Tactical Asset Allocation (TAA) Signals

Tactical asset allocation (TAA) is the practice of tilting a portfolio away from its long-run strategic asset allocation (SAA) to exploit shorter-horizon shifts in capital-market expectations. The directions and conviction of those tilts come either from fundamental judgment (discretionary TAA) or from asset-class-level signals known to carry predictive content for some period of time (systematic TAA). Event studies are the right tool for building the systematic kind, because the signals that move asset classes are overwhelmingly events: scheduled macroeconomic releases, monetary-policy decisions, and sector-relevant news. An event study measures the abnormal return, volume, or volatility an asset earns around a dated event, isolating the market's response from its normal behavior. Estimate that response asset by asset and sector by sector, condition it on the macro regime, and aggregate it across a rolling history, and you have a disciplined, cost-aware engine for allocation and sector-rotation tilts rather than a mechanical calendar trade.

The empirical case is unusually strong. A large, recurring fraction of the equity risk premium is earned in narrow windows around a small set of pre-scheduled events, the same events an event study is designed to measure. This page sets out what the research shows, how to run an event study for this purpose, and how to do it with the calculators on this site. It keeps two honest caveats front and centre: abnormal-return signals decay once they are known and published, and naive business-cycle sector rotation underperforms its reputation net of costs.

The one-sentence intuition. Markets are roughly efficient, so only the unexpected part of an event moves prices, and because the biggest scheduled surprises (FOMC decisions, the jobs report, CPI) arrive on a known calendar, the compensation for bearing macro risk piles up on those few dates. Savor and Wilson (2013) make the hook concrete: more than 60% of the cumulative equity premium is earned on the roughly 13% of days that carry a major macro release. An event study is simply the measuring tape for that compensation, sector by sector, and a TAA signal is what you build once you have measured it.

What the research shows

Four robust facts define this literature: the equity premium is concentrated on scheduled-event days, it is the surprise rather than the level that moves prices, the cross-section of sector responses is systematic, and the sign of the response is regime-dependent. A fifth fact is a warning: these signals are time-varying and can decay. The table below collects the canonical magnitudes; the sections that follow unpack them.

Table 1: Canonical empirical findings (study, event, window, magnitude, sample)

Study	Event / measure	Window	Headline magnitude	Sample
Savor and Wilson (2013)	Scheduled macro releases (FOMC, employment, CPI, PPI)	Announcement day	11.4 bp vs 1.1 bp excess return; >60% of premium on ~13% of days; Sharpe ~10x	1958-2009
Savor and Wilson (2014)	Security market line (beta pricing) on news days	Announcement day	SML slope 6.81 bp per unit beta vs ~0 on quiet days	1964-2011
Lucca and Moench (2015)	Pre-FOMC announcement drift	24h pre-announcement	+49 bp; ~80% of post-1994 premium; Sharpe ~1.14	1994-2011
Cieslak, Morse and Vissing-Jorgensen (2019)	FOMC-cycle even weeks	Bi-weekly calendar	Entire equity premium in even weeks (0,2,4,6); ~0 in odd weeks	1994-2016
Ai and Bansal (2018)	Macroeconomic announcement premium (FOMC + employment)	Announcement day	~55% of equity premium; requires non-expected-utility preferences	1961-2014
Ai, Bansal and Guo (2024)	Announcement premium, longer-sample restatement	~44 announcement days/yr	>71% of aggregate equity risk compensation	1961-2023
Bernanke and Kuttner (2005)	Unanticipated FOMC rate change (index)	1-day	~1% per 25 bp surprise (~5.4% per 100 bp); risk-premium channel	1989-2002
Ehrmann and Fratzscher (2004)	Unanticipated tightening, sector cross-section	1-day	50 bp tightening = ~3% lower broad returns; cyclicals react ~2-3x defensives	1994-2003
Hu, Pan, Wang and Zhu (2022)	Pre-announcement drift beyond FOMC (NFP, GDP, ISM)	Overnight pre-release	NFP +10.1 bp, GDP +9.6 bp, ISM +9.14 bp; concentrated on high-VIX-drop days	1997-2019
Knox, Londono and Samadi (2025)	Options-implied (ex-ante) event premia	Daily-expiry options	CPI ~45 bp, Employment ~53 bp, FOMC ~49 bp annualized; large cross-country spillovers	Sep 2023-Jul 2025

The premium is concentrated on announcement days

Savor and Wilson (2013) show that the average excess stock return is 11.4 basis points on scheduled-announcement days (FOMC decisions, the employment report, CPI and PPI) versus 1.1 basis points on all other days over 1958 to 2009, so more than 60% of the cumulative equity premium is earned on the roughly 13% of days that carry a major macro release, at a Sharpe ratio about ten times higher than non-announcement days. Lucca and Moench (2015) sharpen this to the policy event itself: the S&P 500 rose about 49 basis points on average in the 24 hours before scheduled FOMC announcements over 1994 to 2011, accounting for roughly 80% of the post-1994 equity premium, with a buy-the-day-before strategy earning an annualized Sharpe near 1.14. Crucially, the pre-FOMC drift is present in equities but absent in US Treasuries and the dollar, so the signal is asset-class specific. Cieslak, Morse and Vissing-Jorgensen (2019) document a deterministic event-clock pattern: since 1994 the entire equity premium accrues in even weeks of FOMC-cycle time (weeks 0, 2, 4 and 6 from the last meeting) and is roughly zero in odd weeks, a tactical risk-on/risk-off rule that needs only a calendar and no forecasting.

The pre-announcement drift is not unique to the FOMC. Hu, Pan, Wang and Zhu (2022) document large positive overnight returns with no abnormal variance ahead of nonfarm payrolls (+10.1 bp), GDP (+9.6 bp) and the ISM survey (+9.14 bp), comparable to the pre-FOMC effect. Their mechanism is a premium for heightened uncertainty: the market's uncertainty about the impending news' impact builds up and then resolves before the release, so pre-announcement returns concentrate on the days when implied volatility falls most (102.71 bp on high-variance-reduction days versus 79.54 bp on low). The practical upshot is that the event clock has more hands than just FOMC dates, and a serious TAA calendar should track all of them.

Ai and Bansal (2018) formalize this as a macroeconomic announcement premium, about 55% of the equity premium concentrated specifically in employment-report and FOMC days, and prove that its existence requires non-expected-utility preferences (generalized risk sensitivity of the Epstein-Zin or robust-control type). This is the theoretical microfoundation for treating announcement abnormal returns as compensated, priced risk rather than noise. The most current restatement is Ai, Bansal and Guo (2024), who extend the sample to 1961-2023 and find that roughly 44 announcement days per year (FOMC, employment, CPI, PPI, GDP) carry more than 71% of aggregate equity-market risk compensation, a sharper and fresher figure than the original 60% (1958-2009) and 55% numbers. They also stress that the premium is conditional on the information environment, not a constant, which motivates the regime conditioning below. Savor and Wilson (2014) add the cross-sectional counterpart: market beta is strongly priced on announcement days but essentially unpriced on quiet days. The load-bearing number is a security-market-line slope of 6.81 basis points per unit of beta on announcement days versus statistically indistinguishable from zero on other days, so the CAPM "works" only on the roughly 13% of days that carry macro news. The implication for TAA is direct: high-beta and cyclical tilts should be timed to the event clock.

It is the surprise, not the level, that moves prices

The event in a macro or policy study is the unexpected component, measured against a consensus survey or a futures-implied expectation, not the raw release. Kuttner (2001) introduced the now-standard decomposition of policy actions into expected and surprise components using fed-funds futures, showing that only the surprise component moves asset prices. Bernanke and Kuttner (2005) apply this to equities: an unanticipated 25 basis-point rate cut raises broad indices about 1% on average, and the reaction works mainly through the equity-risk-premium channel (future expected excess returns), not real rates or expected dividends. Scaling that estimate linearly gives the headline figure used throughout this page, roughly 4 x 1% = 5.4% per 100 basis points of surprise (the calculation is approximate because the response is not perfectly linear, but it is anchored directly in Bernanke and Kuttner's per-25-bp coefficient, not asserted independently).

Raw surprises are not exogenous, and separating a pure policy shock from an information shock is the central identification problem. The canonical decomposition is Jarocinski and Karadi (2020), who use the sign of the high-frequency stock/interest-rate co-movement around an announcement: a pure policy tightening raises rates and lowers stocks, while a positive central-bank information shock raises both. Ignoring the information shock biases inference on policy non-neutrality, which is why purified surprises matter for TAA. Nakamura and Steinsson (2018) ground the standard 30-minute identification window and document the "Fed information effect" directly: output-growth forecasts rise after a contractionary policy surprise, the opposite of the textbook sign, because announcements reveal the Fed's private read on the economy. The modern decomposition of the surprise itself is Swanson (2021), who extends the older two-factor model to three orthogonal factors per FOMC announcement (1991-2019): a federal-funds-rate factor, a forward-guidance factor, and a large-scale-asset-purchase (LSAP/QE) factor. Equities load most on the funds-rate factor; forward guidance moves short-term yields most; LSAPs move long-term Treasury and corporate yields most, so the LSAP factor is what drove the largest long-duration sector dispersion during 2008-2021. The originating two-factor study, Gurkaynak, Sack and Swanson (2005), identified the target and path factors but found the path (forward-guidance) factor mainly in long-term Treasury yields, with a weaker and less robust equity effect, so the precise claim is that forward guidance and asset purchases are increasingly important channels for equities alongside the funds-rate factor, not that equities respond "far more" to the path factor. Andersen, Bollerslev, Diebold and Vega (2003) establish the surprise-standardization framework for macro data (actual minus consensus, scaled by the historical standard deviation of the surprise) and document a sign-asymmetry effect, with bad news moving prices more than good news of equal magnitude.

The cross-section of sector responses is systematic

Sector responses to policy and growth surprises are systematic and sign-consistent. The dedicated sector study is Ehrmann and Fratzscher (2004): a 50 basis-point surprise tightening lowers broad equity returns by about 3% on average, the reaction is larger when no change was expected and when policy direction reverses, and the cross-section is sharp. Cyclical and capital-intensive industries (technology, communications, consumer durables) react two to three times more than defensives, and firms that are financially constrained (low cash flow, small size, poor credit ratings, low debt-to-capital, high price-earnings, high Tobin's q) respond significantly more to the same surprise. Bernanke and Kuttner (2005) report the same direction across sectors at the index level. This dispersion is the raw material for surprise-driven sector rotation: estimate each sector's abnormal-return sensitivity to a standardized surprise, and the betas become tilt weights. The cash-flow versus discount-rate split refines this further, since firms with floating-rate or short-duration debt and high operating leverage carry a distinct cash-flow sensitivity to rate surprises, separate from pure long-duration discount-rate names.

As a starting prior, the table below maps the broad sectors to their typical sign and magnitude of response to a hawkish (tightening) surprise. Treat it as a hypothesis to re-estimate on your own sample and regime, not as fixed truth, since the next section shows the sign itself can flip with the cycle.

Table 2: Sector sensitivity prior (response to a hawkish surprise, to be re-estimated)

Sector group	Examples	Dominant channel	Typical surprise beta (hawkish surprise)
Cyclical / rate-sensitive	Technology, consumer discretionary, communications	Long-duration discount-rate	Large negative (high beta)
Financials / real estate	Banks, REITs, homebuilders	Mixed (margin vs duration)	Large, sign mixed by sub-industry
Defensive	Consumer staples, utilities, healthcare	Stable cash-flow	Small negative (low beta)
Energy / materials	Oil and gas, mining	Commodity / growth	Small, often dominated by the growth read

The response sign is regime-dependent

The single most important interpretive rule for TAA is that the same macro surprise can flip sign with the business-cycle regime. Boyd, Hu and Jagannathan (2005) show that rising-unemployment news raises stocks in expansions (the lower-rate, discount-rate channel dominates) but lowers them in contractions (the earnings channel dominates), with the effect strongest in cyclical stocks. The cleanest single illustration is their August 3, 1984 episode: the unemployment rate jumped from 7.2% to 7.5%, unambiguously bad news for the real economy, yet the S&P 500 rose about 5.4% around the announcement because, in that expansion, the market read it as relief on the rate path. This is the substance behind "regime awareness": a pooled, regime-unconditional beta is biased toward zero and can have the wrong sign for the state you are actually in. A defensible TAA signal therefore interacts every surprise with the cycle state (an NBER recession indicator, the yield-curve slope, or a recent-macro-momentum gauge). Consistent with conditionability, Lucca and Moench (2015) find pre-announcement abnormal returns are larger when the yield-curve slope is low, implied volatility is high, and recent pre-event returns were high: the event premium is time-varying, not constant.

Recency: ex-ante premia, the 2022-23 inflation regime, and global spillovers

The 2013-era literature emphasized realized abnormal returns on FOMC and employment days. Two recent developments update that picture. First, the event premium can now be priced ex ante. Knox, Londono and Samadi (2025) use daily-expiry index options to back out forward-looking, options-implied premia: over September 2023 to July 2025 the annualized event premium ran about 45 bp for US CPI, 53 bp for US Employment, 49 bp for FOMC, and 28 bp for the euro-area policy rate. This is a different object from the realized CAR that the Abnormal Return Calculator computes after the fact, and the two are complementary: realized abnormal returns tell you how the market reacted, options-implied premia tell you how much risk the market is pricing into an event before it happens. They also document large cross-country spillovers (US employment is priced in euro-area markets, euro-area CPI is priced in US markets), so a serious TAA event calendar and surprise-beta panel should be multi-region, not US-only.

Second, the premium is demonstrably time-varying, and the 2022-23 inflation surge is the cleanest recent illustration of the page's own "can decay, can spike" thesis. As the Fed hiked from March 2022 into 2023 and the S&P 500 fell more than 20% into its trough, stock-market sensitivity to inflation surprises rose sharply and CPI-release days came to rival or even eclipse FOMC days in priced risk. The options-implied CPI and FOMC premia were highly elevated from mid-2022 to mid-2023 and then waned as inflation moderated, with the March 2022 FOMC carrying the single largest options-implied premium of the recent sample. The lesson for a TAA engine is to condition surprise betas on the inflation regime and to expect the relative salience of CPI versus FOMC versus payrolls to rotate over time.

Decay, confounders, and the sector-rotation myth

An honest authority page must state the limits. First, abnormal-return signals decay once published, but the decay is anomaly-specific and asset-class-specific rather than uniform. Kurov, Wolfe and Gilbert (2021) extend the sample through 2019 and find the pre-FOMC drift essentially vanished after about 2011 to 2015, a textbook case of an anomaly arbitraged away after it became known; a 2024 follow-up in the same line finds the drift faded in equities but left residual, detectable footprints in volatility markets, so "arbitraged away" is too absolute for all asset classes. Any drift-style calendar signal needs an out-of-sample monitor that retires it when its abnormal return collapses, and the monitor should be run per signal and per asset class. Second, high-frequency policy surprises are not cleanly exogenous: Bauer and Swanson (2023) show raw surprises are predictable from publicly available pre-announcement macro data with an R-squared of 10% to 40% (a "Fed response to news" channel), so surprises must be orthogonalized against prior macro and financial data before they are treated as exogenous signals. Importantly, after orthogonalization the high-frequency asset-price effects are largely unchanged and there is little evidence of a stock-market information effect, so the confounder to neutralize is the Fed's response to news, not a genuine equity information effect. Third, naive business-cycle sector rotation is weaker than its reputation: Molchanov and Stangl (2024) find no robust evidence of systematic sector outperformance at the cycle phases conventional quadrant charts predict, with any gains vanishing after transaction costs and realistic cycle-timing error.

The constructive counterweight to this decay-heavy tone is twofold. The pre-FOMC drift has a tradable, path-conditioned cousin: Neuhierl and Weber (2019) build a "monetary-momentum" slope factor from changes in fed-funds futures across horizons and show that stocks drift up roughly 25 days before expansionary FOMC surprises and down before contractionary ones, with a cumulative expansionary-minus-contractionary spread reaching about 2.5% by the meeting day and more than 4.5% fifteen days after, concentrated in high-uncertainty periods. And the aggregation template that survives all of these critiques is macro-momentum or economic-trend investing (Brooks et al., 2017), which aggregates persistent fundamental trends across growth, inflation, monetary policy, and risk sentiment into asset and sector tilts; AQR's 2024 "Economic Trend" update confirms the approach has retained durable, low-correlation, positive returns across growth, inflation, policy and risk-sentiment regimes over more than fifty years, so the surviving signal is the slow aggregation of trends, not the fast calendar trade.

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. Macro, policy, and sector-news studies add several event-type-specific design choices that matter a great deal in practice.

Event-date identification: scheduled versus unscheduled, and which releases to use

TAA-relevant events are pre-scheduled (FOMC dates, the BLS employment and CPI release calendar), so use the official release timestamp, not the calendar day, and separate scheduled releases from unscheduled inter-meeting policy actions, which behave differently. Day-0 timing is acute: a release after the market close should be coded to the next trading day. Event selection matters as much as event dating: not every macro release is priced. Flannery and Protopapadakis (2002) screen 17 macro series in a GARCH model of daily returns and find only six are robust priced factors for the return mean (CPI, PPI, a monetary aggregate, the balance of trade, the employment report, and housing starts), while headline activity measures such as industrial production and GNP/GDP are not priced for returns (several series move conditional volatility without moving the mean). A disciplined TAA event set should therefore pre-screen releases for demonstrated abnormal-return content rather than throwing every release into the panel. Pinning down the exact event date and reconciling release timestamps is exactly what our Event Date Identifier (EDI) is built for.

Window choice scales with event sharpness

For monetary policy and high-frequency macro releases, use narrow intraday windows: the standard, grounded in Nakamura and Steinsson (2018), is a roughly 30-minute window (about -10 minutes to +20 minutes around the announcement) to isolate the news and exclude confounders. Central-bank events carry a further trap: they have multiple intraday windows that must be measured separately. Altavilla et al. (2019) show that the policy-decision-release window and the press-conference window carry different factors (target, forward guidance, QE, and a post-2009 sovereign-spread factor), so collapsing both into one daily abnormal return contaminates the signal; their euro-area monetary-policy-event database operationalizes the split, and the same statement-then-press-conference structure now applies to the Fed. For daily-data event studies a (-1,+1) or (0,+1) day window is typical, and pre-FOMC-drift work specifically uses the 24-hour pre-announcement window. State the window explicitly, because it determines what the measured abnormal return actually captures.

Extract the surprise, decompose it, and orthogonalize it

Only the unexpected component moves prices, so compute the surprise from fed-funds futures for policy (Kuttner, 2001) or as (actual minus consensus) scaled by the historical standard deviation of the surprise for macro data (Andersen et al., 2003), so coefficients are comparable across release types. For FOMC events, use the three-factor decomposition of Swanson (2021) into a funds-rate factor, a forward-guidance factor, and an LSAP/QE factor (the modern standard, with the 2005 two-factor model of Gurkaynak, Sack and Swanson as its origin), because equities load most on the funds-rate factor while the LSAP factor drives long-duration sector dispersion. Then orthogonalize, following the explicit recipe of Bauer and Swanson (2023): raw surprises are predictable from public pre-announcement data with an R-squared of 10% to 40%, and they remove that predictability by regressing the raw surprise on six predictors (the most recent nonfarm-payroll surprise, employment growth over the prior year, the log change in the S&P 500 from about three months before to the day before the meeting, the change in the yield-curve slope over that window, the log change in a commodity-price index over that window, and the option-implied skewness of the 10-year Treasury yield), and using the residual as the exogenous signal. The principled basis for separating a policy shock from an information shock is the sign of the stock/rate co-movement (Jarocinski and Karadi, 2020; Nakamura and Steinsson, 2018).

Use real-time (as-released) data, not later revisions, and fix one vintage-consistent expectations vendor. Macro consensus and surprise-data vendors (Bloomberg median, Action Economics/MMS, Refinitiv) differ, and the choice of expectation proxy materially changes the measured surprise, so mixing vendors or using revised release values introduces look-ahead bias.

Benchmark model, regime conditioning, and clustering

Estimate abnormal returns with a market or factor model (CAPM, Fama-French) on an estimation window cleanly separated from the event windows; for the cross-section, estimate per-asset or per-sector surprise betas, the sensitivity of the abnormal return to the standardized surprise. See Expected Return Models for the benchmark choice. Regime conditioning is mandatory: interact the surprise with the cycle state (NBER recession indicator, yield-curve slope, recent macro momentum), because the response sign flips with regime (Boyd, Hu and Jagannathan, 2005). Macro event dates cluster within calendar months (an employment report can fall in an FOMC week), inducing cross-sectional correlation in residuals that inflates t-statistics. The standardized cross-sectional BMP (Boehmer, Musumeci and Poulsen) test corrects for event-induced variance but not for the cross-sectional correlation that clustering creates, so it is a complement, not a substitute, for two further remedies: the calendar-time-portfolio (Jensen-alpha) approach, which is immune to cross-sectional-correlation bias because it aggregates event observations into calendar-month portfolios, and the cross-correlation-robust test of Kolari, Pape and Pynnonen (2018), which shows that partially overlapping event windows inflate test statistics by a factor of two or more even at modest cross-correlation and provides a corrected statistic that adjusts for the average window-overlap percentage. See Significance Tests for the full battery.

Aggregate to a tradable signal, then validate net of costs

Aggregate per-event abnormal responses (a cumulative abnormal return or a Treynor or Sharpe ratio of the response) across a rolling history into per-sector or per-asset-class tilt scores, weighting by statistical significance and decaying older observations, in the spirit of macro-momentum aggregation (Brooks et al., 2017). Then impose two disciplines. Validate every signal out-of-sample and monitor it for decay, given the pre-FOMC precedent (Kurov, Wolfe and Gilbert, 2021), and report performance net of transaction costs and cycle-timing error, since naive rotation gains are fragile (Molchanov and Stangl, 2024). Finally, apply multiple-testing discipline: scanning many release types, sectors, and windows invites data snooping, so pre-register the hypothesis set and apply significance adjustments before promoting a signal to live tilts.

The Kuttner surprise-scaling formula

Worked example: from a CPI print to a sector tilt

Expand: end-to-end CPI-surprise-to-tilt calculation

This carries a single number from a raw release all the way to a sector tilt weight. Every step is visible.

1. Raw surprise. CPI comes in at 0.5% month-on-month against a consensus of 0.3%, so the raw surprise is +0.2 percentage points (a hot print).

2. Standardize. The historical standard deviation of the CPI surprise is about 0.1 pp, so the standardized surprise is +0.2 / 0.1 = +2.0, a two-sigma upside surprise (Andersen et al., 2003). Standardizing makes the coefficient comparable across release types.

3. Abnormal return. Run the Abnormal Return Calculator with a market model on a (-1,+1) window for each sector ETF. Suppose, over the history of hot CPI prints, technology shows a mean CAR of -1.6% and utilities a mean CAR of -0.2% on a two-sigma surprise.

4. Surprise beta. Divide the abnormal return by the standardized surprise: technology beta = -1.6% / 2.0 = -0.8% per sigma; utilities beta = -0.2% / 2.0 = -0.1% per sigma. Technology is roughly eight times more rate-surprise-sensitive than utilities, consistent with the Ehrmann and Fratzscher (2004) cyclical-versus-defensive dispersion.

5. Tilt weight. A fresh +2-sigma hot CPI print implies an expected technology abnormal return of -0.8% x 2.0 = -1.6% and utilities about flat. Translate to a tilt by underweighting technology by roughly 1.6% (scaled down by conviction and inversely to the noise in the estimated beta) and leaving utilities near neutral; reverse the signs on a cool print.

The whole pipeline is just: standardize the surprise, regress the abnormal return on it to get a surprise beta, then turn the beta into a tilt. Condition the beta on the inflation and cycle regime (it can flip sign), and re-estimate as it decays.

Worked examples

• FOMC-surprise sector tilt. For each FOMC meeting, extract the fed-funds-futures surprise and the three Swanson (2021) factors, and estimate each sector's (-1,+1) abnormal-return surprise beta with the abnormal return calculator. Cyclicals and long-duration sectors (technology, discretionary, financials, real estate) load far more negatively on hawkish surprises than defensives (staples, utilities, healthcare), with the Ehrmann and Fratzscher (2004) dispersion (50 bp = ~3% at the index level, cyclicals ~2-3x defensives) as a calibration anchor. The rule: on a more-hawkish-than-expected outcome, underweight high-beta sectors and overweight defensives, scaled by each sector's historical beta; reverse on dovish surprises. Benchmark the aggregate move against the roughly 5.4% S&P move per 100 basis points of surprise.

• Monetary-momentum slope overlay. Build the Neuhierl and Weber (2019) slope factor from changes in fed-funds futures across horizons and tilt toward equity beta when the expected policy path is easing and away when it is tightening, exploiting the predictable ~25-day pre-FOMC drift (cumulative expansionary-minus-contractionary spread of about 2.5% by the meeting and more than 4.5% fifteen days after). This is the constructive, path-conditioned counterpart to the decaying calendar drift.

• FOMC-cycle even-week overlay. Implement the Cieslak, Morse and Vissing-Jorgensen (2019) rule as a zero-forecast overlay: hold full equity beta in even weeks (0, 2, 4, 6 from the last meeting) and cut to cash or bonds in odd weeks. This is the simplest possible event-response-into-allocation signal, needing only the meeting calendar.

• Employment-report regime switch. A worse-than-consensus payrolls print triggers a risk-on tilt in a confirmed expansion (rate-relief channel) but a defensive tilt in a contraction (earnings channel), operationalizing the Boyd, Hu and Jagannathan (2005) state-dependence with a yield-curve-slope or NBER regime gate.

• Macro-surprise risk-on/off composite. Aggregate standardized surprises (CPI, payrolls, ISM, retail sales, pre-screened for priced content per Flannery and Protopapadakis, 2002) into a rolling composite akin to an economic-surprise index. When the composite turns up (data beating expectations), tilt toward equities, cyclicals, and credit; when it rolls over, de-risk. For a concrete, citable construction, the Scotti (2016) surprise index weights standardized actual-minus-consensus surprises by their estimated impact on a nowcast of current GDP over a rolling three-month window with explicit time-decay weighting, which is the academic backbone of the commercial economic-surprise indices below.

How practitioners actually use this

The academic facts above map onto a small number of working investor playbooks. Four user types dominate.

• Global-macro and CTA / managed-futures funds run the macro-momentum and event-clock signals as systematic overlays across equities, rates and FX, sizing exposure off surprise composites and the policy path.

• Multi-asset and risk-parity managers use surprise composites to scale risk-on/off and to time equity beta to the event clock, dialing leverage up into low-priced-risk windows and down into high-priced-risk ones.

• Sell-side strategists publish surprise-driven sector-rotation calls off the same cross-sectional surprise betas, framing them for institutional clients.

• The buy-side as a whole names the regime-dependent, asymmetric equity response to policy the "Fed put" or "Powell put", which is exactly the state-dependence the regime section describes academically.

Where to get the surprise data (free)

You can reproduce these signals with our calculators at near-zero data cost. For monetary-policy surprises, the San Francisco Fed US Monetary Policy Event-Study Database (Acosta, Bauer and Swanson) is free and public, begins in 2020 and is regularly updated, and ships ready-made MP1 (current-meeting), MP2 (next-meeting) and ED1 to ED4 path surprises around every FOMC announcement, press conference and minutes release, plus downloadable R code that is compatible with the GSS-2005 decomposition. For the macro-surprise composite, the Citigroup Economic Surprise Index and Bloomberg's economic-surprise series are the commercial instantiations, with the Scotti (2016) surprise and uncertainty indexes as the open academic backbone covering the US, euro area, UK, Canada and Japan. One honest caveat: the Citigroup index was engineered for FX, and a widely cited critique finds it has little direct bearing on equity performance, so for a TAA equity or sector application estimate your own equity and sector surprise betas directly with ARC rather than reusing an off-the-shelf FX-weighted index. EDI reconciles the release timestamps, and ARC ingests the surprise as a regressor.

Volatility, the event premium, and AVyC

Narrow event windows and the volatility response are not academic niceties; they are how desks actually price these events. Options markets pre-price an event premium that collapses (an "IV crush") once the number prints, and the average S&P 500 move on a CPI release day has run on the order of plus or minus 0.64% versus an all-day average near 1.07% in recent samples. The forward-looking, options-implied premia of Knox, Londono and Samadi (2025) formalize exactly this ex-ante event risk. The Abnormal Volatility Calculator (AVyC) measures the realized volatility response around releases, which is the right conditioning variable for sizing tilts and for confirming that an event carried information even when the directional move is ambiguous.

Run it with our tools

The applications on this site implement the workflow above end to end. Note that the abnormal-effect calculators map to signal types: returns to ARC, volume to AVC, volatility to AVyC, and news tone to CATA.

• Abnormal Return Calculator (ARC) is the core engine. Supply an event file keyed to release timestamps, choose your estimation and event windows (a narrow window for policy, (-1,+1) or (0,+1) for daily macro), pick an expected-return model (Market Model, CAPM, Fama-French 3-factor, Fama-French 5-factor, Carhart 4-factor, or comparison-period mean), and run the parametric and non-parametric significance tests built for clustered, calendar-bunched samples. Group your events by sector to estimate the surprise betas that become tilt weights, and maintain a rolling Treynor or Sharpe ratio per sector or geography to inform systematic shifts between sub-asset classes.

• Event Date Identifier (EDI) pins down the exact release date, helps separate scheduled releases from unscheduled inter-meeting actions, and reconciles the release timestamps from a free surprise feed, the most important data-quality decision in a macro study.

• Abnormal Volume Calculator (AVC) measures the trading-volume reaction around releases, corroborating that information arrived even when the directional price move is ambiguous. Abnormal Volatility Calculator (AVyC) captures the volatility response and the IV-crush dynamic around scheduled releases, useful as a regime and dispersion gauge for conditioning tilts.

• News Analytics (CATA) scores the tone of policy statements and macro headlines, converting language into a surprise-like sentiment signal. Combine it with the quantitative surprise betas from ARC to refine tilt conviction, and watch for long-persistence sentiment as a reversal warning.

Common misconceptions and pitfalls

Frequently asked questions

What is a macroeconomic announcement premium?

It is the empirical fact that a large, disproportionate share of the equity risk premium is earned in narrow windows around a small set of scheduled macro events. Savor and Wilson (2013) find more than 60% of the cumulative premium on the roughly 13% of days carrying a major release; the longer-sample restatement by Ai, Bansal and Guo (2024) puts it at more than 71% of aggregate equity risk compensation on about 44 announcement days per year over 1961-2023.

Why is bad economic news sometimes good for stocks?

Because the response sign depends on the regime. In an expansion, weak data signals lower future rates, and the discount-rate channel dominates, so stocks can rise; in a contraction, the same data signals weaker earnings, and the earnings channel dominates, so stocks fall. Boyd, Hu and Jagannathan (2005) document this directly: the S&P rose about 5.4% on a rising-unemployment print in August 1984. A pooled beta that ignores the regime is biased toward zero.

How do you measure a Fed rate surprise?

From fed-funds futures, following Kuttner (2001): the surprise is the change in the implied futures rate around the decision, scaled by the days remaining in the month, because the contract settles on the month's average funds rate. For richer events, decompose the surprise into Swanson (2021) funds-rate, forward-guidance and LSAP factors, and orthogonalize against pre-announcement public data (Bauer and Swanson, 2023). Free MP1/MP2/ED1-ED4 series are published in the San Francisco Fed's US Monetary Policy Event-Study Database.

What event window should I use for FOMC versus CPI?

For monetary policy, use a narrow intraday window, the roughly 30-minute window grounded in Nakamura and Steinsson (2018), and measure the decision-release and press-conference windows separately (Altavilla et al., 2019). For daily macro releases such as CPI, a (-1,+1) or (0,+1) day window is standard, and pre-announcement-drift studies use the 24-hour pre-release window. State the window explicitly, because it determines what the abnormal return captures.

Does the pre-FOMC drift still work?

Largely not in equities. Kurov, Wolfe and Gilbert (2021) find it essentially vanished after about 2011 to 2015 once it became widely known, a textbook arbitraged-away anomaly, though follow-up work finds residual footprints in volatility markets. The constructive, path-conditioned cousin that has held up better is monetary momentum (Neuhierl and Weber, 2019). Treat decay as anomaly-specific and monitor every calendar signal out of sample.

Why standardize the surprise instead of using the raw beat or miss?

Standardizing (actual minus consensus, divided by the historical standard deviation of that release's surprise) puts every release on a common scale, so a coefficient estimated on CPI surprises is comparable to one estimated on payroll or ISM surprises (Andersen et al., 2003). It also lets you express a tilt as a response per sigma, which is the natural unit for aggregating heterogeneous releases into one composite.

Which sectors are most rate-sensitive?

Cyclical and long-duration sectors (technology, consumer discretionary, communications, real estate) carry the largest surprise betas; defensives (consumer staples, utilities, healthcare) carry the smallest. Ehrmann and Fratzscher (2004) find a 50 bp tightening lowers broad returns by about 3% and that cyclicals react roughly two to three times more than defensives, with financially constrained, high-q firms responding most. Treat any sector table as a prior to re-estimate, since the sign can flip with the regime.

What is the difference between TAA and SAA?

Strategic asset allocation (SAA) is the long-run policy mix that reflects an investor's objectives and risk tolerance and changes rarely. Tactical asset allocation (TAA) is the deliberate, shorter-horizon tilt away from the SAA to exploit time-varying expected returns, such as a surprise-driven sector overweight. Event studies build the systematic version of TAA by measuring how asset classes and sectors respond to dated events and turning those responses into tilt weights.

Related use cases

TAA signals sit alongside the other investing applications of this methodology. See the closely related Investment Strategies page on event-driven and signal-driven quantitative strategies, the Investment Weather and Investment Clock page on macro-regime framing, the Bottom-Fishing page on abnormal-response signals at the single-instrument level, and the Stock-Market Responses to Economy-Wide Events page on the macro events themselves. Earnings-driven drift signals are covered under Earnings Announcements, and the broader Comparative Event-Type Analyses page compares response patterns across event types. For the full catalogue, return to the Practical Applications overview.