cash +2.99% per +1 SD (coeff +0.112)	Fever phase (Feb 24-Mar 20 2020)
Russia-Ukraine proximity	Federle et al. 2026	−	~-2.6pp per 1,000km closer to Ukraine; trade spillovers ~two-thirds of penalty	4-week window
Policy uncertainty (investment)	Gulen-Ion 2016	−	~-8.7% firm investment per doubling of EPU; ~one-third of the 2007-09 investment collapse	Following ~8 quarters
Policy uncertainty (priced factor)	Brogaard-Detzel 2015	−	-5.53%/yr high-minus-low EPU-beta portfolio, net of Carhart-4 (negative price of risk)	Cross-section, monthly

¹ The FOMC magnitude is the page's clearest decay-and-reinterpretation case. The 2005 Bernanke-Kuttner coefficient (-4.68 per 1pp) is confirmed and slightly enlarged by the modern high-frequency replication of Bauer-Swanson (2023), -5.4% per 100bp surprise, and corroborated by an orthogonal identification: Rigobon and Sack (2004) use identification through heteroskedasticity to find about -1.7% in the index (and -2.4% in the Nasdaq) per 25bp hike, so the magnitude is method-independent. But the interpretation is reversed by Nagel and Xu (2024), who attribute most of the reaction to the default-free yield curve rather than the equity premium. The number holds up; the channel is contested.

Monetary policy and FOMC announcements

The reference result is Bernanke and Kuttner (2005): an unanticipated 25 basis-point cut in the federal funds target raises broad US equity indices (the CRSP value-weighted index) by roughly 1% on the announcement day. The precise anchor is a surprise-regression coefficient of -4.68 (a -4.68% return per +1 percentage-point surprise rate cut), with the policy surprise explaining R² = 0.17 of event-day equity variance; dropping six outliers lowers the coefficient to -2.55. The crucial qualifier is that only the surprise moves prices. Kuttner (2001) decomposes a target-rate change into expected and unexpected components using the change in the fed funds futures rate, and shows that asset prices respond strongly to the surprise and barely at all to the fully anticipated part, which is already in prices. Bernanke and Kuttner attribute most of the equity reaction to news about the equity risk premium and future dividends rather than the real risk-free rate: in their Campbell-Ammer decomposition of the hypothetical surprise effect, just over half of the total response traces to revised expected future excess returns (the equity risk premium), with dividends contributing the next-largest share and the real interest rate under one percentage point. Gurkaynak, Sack and Swanson (2005) refine the identification further: FOMC effects need two factors, a current-rate target factor and a path (forward-guidance) factor tied to the statement, and the path factor drives much of the equity and bond response. This justifies separating the action (the rate decision) from the communication (the statement).

This result anchors a clear methodological lineage: from the founding market-adjusted design of Fama, Fisher, Jensen and Roll (1969), to the regulatory multi-date template of Schwert (1981), to the surprise decomposition of Kuttner (2001) and Bernanke and Kuttner (2005), and on to the modern cross-sectional designs of Wagner, Zeckhauser and Ziegler (2018), Ramelli and Wagner (2020) and Federle et al. (2026). Each step is a response to benchmark contamination as the events grow more market-wide.

The magnitude has held up and even grown in modern high-frequency replication: Bauer and Swanson (2023) estimate that a 100bp monetary-policy surprise moves the S&P 500 by about -5.4% (t-statistic near 8), roughly -1.35% per 25bp, slightly exceeding the Bernanke-Kuttner coefficient rather than contradicting it. But two qualifiers matter. First, the headline magnitude is not stable across regimes: the 1%-per-25bp result is a pre-2008 normal-times number, and at the zero lower bound the conventional short-rate channel is muted, with a policy-induced 100bp decline in the 10-year Treasury yield mapping to only about 1.5% to 3% higher equity. Second, the interpretation is contested: Nagel and Xu (2024), using a model-free dividend-futures counterfactual, attribute most of the equity reaction to the default-free yield curve rather than the equity premium, directly reversing the Bernanke-Kuttner channel. The Federal Reserve FEDS working paper 2026-023, The Effect of the Federal Reserve on the Stock Market, consolidates these magnitudes, channels and shocks across 1994 to 2019 and is a useful current meta-citation for the section.

Two caveats keep this area live. First, the pre-FOMC announcement drift: Lucca and Moench (2015) document an average US equity excess return of about 49 basis points in the 24 hours before scheduled FOMC meetings (1994 to 2011), historically around 80% of the annual equity premium, absent in Treasuries. The named replication closes the loop: Kurov, Wolfe and Gilbert (2021) show the drift held at 44.5bp through 2015 and then collapsed to 9.2bp over 2016-2019 (a Wilcoxon test rejects equality at the 1% level), and is effectively zero on days without a press conference, attributing the fade to a falling VIX. A second window-placement warning runs alongside it: Cieslak, Morse and Vissing-Jorgensen (2019) show that since 1994 the entire average equity premium accrues in even weeks of FOMC-cycle time (weeks 0, 2, 4, 6 after a meeting), with odd weeks earning roughly zero, so where you center an event window matters as much for a macro study as the surprise measure. Returns can accrue before a naive (0,+1) window opens, and the 49bp figure should not be over-applied to recent samples.

Second, the surprise itself is not exogenous, and the index reaction is not a single shock. Nakamura and Steinsson (2018) launched the modern debate with the high-frequency information effect: around FOMC announcements a surprise tightening raises real rates roughly one-for-one several years out and coincides with upward revisions to expected output growth, the opposite of the textbook channel, implying announcements move beliefs about fundamentals, not just policy. Their "policy news shock," the first principal component of 30-minute changes across fed-funds and Eurodollar futures out to about one year, is the multi-instrument modern upgrade to single-contract Kuttner scaling. Jarocinski and Karadi (2020) then supplied the practical disentangling rule, the "poor man's sign restriction": in the (rate surprise, S&P surprise) plane, a pure monetary-policy shock moves rates and stocks in opposite directions (rates up, stocks down), while a central-bank information shock moves them in the same direction (rates up, stocks up), so the sign of the high-frequency co-movement classifies each FOMC. Their updated shock series is publicly maintained (jkshocks). Building on this, Bauer and Swanson (2023) show high-frequency surprises are about 16% predictable (R²) from public pre-announcement data across 322 FOMC announcements (1988-2019), with strong payrolls, a strong stock market and high commodity prices forecasting hawkish surprises; their fix is to orthogonalize the raw surprise on six pre-announcement predictors and use the residual, which yields estimated macro effects up to four times larger and removes the price and activity puzzles, so residualization is magnitude-changing, not cosmetic. As a complementary structural channel, Cieslak and Vissing-Jorgensen (2021) show a large share of the FOMC-cycle equity premium reflects the Fed put, larger-than-expected accommodation after stock declines. The narrative therefore runs 2018 to 2023 to 2024: conclusions hinge on the benchmark and the decomposition, so report which channel your interpretation assumes.

Pandemics: the COVID-19 cross-section

The model pandemic study is Ramelli and Wagner (2020), which splits the COVID-19 shock into three phases (Incubation, Outbreak, Fever) rather than forcing it into a single date. Because the pandemic hit the whole market, the signal is extracted cross-sectionally.

Phase	Approx. dates (2020)	What the market priced	Implied benchmark / window
Incubation	Jan 2-17	China-trade and internationally exposed firms fell first	Foreign / peer benchmark; short windows
Outbreak	Jan 20-Feb 21	Broad risk-off begins; +1 SD China exposure ~-1.36%, +1 SD foreign revenue ~-1.25%	Cross-sectional on exposure
Fever	Feb 24-Mar 20	Leverage punished (~-3.06% per +1 SD), cash rewarded (~+2.99% per +1 SD)	Phase window net of beta; net out Fed bond intervention

In the Fever phase the cross-section is the story: a one-standard-deviation rise in leverage (coefficient -0.135, SD of leverage 22.69pp) maps to roughly 3.06% lower cumulative returns, net of beta, and the symmetric result holds for liquidity, a one-standard-deviation higher cash-to-assets ratio (coefficient +0.112, SD 25.80pp) maps to roughly +2.99% higher returns, as a health crisis became a feared financial crisis and financial resilience was repriced. A model-free companion isolates the policy response: Gormsen and Koijen (2020) read growth-expectation revisions directly off aggregate dividend-futures prices and bound them by horizon, finding the lower bound on expected GDP growth revised down by up to about 10% (US) and 12% (EU) by mid-March 2020, while fiscal-stimulus news around 24 March lifted the market and long-horizon growth but barely moved short-term growth expectations, neatly isolating the fiscal event from the disease shock. The extension matters too: Glossner, Matos, Ramelli and Wagner (2023) show stocks with higher and more-active institutional ownership performed worse in the Feb-Mar 2020 crash, as institutions fire-sold toward cash (not ESG), amplifying the crash, which extends the second stage from firm fundamentals to the investor base. Baker, Bloom, Davis, Kost, Sammon and Viratyosin (2020) add the long view: of 1,116 daily market jumps greater than 2.5% attributed by newspapers since 1900, not one traced to a prior infectious-disease outbreak, the 1918 influenza produced zero such jumps despite roughly 14 times the mortality, and COVID-19 produced 18 jumps in the 22 trading days from Feb 24 to Mar 24 2020 (the S&P 500 fell 33% to its Mar 23 trough, then rose 30% by end-April). The driver was government restrictions and distancing in a service economy, not lethality, and the result demonstrates newspaper or text-based event identification when there is no clean announcement date.

Elections and political shocks

The definitive election example is Wagner, Zeckhauser and Ziegler (2018) on the surprise 2016 US result. Because the election moved the whole index (the S&P 500 rose 4.64% from election to year-end), the authors regress firm-level abnormal returns on firm characteristics, with market-model betas estimated September 2015 to September 2016. The quotable magnitude: a one-standard-deviation higher cash effective tax rate (SD = 15.4) is associated with a +0.41 percentage-point abnormal return on the day after the election (Nov 9, coefficient 0.032), so high-tax firms gained on the anticipated cut from 35% to 21%, while firms rich in net-operating-loss deferred tax assets, and internationally exposed firms, lost on trade and tariff fears. The interesting result for an economy-wide event is the cross-section of responses, not the index move, and easily quantified consequences (statutory tax rates) are priced faster than complex ones (net foreign exposure), so price discovery is gradual even for high-salience events. The authors' own follow-up, Paths to Convergence (Wagner, Zeckhauser and Ziegler, 2018), quantifies that gradualism: the tax and foreign-revenue characteristics that drove first-day returns produced about three days of cross-sectional momentum, a brief reversal, then convergence. And the design generalizes out of sample: Ferriani, Gazzani and Taboga (2025) re-run the same firm-characteristic approach on the 2024 election and find Trump-aligned financials, energy and industrials earned significant positive abnormal returns even as aggregate uncertainty measures rose.

The deeper lesson is that the politically salient average is uninformative. Santa-Clara and Valkanov (2003), in the Presidential Puzzle, find the excess market return is about 9 percentage points per year higher (value-weighted, about 16 points equal-weighted) under Democratic than Republican presidents, yet there is no significant return move right around the election date itself: the gap is spread across the term. This is the page's whole thesis in one finding: a politically charged average can be large and economically meaningful while the event-date reaction is statistically silent, which is exactly why an economy-wide design studies the cross-section and the surprise rather than the headline average. Pastor and Veronesi (2012, 2013) supply the theory: the average reaction to a policy change is negative and larger under high prior uncertainty, and political uncertainty is a priced, non-diversifiable risk.

Policy uncertainty as a continuous economy-wide signal

Not every economy-wide shock has a date. For diffuse, slow-moving policy uncertainty the right regressor is a continuous intensity index rather than a 0/1 dummy, and the canonical real-effects result is Gulen and Ion (2016). Using the news-based Economic Policy Uncertainty index of Baker, Bloom and Davis (2016), they find that a doubling of policy uncertainty is associated with roughly an 8.7% decline in firm-level capital investment, with the 2007-2009 rise in uncertainty plausibly accounting for about one-third of the roughly 32% drop in aggregate corporate investment; the effect is strongest for firms with more irreversible investment and greater reliance on government spending. This is an investment, not a stock-return, study (Review of Financial Studies 29(3): 523-564, not a CAR finding), and it belongs here as the motivation for the continuous-intensity approach: it is the real-economy counterpart to the price-reaction studies and the channel that links these event studies to capital budgeting and the macro-timing work on our Investment Weather page. Crucially, policy uncertainty is not only a real-investment driver but a priced equity risk factor: Brogaard and Detzel (2015) show EPU carries a negative price of risk in the cross-section, with the highest-EPU-beta portfolio underperforming the lowest by about 5.53% per year net of the Carhart four factors, tying the continuous-intensity regressor directly back to the cross-section of returns. The Gulen-Ion result replicates internationally with sample dependence (more persistent, up to four years, in Australia; larger and election-driven in emerging markets; shorter-lived in the US).

The value of political connections (sub-case)

A related strand asks whether connections to political figures carry value, and increasingly finds that they do even in highly developed economies, not only where institutions are weak. Acemoglu, Johnson, Kermani, Kwak and Mitton (2016) studied the market reaction to the November 2008 nomination of Timothy Geithner as Treasury Secretary and identified an effect reminiscent of emerging-market connection value: financial firms personally connected to Geithner earned a cumulative abnormal return of about 6% after the first full trading day and about 12% after ten trading days, with negative abnormal returns when his confirmation was threatened by tax issues, robust to a large array of tests. Earlier, Johnson and Mitton (2003) argued that connection value operates across political contexts, showing that connected firms in Malaysia lost value as the Asian crisis eroded the expected value of their connections and then gained when capital controls in 1998 favoured firms tied to the prime minister. A formal 2021 meta-analysis of political connections and firm performance puts these anecdotes in context: the pooled effect is moderated by corruption, regulation intensity and development level and can flip sign between accounting and market measures, with event-study CARs in the literature ranging from about +1.36% (a close-election winner) to about -1.34% (a five-day connection-announcement window). This belongs under the economy-wide umbrella as one type of policy or appointment event, benchmarked and tested like the others, rather than as the whole of the use case.

Geopolitical and regulatory shocks

For wars and crises, broad markets fall and volatility rises on escalation while cooperation lifts them, and defense and aerospace stocks (the war stocks) show positive abnormal returns (Schneider and Troeger, 2006). The cleanest modern demonstration of the cross-country benchmark solution is Federle, Meier, Muller and Sehn (2026) on Russia's 2022 invasion of Ukraine: across 66 countries in a four-week window they identify a proximity penalty, where equity returns fall by about 2.6 percentage points more for every 1,000 kilometres closer a country, and within countries a firm, sits to the conflict, with trade-related spillovers explaining about two-thirds of the penalty. Because the war hit all markets, identification comes from the cross-section of distance to conflict, not from any single index. For a continuous event-intensity regressor, Caldara and Iacoviello (2022) construct the newspaper-based Geopolitical Risk (GPR) index, whose spikes predict lower near-term returns and higher volatility, and which together with the EPU index is one of the two off-the-shelf intensity indices practitioners plug in instead of dummies. For regulation, the classic template is Schwert (1981), with Binder (1985) as its seminal companion on multi-stage regulatory dates: track the abnormal returns of affected industry portfolios across the multi-stage legislative and regulatory process, accepting that reactions spread across many small dates rather than one clean announcement.

Who uses this in practice

This is not a purely academic exercise; large sums ride on getting the benchmark and the surprise right. As Savor and Wilson (2013) show, the majority of the equity risk premium is earned on the handful of days carrying scheduled macro news, so the economy-wide event is where the money is. Four practitioner communities run these event studies, and each maps to one of our calculators.

• Litigation economists. Securities-fraud class actions are the single biggest real-money application, and they depend on exactly the benchmark-contamination thesis of this page. Since Basic v. Levinson (1988) and Halliburton II (2014), a court-admissible event study must residualize out both a market index and an industry index, so that a market-wide macro move on the disclosure day is not misattributed to firm-specific fraud. The dollar stakes are concrete: NERA Economic Consulting's 2024 review reports median investor losses of $1.76 billion (a ten-year high) and aggregate settlements of $3.8 billion across 229 federal filings, and Cornerstone Research reports a 2025 median settlement of $17.3 million, near a three-decade high. NERA, Cornerstone, The Brattle Group and Fideres are the consultancies whose experts run these studies in court. The non-parametric test battery is legally load-bearing, not just statistically tidy: single-firm returns are non-normal so standard t-tests over-reject, which is why ARC's BMP and Kolari-Pynnonen tests are what survives a Daubert challenge (Fisch, Gelbach and Klick, 2018). On a disclosure day that coincides with an FOMC decision or a macro shock, an event study that controls only for the market (and not the industry) overstates the firm-specific abnormal return, the same contamination error this page is built around; see Litigation.

• Macro and event-driven asset managers and sell-side strategists. FOMC-day surprise trading, election and geopolitics positioning, and tactical reallocation all rest on the surprise-measurement and cross-sectional designs here. The two-stage cross-section is productized as tradable instruments: Goldman's domestic-versus-international-sales baskets diverged about 9 percentage points in the 2018 tariff war, and Morgan Stanley's tariff-exposed basket flipped on the November 2024 election, so increasingly the buy-side buys ready-made policy-sensitivity baskets and screens from sell-side research rather than rebuilding the second stage. This feeds Tactical Asset Allocation Signals.

• Corporate treasurers and CFOs. The same macro shocks feed capital-budgeting decisions: a doubling of policy uncertainty is associated with roughly an 8.7% decline in firm investment (Gulen and Ion 2016), the real-options wait-and-see channel that links these event studies to corporate finance.

• Central banks and regulators. The OCC, the Office of Financial Research, the IMF and the Fed both produce the shock datasets and consume event-study evidence on transmission and financial stability; the OCC and OFR are documented event-study-software users.

How to run this kind of event study

The general workflow (estimation window, expected-return model, event window, abnormal returns, significance testing) is covered in our Introduction to Event Study Methodology and the step-by-step Event Study Application Blueprint. Economy-wide events add five design choices that matter more here than anywhere else.

Benchmark choice: do not regress against the contaminated index

This is the defining problem. A market-wide event moves the very index you would normally use as the market-model benchmark, so estimated abnormal returns shrink toward zero as the benchmark absorbs the event. A two-number illustration makes it concrete: suppose a defense stock falls 8% on an invasion day while the home index falls 7%. Against the home index the abnormal return is only -1%, the event has been benchmarked away. Against an unaffected foreign index that fell 1%, the abnormal return is -7%. Same stock, same day; the benchmark decides what you measure. There are four escape routes. First, a cross-country or foreign benchmark (for example MSCI World or an unaffected-country index for a single-country shock), the Federle et al. proximity design. Second, a sector or peer benchmark to isolate the differential within-market effect. Third, factor benchmarks (Fama-French and momentum) so the common shock loads on the factors while you study the residual cross-sectional dispersion. Fourth, a placebo or synthetic-control benchmark: build a donor-weighted synthetic counterfactual from unaffected countries or sectors (or run placebo events on non-event dates) so the common shock is differenced out and the residual is the treatment effect; this is also what the 2025 tariff studies fell back on when they used a historical-mean rather than market model. Choose deliberately and state which one you used; see Expected Return Models. When the aggregate move genuinely is the object of study (the average market reaction to FOMC), no relative benchmark works, and you must instead measure the surprise, and even then the decomposition is contested (the same index reaction reads as equity-premium news under Bernanke-Kuttner or as yield-curve news under Nagel-Xu), so report the dividend-futures decomposition or at least state which channel your interpretation assumes.

Measure the surprise, not the headline event

For pre-scheduled releases (FOMC, ECB, CPI, payrolls, and elections with polls), decompose the announcement into expected and unexpected components using futures, forwards or poll-implied probabilities. The Kuttner (2001) method computes the surprise as the change in the spot-month fed funds futures rate, scaled by D/(D−d) where d is the announcement day and D the days in the month, with Kuttner's own published rule that when the meeting falls in the final roughly seven days of the month the scaling explodes the noise, so you switch to the unscaled next-month contract. Regressing returns on the raw policy change is mis-specified because anticipation is already priced. The modern multi-instrument upgrade to single-contract scaling is the Nakamura-Steinsson "policy news shock," the first principal component of 30-minute changes across fed-funds and Eurodollar futures out to about one year. The standard high-frequency convention is a 30-minute window from 10 minutes before to 20 minutes after the statement release.

Two refinements are now mandatory. First, disentangle policy from information shocks. The high-frequency surprise mixes a pure monetary-policy shock with a central-bank information shock, and Jarocinski and Karadi (2020) give a rule you can apply with just the rate surprise and the index move: if rates and stocks move in opposite directions (rates up, stocks down) the shock is pure policy; if they move together (rates up, stocks up) it is a Fed information shock, the central bank revealing news about fundamentals. This is the modern answer to the "is the index reaction policy or information?" question (Nakamura and Steinsson, 2018), and a maintained shock series (jkshocks) is public. Second, purge the predictable component: Bauer and Swanson (2023) recommend orthogonalizing the raw surprise on six pre-announcement predictors and using the residual, which removes the price and activity puzzles and yields effects up to four times larger, so residualization changes magnitudes rather than tidying them. Decompose further where relevant: separate the action from the communication (the two-factor target and path decomposition of Gurkaynak, Sack and Swanson), now that post-meeting press conferences have become the single most important source of policy news, which argues for separate windows around the statement and the press conference. Practitioners rarely rebuild surprises from scratch: ready-made series include the Bauer-Swanson San Francisco Fed monetary-policy surprises, the U.S. Monetary Policy Event-Study Database (USMPD, which also covers press conferences and minutes), the maintained Jarocinski-Karadi series, and a 2024 IMF dataset of 3,545 high-frequency monetary-policy shocks across 20 central banks (2000-2022), extending Kuttner (2001) and Nakamura-Steinsson. For continuous diffuse shocks the GPR and EPU indices, both distributed via Bloomberg, FRED, Haver and Reuters, are the off-the-shelf intensity regressors.

The cross-sectional second stage is the point

For most economy-wide events the right question is not whether mean CAR is significant (it can be zero while the cross-section is highly informative) but which firms react and why. Estimate firm-level abnormal or relative returns in stage one, then regress them on ex-ante characteristics (leverage, cash, foreign or China sales, effective tax rate, deferred tax position, distance to conflict, regulatory exposure, institutional ownership, beta and factor loadings) in stage two, as in the worked second-stage example above. This two-stage design is how Wagner, Zeckhauser and Ziegler, Ramelli and Wagner, and Federle et al. actually identify their effects. Set expectations on fit: cross-sectional event-study regressions routinely explain only 5% to 25% of the variation, so a low R² is normal and does not mean the characteristic is unpriced; the slope and its t-statistic are what matter, not the fit. The classic correctly-sized estimator under cross-sectional dependence is the portfolio-OLS approach of Sefcik and Thompson (1986), which is equivalent to forming characteristic-weighted portfolios and testing their in- versus out-of-window returns, and the modern, tooled complement is the placebo approach discussed next.

Event-date identification and window length

Match the window to the diffusion speed of the event. Monetary-policy surprises price within minutes, so a one-day or even 30-minute intraday window is appropriate, and the pre-FOMC drift and even-week FOMC-cycle pattern warn you to center windows carefully rather than defaulting to (0,+1). Pandemics, wars and regulatory processes are slow-burn, multi-date shocks with no single clean date: define event dates from a documented escalation timeline, treat each milestone as a separate event, use phase-based windows (the Ramelli-Wagner Incubation, Outbreak, Fever split), or use newspaper or narrative attribution of market jumps (Baker et al.). A continuous event-intensity index such as GPR or Economic Policy Uncertainty (Baker, Bloom and Davis, 2016) can replace a 0/1 dummy for diffuse shocks, and our News Analytics tool can reproduce a bespoke newspaper-tone index for a custom shock. Longer windows raise the benchmark-misspecification and confounding-event risk, so widen them only as far as the event genuinely requires.

Clustering, confounders and single-event inference

Every firm shares the same calendar date, so abnormal returns are cross-sectionally correlated and naive cross-sectional t-tests over-reject the null of zero average abnormal return, sometimes badly. The state-of-the-art treatment of exactly this problem is Cohn, Johnson, Liu and Wardlaw (2026): standard cross-sectional regressions of returns on firm characteristics around a single common event reject the null at the 1% level in more than 20% of non-event (placebo) periods, roughly twenty times the nominal size, and clustering standard errors yields only a modest reduction in over-rejection. Their recommended fix is to build the test distribution from the cross-sectional regression coefficients on pre-event (placebo) days and judge the event-day coefficient against that empirical distribution, implemented in the public Stata command csestudy; because the placebo days carry whatever else was going on, this also addresses same-day confounding news, not just clustering, making it the most complete single fix for the "every firm shares the date and the date has other news" problem this page is built around. Use it alongside the classic clustering-robust tools: the Boehmer, Musumeci and Poulsen standardized cross-sectional test adjusted for cross-sectional correlation per Kolari and Pynnonen (2010), the Kolari-Pynnonen rank test, the portfolio-OLS estimator of Sefcik and Thompson (1986), or a calendar-time portfolio (CTAR) approach, especially for multi-date shocks. Kothari and Warner (2007) flag event-date clustering as a dominant pitfall for this event type. The sharper, often-missed trap is degrees of freedom: with N firms reacting on a single common date, the effective number of independent observations is the number of events (one), not N, so pooling asset-event observations as if they were i.i.d. radically inflates power and t-statistics. The fix is an event-level block bootstrap (resample whole events, keeping all firms together) plus few-cluster-robust corrections (the wild-cluster bootstrap; MacKinnon-Webb 2018; Conley-Taber 2011); see Cameron and Miller (2015). Watch confounders too: economy-wide windows frequently overlap with other macro releases (jobs, CPI, simultaneous fiscal actions), and for COVID the Fed's bond-market intervention is itself a confounder that Ramelli and Wagner net out. Finally, a market-wide event has a near-universe sample, so survivorship, thin trading and beta non-stationarity matter, and betas jump in crises: estimate them in a pre-event window unaffected by the shock and consider shrinkage or time-varying betas. The full battery is in Significance Tests.

Run it with our tools

The applications on this site implement the workflow above end to end:

• Abnormal Return Calculator (ARC) is the core tool. Supply an event file keyed to the announcement or phase dates, choose your estimation and event windows (a tight one-day window for an FOMC surprise, multi-week phase windows for a pandemic or war), and pick an expected-return model. For economy-wide events the model choice is the benchmark-contamination decision: select a sector, factor (Fama-French 3-factor, 5-factor, Carhart 4-factor) or peer benchmark rather than the contaminated home index, then run the parametric and non-parametric tests built to handle clustered, common-date samples (BMP and Kolari-Pynnonen). Group firms by leverage, foreign exposure, tax position, institutional ownership or distance-to-conflict to reproduce the cross-sectional second stage. This is also the engine litigation experts use to defend market-plus-industry residualization, where the non-parametric tests are what survives a Daubert challenge.

• Abnormal Volume Calculator (AVC) detects when information actually arrived, which is invaluable for slow-burn events with no clean date and for confirming the volume and trading spike around FOMC announcements. Abnormal Volatility Calculator (AVyC) captures the volatility response, which rises sharply on geopolitical escalation and around macro releases.

• Event Date Identifier (EDI) helps pin down event dates and build the escalation timeline for diffuse shocks (pandemics, wars, multi-stage regulation) where the date is not given.

• News Analytics (CATA) lets you build a narrative or tone signal from policy statements, election coverage or crisis news, supporting the text-based event identification of Baker et al. and the action-versus-communication split for central-bank statements. It is the operational way to turn a 0/1 dummy into a continuous EPU- or GPR-style intensity regressor for a bespoke shock.

Frequently asked questions

How is an economy-wide event study different from a normal (firm-specific) one?

In a firm-specific study (an earnings release, a merger) the market index is a clean benchmark, because the event does not move the whole market, so a standard market-model abnormal return works. In an economy-wide study the benchmark is contaminated: the event moves the index itself, so you switch to a foreign, sector, factor or synthetic benchmark, or you measure the policy surprise directly. The object of study also shifts, from the average reaction to the cross-section of who reacts more and why. That double switch, clean-to-contaminated benchmark and average-to-cross-section, is the whole use case.

Why can't I use the S&P 500 as the benchmark for a market-wide event?

Because the event moves the S&P 500 itself, so a market-model abnormal return nets out the very effect you want to measure: the benchmark absorbs the shock and your abnormal returns shrink toward zero. For a market-wide event use a foreign or cross-country index, a sector or peer portfolio, factor-mimicking portfolios, or a synthetic control, none of which is contaminated by the domestic shock.

How do I measure the stock market reaction to an FOMC or interest-rate decision?

Measure the surprise, not the headline rate change. Decompose the decision into anticipated and unanticipated components using fed funds futures (Kuttner 2001), apply it in a tight window (one day or the 30-minute high-frequency window), and remember that only the surprise moves prices. Bernanke and Kuttner (2005) find about a 1% index move per unanticipated 25bp cut, and the modern high-frequency replication of Bauer and Swanson (2023) finds about -5.4% per 100bp surprise, a magnitude weaker at the zero lower bound and whose channel (equity premium versus yield curve) is contested.

Why did stocks fall when the Fed cut rates?

Because the sign of the surprise is necessary but not sufficient. A surprise cut can carry a Fed information effect: the cut signals that the central bank has bad news about growth, so equities can fall even on a dovish surprise (Nakamura and Steinsson 2018). The practical test is the high-frequency co-movement: if rates fell and stocks fell together, that is the information-shock quadrant (Jarocinski and Karadi 2020), not a pure easing. Whether this effect is genuine or an artifact of predictable surprises is contested (Bauer and Swanson 2023), so classify the shock before interpreting the move.

What is a monetary-policy surprise and how is it calculated?

It is the unanticipated part of a rate decision, computed as the change in the spot-month fed funds futures rate on the announcement day, scaled by D/(D−d) where d is the meeting day and D the days in the month. The scaling undoes the contract's monthly-averaging convention, and Kuttner's own rule is to use the unscaled next-month contract when the meeting falls in the final roughly seven days of the month. If 40bp of a 50bp cut was already priced, only the 10bp surprise is news. Ready-made surprise series (SF Fed, USMPD, the Jarocinski-Karadi series, the 2024 IMF dataset) save you from rebuilding them.

Does the stock market do better under Democrats or Republicans?

Historically better under Democrats: Santa-Clara and Valkanov (2003) find the excess market return is about 9 percentage points per year higher (value-weighted, about 16 equal-weighted) under Democratic than Republican presidents. The crucial event-study lesson, though, is that there is no significant return move right around the election date itself: the gap is spread over the term. That is the canonical illustration of why an economy-wide event study examines the cross-section and the surprise rather than the headline average, which can be large yet statistically silent at the event date.

How do you do an event study when there is no single event date (a pandemic or war)?

Treat it as a multi-date or phased event. Build an escalation timeline, treat each milestone as a separate event, use phase-based windows (Ramelli and Wagner's Incubation, Outbreak, Fever split), or attribute market jumps from newspaper text (Baker et al.). For diffuse policy uncertainty, replace the dummy with a continuous intensity index such as EPU or GPR.

Can the average abnormal return be zero but the event still matter?

Yes, and for economy-wide events that is usually the point. The average index move can be uninformative while the cross-section of who reacts more is highly informative. The mean CAR across high-tax and low-tax firms can net to roughly nothing even though the tax characteristic is strongly priced, which is exactly why the cross-sectional second stage, not the average, is the object of study.

What event window should I use for a macro event?

Match the window to how fast the event prices in. Monetary-policy surprises price within minutes, so use a one-day or 30-minute window and watch the pre-FOMC drift and the even-week FOMC-cycle pattern. Pandemics, wars and regulatory processes are slow-burn, so use phase windows or multiple milestone dates. Longer windows raise benchmark-misspecification and confounding risk, so widen them only as far as the event genuinely requires.

How do I handle every firm sharing the same event date?

Abnormal returns are cross-sectionally correlated and the true degrees of freedom equal the number of events (one), not the number of firms, so naive t-tests over-reject badly: a 1% test can reject in more than 20% of non-event placebo periods (Cohn, Johnson, Liu and Wardlaw 2026). Use the BMP test adjusted for cross-sectional correlation (Kolari-Pynnonen 2010), the rank test, the Sefcik-Thompson portfolio-OLS, a pre-event placebo coefficient distribution (the csestudy command), a calendar-time portfolio, or an event-level block bootstrap that resamples whole events keeping all firms together (Cameron and Miller 2015).

Should I use a 0/1 dummy or a continuous index (EPU/GPR) for a diffuse shock?

Use a continuous intensity index when the shock has no clean date and arrives in waves, such as policy uncertainty or geopolitical risk. A 0/1 dummy forces a diffuse shock into a single date it does not have. The EPU index (Baker, Bloom and Davis 2016) and the GPR index (Caldara and Iacoviello 2022) are the canonical continuous regressors, Gulen and Ion (2016) is the classic real-effects application, and Brogaard and Detzel (2015) show EPU is also a priced equity risk factor.

Related use cases

Economy-wide events share the surprise-measurement and clustering challenges with Earnings Announcements and the cross-sectional benchmark question with Comparative Event-Type Analyses. The signals studied here feed directly into the macro-timing work on the Investment Weather and Investment Clock and Tactical Asset Allocation Signals, and into Investment Strategies more broadly. For policy and regulatory shocks affecting specific firms, see Litigation and Competitive Dynamics. For the full catalogue, return to the Practical Applications overview.