Investment Weather and Clock

Asset classes rotate with the economic cycle: bonds lead when growth and inflation fall, equities when growth recovers, commodities when the economy overheats, and cash when growth stalls while inflation runs hot. The framework is the easy part. Knowing where on the map the economy actually sits right now is the hard part, and the event study is the natural instrument for reading it from how markets react to scheduled macro news.

Asset classes move in and out of favor with the economic cycle: government bonds tend to lead when growth and inflation are falling, equities when growth is recovering, commodities when the economy overheats, and cash when growth stalls while inflation stays high. The hard part is never the map; it is knowing where on the map the economy actually is right now. This page is about reading that position, the macro "weather", from how markets react to scheduled macroeconomic news, and why the event study is the natural instrument for the job.

Two practitioner frameworks dominate this space, both built on the same two axes (the direction of growth and the direction of inflation relative to trend) and both using an everyday-life analogy to make the message intuitive. The Investment Clock, popularized by Merrill Lynch in Greetham and Hartnett (2004), places the cycle on a clock face with four phases: Reflation (growth down, inflation down, favoring bonds), Recovery (growth up, inflation down, favoring stocks), Overheat (growth up, inflation up, favoring commodities), and Stagflation (growth down, inflation up, favoring cash). The related Investment Weather map renders the same growth-by-inflation space as a 2x2 forecast: same axes, different visual metaphor. Within equities, the same two axes rotate sector leadership: rate-sensitive and early-cycle names (financials, consumer discretionary) lead first, then cyclicals and technology lead as growth accelerates, resources and energy lead when inflation accelerates, and defensives (staples, utilities, healthcare) lead when growth rolls over (Stovall, 1996).

The premise underneath both frameworks is that macroeconomic risk is priced. Chen, Roll and Ross (1986) established that systematic macro innovations (unexpected inflation, changes in expected inflation, industrial-production growth, the default spread, and the term spread) carry priced risk premia in the cross-section of equities. If the cycle is priced, then the cleanest signal of how markets read the cycle is their reaction to the unanticipated part of a macro release, which is exactly what an event study isolates.

How to read the 2x2: the generative rule

Rather than memorize four facts, learn the rule that generates them. Each quadrant is defined by a growth arrow and an inflation arrow; together they pick the winning asset through a dominant channel:

Read clockwise from the bottom left and the winners rotate Bonds, Stocks, Commodities, Cash in turn. The diagram opposite each quadrant names the loser: Reflation's opposite is Overheat, so commodities are the worst Reflation asset; Recovery's opposite is Stagflation, so cash lags badly in Recovery. (Designer note: render this as a theme-CSS SVG clock-face 2x2 with the four phase winners on the diagonal corners and the cyclical / value / defensive / growth sector quadrants from Chart 1 of the original report, so the verbal rule gets a canonical visual anchor.)

Event studies are the right tool here for a specific, often-missed reason. The frameworks are descriptive scaffolds; they do not tell you which phase you are in. But markets price the cycle continuously, and the cleanest signal of how markets read the economy is their abnormal reaction to a macro surprise: the return of an asset class, factor, or sector around an FOMC decision, a payrolls print, or a CPI release, net of what would have happened anyway. Measured across asset classes and sectors, daily and rule-driven, the cross-section of those abnormal returns is a quantified, market-based reading of the macro outlook, and therefore of the position on the clock. That turns an analysis usually dominated by opinion into something testable.

The frameworks are not museum pieces. The Investment Clock remains a live, multi-asset investment process: Trevor Greetham, co-author of the original Merrill Lynch report, has run it for roughly two decades, carrying it from Merrill Lynch (where he co-authored it) to Fidelity and then to Royal London Asset Management, where it governs asset allocation across a multi-asset book and a current clock position is published each quarter. The point of this page is the missing instrument underneath that discretion: a rule-based, market-based reading of where the cycle is, derived from event-study reactions rather than judgment alone.

What the research shows

The academic literature underneath this use case is large and mature, and it converges on a handful of facts that make "reading the weather from reactions" defensible rather than wishful. One meta-observation up front, because it answers the obvious "has this held up?" question: the headline magnitudes have if anything strengthened across data vintages (the announcement premium runs from roughly 60% in Savor and Wilson (2013) to about 55% measured differently in Ai and Bansal (2018) to more than 71% on the longest sample in Ai, Bansal and Guo (2023)), while one widely publicized calendar anomaly, the pre-FOMC drift, has decayed to insignificance. What is structural held; what was a tradable anomaly faded once known.

The original back-test, in numbers

The single most authoritative artifact for this use case is the original Merrill Lynch back-test, and it is worth reproducing because the framework's own numbers validate the rotation map at exceptional significance. Greetham and Hartnett (2004) classify each month from April 1973 to July 2004 (375 months) into a clock phase using the OECD output gap and CPI inflation, then compute real (inflation-adjusted) geometric annualized total returns for four U.S. dollar asset classes. The asset proxies are the ML U.S. Treasury/Agencies Master index (bonds), the S&P 500 Composite (stocks), the GSCI Total Return index (commodities), and 3-month T-Bills (cash).

The statistical headline is exactly this page's thesis. One-way ANOVA rejects equal returns across phases at extremely high confidence for every asset class (the report's confidence levels are 99.93% for bonds, 99.90% for stocks, 99.99% for commodities, and 99.99% for cash), and the four "clock-opposite" pair trades are all at least 95% significant (stocks beat cash in Recovery at more than 99.9%). Two honesty notes the report makes itself, and that any serious treatment must keep. First, the bolded theory winner is not always the numeric top performer: in Stagflation, commodities returned +28.6% versus cash at -0.3%, yet cash is the designated winner because it is "the best of a bad bunch" for capital preservation, and that +28.6% is dominated by the 1970s oil shocks while non-oil commodity prices fell, a textbook case of an episode-driven average. Second, and more important for credibility, the same report concedes from the inside that the framework works far better for broad asset rotation than for sector rotation: its own sector ANOVA is significant for only about half of the broad equity sectors, with Consumer Discretionary and Oil & Gas the most macro-driven and Telecoms, Utilities and Basic Materials "poorly explained."

Macro-announcement days carry the premium

The deepest reason event windows around macro releases are informative is that they are where the macro risk premium is actually earned. Savor and Wilson (2013) show that over 1958 to 2009 the average excess return on the U.S. equity market was about 11.4 basis points on the roughly 13% of trading days carrying a scheduled CPI, employment, or FOMC announcement, versus about 1.1 basis points on all other days. More than 60% of the cumulative annual equity premium was earned on those announcement days, the Sharpe ratio was an order of magnitude higher, and the CAPM beta-return relation holds on announcement days but essentially vanishes on non-announcement days. The cross-asset payoff is sharper still: Savor and Wilson (2014) show that the positive beta-return relation on scheduled-announcement days holds not only for equities but also for Treasury bonds and currencies, which is the direct justification for reading the macro weather across asset classes on event days rather than in equities alone. Ai and Bansal (2018) put theory under this fact: returns around pre-scheduled announcements (the employment report and FOMC) account for roughly 55% of the market equity premium, and a positive announcement premium requires generalized risk sensitivity (recursive or robust-control preferences rather than time-separable expected utility). The conclusion has held up and strengthened with more data: Ai, Bansal and Guo (2023) report that over 1961 to 2023, the roughly 44 announcement days per year carry more than 71% of aggregate equity-market risk compensation, a larger share on a longer sample than the earlier estimates. Two implications follow directly. First, the event window is not an arbitrary measurement convenience; it is the part of the calendar where the macro information and the compensation for bearing macro risk are concentrated. Second, pooling announcement and non-announcement days dilutes the signal mechanically, the same way pooling across regimes does (below). The premium and the cross-section both live on the event days.

Only surprises move prices

The single most important methodological fact is that markets respond to the unanticipated component of news, not the headline. The early equity evidence is Pearce and Roley (1985), who show that only the surprise component of money-supply, inflation and real-activity announcements moves stock prices, with anticipated news priced in advance. Kuttner (2001) sharpened this for monetary policy: anticipated changes in the federal funds target move bond yields close to zero, while the surprise component, extracted from fed funds futures, produces large and significant moves. This is precisely why an event study is the right instrument: it isolates the surprise (actual minus consensus, measured in a tight window) instead of relying on the naive and frequently wrong reasoning that "the data was strong, so stocks should rise." Bernanke and Kuttner (2005) make the magnitude concrete: an unanticipated 25 basis point rate cut is associated with roughly a 1% rise in broad equity indices on the announcement day, working mainly through the equity risk premium, with cyclical, rate-sensitive sectors reacting far more than defensives. (That 25bp-for-1% mapping is the calibration anchor for the worked example below.) Which specific releases carry priced surprises is itself an empirical question: Flannery and Protopapadakis (2002), modeling daily S&P returns on seventeen macro series, identify six priced factors (three nominal: CPI, PPI, and a monetary aggregate; three real: the balance of trade, the employment report, and housing starts) and find that Industrial Production and GNP are not priced, with several releases affecting volatility more than the mean.

State-dependence: the same surprise can flip sign

The load-bearing finding for this page is state-dependence: the sign of the equity reaction to a fixed macro surprise depends on the business cycle. McQueen and Roley (1993) show that the stock-price reaction to macroeconomic news is weak unconditionally but strong once you condition on the cycle stage, and that in a strong economy the market reacts negatively to good real-activity news because the discount-rate channel dominates expected cash flows. Boyd, Hu and Jagannathan (2005) sharpen this into the cleanest available example: a rising-unemployment surprise (bad growth news) on average raises stock prices during expansions, where the rate-relief channel dominates, but lowers them during contractions, where the expected-dividend channel dominates. The intuition behind the paradoxical "bad news is usually good for stocks" is a base rate: over their 1961 to 1995 sample the economy was in expansion in 351 of 408 months (about 86% of the time), so the rate-relief sign wins on average, then flips in the rare recession, and that flip is itself the readable cycle indicator. This is the exact mechanism that lets a measured abnormal return read the regime rather than just the data, and it is why pooling across regimes destroys the signal: averaged over the whole cycle the reaction is near zero, and the information is invisible. The mechanism is not a 1990s artifact: a 2025 Federal Reserve study (Board of Governors, FEDS 2025-007) re-confirms that ignoring state-dependence yields weaker or insignificant estimates of the stock reaction, because the relative strength of the cash-flow, risk-premium and interest-rate channels shifts over the cycle. The regime logic also has a peer-reviewed antecedent that predates the practitioner clock: Jensen, Mercer and Johnson (1996) classify the monetary environment as expansive or restrictive by the direction of the most recent Fed discount-rate change and find that the business-condition predictability of stock and bond returns is concentrated in, and reverses across, those regimes, with sizeable cross-industry differences.

Cross-asset comovement and the growth-versus-inflation distinction

The Investment Clock's growth-by-inflation quadrants have a direct empirical counterpart in the regime-dependent stock-bond correlation. Campbell, Pflueger and Viceira (2020) show that bonds are "safe" (a negative stock-bond return correlation) when shocks are growth-driven, and "risky" (a positive correlation) when inflation and the output gap comove, as they did before 2001 and again in 2022. Andersen, Bollerslev, Diebold and Vega (2007) document the same logic at event frequency across global stock, bond and FX markets: the equity response to news is conditional on the cycle stage, which is why the unconditional stock-bond correlation is near zero (it flips sign across regimes). Operationally, this means the sign of the joint stock-and-bond reaction to a release is a fast read of whether the market fears growth or inflation: a growth surprise that lifts stocks and sells bonds (negative price comovement) flags a growth-driven regime, while an inflation or policy surprise that sells both (positive comovement) flags the regime in which bonds stop hedging.

The inflation threshold matters quantitatively, not just directionally. The realized stock-bond correlation is roughly 14% when inflation runs in a benign 2% to 4% band but climbs to about 32% above 4% inflation, and lifting the correlation from -0.5 to +0.5 raises a 60/40 portfolio's volatility from about 7.7% to about 10.4% (roughly a third higher). The 2022 to 2024 inflation episode is the textbook confirmation: the realized stock-bond correlation flipped from roughly -0.37 over the post-financial-crisis decade to about +0.41 over 2022 to 2024, and the classic 60/40 portfolio fell about 17.5% in 2022, its worst calendar year since 1937. Positive stock-bond correlation is in fact the historical norm, holding in roughly four of five years (about 78%) since 1870; the negative-correlation decade the 60/40 relied on was the anomaly. In the same period, hot CPI and activity prints repeatedly produced negative equity reactions through the rate and discount-rate channel, the "good news is bad news" regime, while soft inflation prints drove sharp rallies as yields fell. This is precisely the machine-readable signature an event study captures: when both the equity-index and the long-Treasury abnormal returns are negative on a hot inflation print, the cross-asset reaction is telling you bonds have stopped hedging.

FOMC dominance, the policy factors, and the calendar anomalies

FOMC announcements are the dominant scheduled event for equities, and the policy reaction is multi-dimensional. Gurkaynak, Sack and Swanson (2005) show that FOMC effects require a "target" (current-rate) surprise and a "path" (forward-guidance) surprise, with the statement-driven path factor dominating longer-term yields and equities. Swanson (2021) extends this to three orthogonal factors (the federal funds rate, forward guidance, and large-scale asset purchases) over 1991 to 2019, and finds that guidance and asset-purchase surprises moved yields, equities and exchange rates by amounts comparable to conventional rate surprises. This matters for the structural-break problem below: during the 2008 to 2015 zero-lower-bound and QE era the conventional rate surprise was near zero, so a single rate-change regressor would have shown no effect even as policy moved markets; the reaction loaded almost entirely on the guidance and asset-purchase factors. A single "rate change" variable is therefore misspecified. A further confounder is the Fed information effect: Nakamura and Steinsson (2018) find that in a tight 30-minute window a hawkish surprise can coincide with upward revisions to expected growth, because the announcement reveals the Fed's private read of the economy. The contemporary correction is Bauer and Swanson (2023): raw high-frequency monetary surprises are predictable from publicly available pre-announcement macro and financial data (regression R-squared of roughly 10% to 40%), so the "information effect" is better read as a "Fed response to news" channel. Their recommended fix is to orthogonalize each surprise against the pre-announcement data before using it as the event regressor, which restores exogeneity, leaves the asset-price effects largely unchanged, and makes the estimated macro effects substantially larger and more significant.

FOMC dates also display two distinct calendar regularities that any event window should respect. Lucca and Moench (2015) document the pre-FOMC drift: from 1994 to 2011, U.S. equity excess returns averaged about 49 basis points in the 24 hours before scheduled FOMC announcements, accounting for roughly 80% of the annual equity premium over that period (Sharpe near 1.1), with returns at and after the announcement close to zero. Crucially, this anomaly did not merely soften: out-of-sample replication by Kurov, Wolfe and Gilbert (2021) extends the sample to December 2019 and finds the drift fell from about 0.5% to about 0.1% per meeting and became statistically insignificant, essentially disappearing after about 2015 for meetings both with and without press conferences (their proposed mechanism is less pre-announcement uncertainty resolution). This is the cleanest available "documented patterns decay once publicized" example, and it is exactly why this page insists on re-testing any reaction out of sample. Operating at a lower frequency, Cieslak, Morse and Vissing-Jorgensen (2019) show that since 1994 essentially the entire equity premium has been earned in even weeks (0, 2, 4, 6) of FOMC-cycle time measured from the last meeting, with odd weeks earning on-average negative excess returns, a biweekly pattern they tie causally to the Fed via intermeeting communication. Both findings argue for reporting a pre-event cumulative abnormal return alongside the event-window measure, because an event-only window understates the policy-related move.

Which releases matter, and the bad-news asymmetry

Among scheduled data, the rough ranking by equity and Treasury impact is FOMC and monetary policy first, then nonfarm payrolls (the most consistently market-moving release), then CPI, GDP, and the ISM and PMI surveys. This ranking is not folklore: Gilbert, Scotti, Strasser and Vega (2017) show that a release's price impact is driven chiefly by its "intrinsic value", above all its timing (how early in the data cycle it arrives and how much it forecasts later releases), more than by its relation to fundamentals or its revision noise. Crucially, which release dominates rotates with the regime, so the identity of the market-moving release is itself a weather reading: CPI surprises moved markets far more in the 2022 to 2024 high-inflation regime than in the prior low-inflation decade, and the dominant release also differs by asset class (employment reports tended to drive equity reactions even through the recent inflation surge, while CPI dominated interest-rate volatility around recession risk). Reactions are also asymmetric: bad news generates larger absolute moves than good news of equal size, a result that goes back to the real-time foreign-exchange evidence of Andersen, Bollerslev, Diebold and Vega (2003) and carries through to global equity and bond markets in Andersen, Bollerslev, Diebold and Vega (2007).

A caution: rotation as diagnosis, not turnkey alpha

The honest counterweight matters for credibility, and the strongest version of it comes from inside the framework. As noted above, the original report's own sector ANOVA is significant for only about half of the broad equity sectors, so even the authors concede that the clock works for broad asset rotation far better than for sector rotation. The external evidence agrees: academic back-tests find that cyclically sensitive sectors do outperform in expansions and defensives in contractions, but Molchanov and Stangl (2024) find no systematic sector outperformance where the clock predicts it once transaction costs and realistic cycle-timing error are accounted for, with cross-sector predictability close to chance, and earlier back-tests of a naive Stovall-style list found only about 2.3% per year of gross excess return (Stangl, Jacobsen and Visaltanachoti, 2009). A regime-classified rotation can do better in some samples (Conover, Jensen, Johnson and Mercer, 2008 find sizeable Fed-state sector spreads, for example apparel cyclically sensitive at roughly 50% and energy near 20%), but the framework itself is sample-and-market dependent out of sample: clock-style rotation that "worked" in one market and period has broken down in others (for instance it was effective in China before about 2012 and failed thereafter). The defensible position is therefore that event studies are best used here as a measurement and nowcasting diagnostic ("where is the market saying we are in the cycle?"), not as a guaranteed rotation strategy. Note the symmetry with the announcement-premium evidence: the macro-event signal has a real economic basis (the premium is positive and risk is rewarded on those days), but extracting reliable rotation alpha from it after costs is a separate and much harder claim. Inferring the cycle from reactions is inferential, not an established causal measurement, and should be flagged as such.

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-regime studies add several event-type-specific choices that change the design substantially.

Define the event as the surprise, not the outcome

Scheduled releases have exact timestamps (FOMC at 14:00 ET; nonfarm payrolls and CPI at 08:30 ET), so the event date is unambiguous. The "event", however, is the surprise. The canonical standardization is from Balduzzi, Elton and Green (2001): surprise = (actual minus median consensus) divided by the time-series standard deviation of that gap, which makes magnitudes comparable across heterogeneous releases. Then regress the windowed abnormal return on the standardized surprise. A concrete data-choice warning from the same source: use the latest available consensus (forecasters update until release time) rather than a stale survey, or the surprise is contaminated by expectations that were already revised. For monetary policy, extract the surprise from market instruments rather than the headline rate: the current-month fed funds futures contract (scaled for its monthly-averaging convention) for the target surprise (Kuttner, 2001); OIS or Eurodollar curves, or the two-factor target-plus-path decomposition of Gurkaynak, Sack and Swanson (2005), for forward guidance; and the third asset-purchase factor of Swanson (2021) for the ZLB and QE era. The public FRBSF U.S. Monetary Policy Event-Study Database, documented in Acosta, Ajello, Bauer, Loria and Miranda-Agrippino (2025), packages these high-frequency surprises directly; that same paper reports that large policy surprises returned after 2020 and that the post-meeting press conference is now the most important single source of policy news. For the euro area, the equivalent public resource is the EA-MPD of Altavilla et al. (2019), which generalizes the target/path/QE factor decomposition outside the United States.

Use very short windows, and bracket the whole policy event

Macro information is impounded in minutes, so the gold standard is an intraday window (often 30 minutes) bracketing the release, where the "normal" return over the window is approximately zero and the raw windowed return is the abnormal return (Andersen, Bollerslev, Diebold and Vega, 2003, 2007). Daily windows already contaminate the reaction with confounding news, but they remain appropriate for cross-sectional and sector-dispersion work where the signal is the spread across portfolios rather than the micro-reaction of a single release. State the convention explicitly. For FOMC events, do not bracket only the 14:00 ET statement: because the press conference (typically 14:30 ET) is now the dominant policy-news channel (Acosta et al., 2025), and because Altavilla et al. (2019) show the euro-area release is one-dimensional while the press conference carries additional forward-guidance and QE dimensions, a window that stops before the presser systematically truncates the move. And because of the pre-FOMC drift and the broader FOMC-cycle pattern, report both a pre-event and an event-window cumulative abnormal return (CAR) so anticipatory returns are visible (Lucca and Moench, 2015; Cieslak, Morse and Vissing-Jorgensen, 2019).

Sample portfolios, not firms, and pick the benchmark with care

The unit of analysis is portfolios: asset-class proxies (equity index, long Treasury, commodity, gold or cash) and sector and factor baskets (the GICS sectors; value versus growth; cyclical versus defensive; duration buckets). The cross-section of their response coefficients is the weather reading. The benchmark choice has a subtle trap for economy-wide events: a market-model normal return can mechanically absorb part of a market-wide macro shock, because the index is the event. Prefer a mean-adjusted return, or a cross-sectionally differenced measure (sector minus market, cyclical minus defensive), so the signal is the rotation rather than the level. See Expected Return Models for the trade-offs.

Regime-conditioning is mandatory

This is the design choice that makes the use case work. Split or interact by regime (NBER expansion versus recession, the sign of the output gap, or the growth-by-inflation quadrant) before pooling. Without the interaction, sign-flipping responses cancel and the average reaction is near zero (McQueen and Roley, 1993; Boyd, Hu and Jagannathan, 2005). Public regime inputs include the OECD composite leading indicator and the Conference Board LEI; an alternative classifier is the expansive-versus-restrictive scheme based on the direction of the most recent Fed discount-rate change (Jensen, Mercer and Johnson, 1996). For a live nowcast, a published practitioner clock position (such as the quarterly Royal London Investment Clock) can serve as an independent sanity check on the regime read.

Confounders, clustering and the right test statistics

Macro dates bring acute confounders: releases cluster within a week (payrolls and ISM; CPI and retail sales), policy events bundle a statement, presser and projections, and the Fed information effect mixes policy with outlook news. Use tight windows, control for simultaneous releases, and treat unscheduled or intermeeting actions separately. When the no-other-news assumption is genuinely doubtful, the principled alternative is the identification-through-heteroskedasticity approach of Rigobon and Sack (2004), which identifies the policy reaction from the rise in shock variance on event days relative to matched non-event windows and so explicitly allows "background noise" inside the window; it yields magnitudes similar to the event study when the assumptions hold, which cross-validates the simpler approach. Statistically, every sector shares the same systematic shock, so ordinary cross-sectional t-tests, which assume independence, over-reject. Use the standardized cross-sectional (BMP) test of Boehmer, Musumeci and Poulsen (1991), the cross-correlation-robust adjustment of Kolari and Pynnonen (2010), and non-parametric rank and generalized-sign tests; see our overview of Significance Tests. Because the number of event dates per regime is small, statistical power is a real constraint: pool the standardized cumulative surprises within a regime and run the clustering-robust cross-sectional regression rather than reading single dates. Finally, be alert to structural breaks (post-1994 FOMC statements, the 2008 zero-lower-bound and QE era when conventional rate surprises were tiny, the 2022 inflation regime, and the post-2020 return of large surprises with the press conference as the dominant channel), and re-test any reaction out of sample, as the pre-FOMC drift's disappearance warns.

A worked example you can reproduce in ARC

To make the method concrete, walk one CPI release end to end. The numbers in the first table are illustrative (chosen to show the arithmetic and the read-out), calibrated so the magnitudes are realistic against the Bernanke-Kuttner anchor that a real 25bp policy surprise maps to roughly a 1% index move.

Step 1: standardize the surprise. Suppose consensus core CPI is +0.3% month-on-month and the actual print is +0.6%, and the historical standard deviation of the (actual minus consensus) gap is 0.12 percentage points. Then the standardized surprise is (0.6 minus 0.3) divided by 0.12 = +2.5 standard deviations, a 2.5-sigma upside inflation surprise. That standardized value, not the raw +0.3pp, is the regressor.

Step 2: read the cross-section of one-day abnormal returns. Build event files for five asset-class proxies, set the CPI release date as the event date, use a short event window with an estimation window ending before the event, and (because the event is economy-wide) use a mean-adjusted or cross-sectionally differenced benchmark rather than a market model so the index does not absorb its own shock. An illustrative cross-section for a 2.5-sigma hot inflation print in an inflation-fearing regime:

Step 3: read out the regime. Equities and long Treasuries are both negative: a positive comovement. Bonds have stopped hedging, which places the economy on the Overheat or Stagflation (inflation-up) side of the clock. The same exercise reads the regime change in reverse: the day the equity and bond abnormal returns stop sharing a sign, the market is telling you the inflation fear has receded and bonds are hedging again.

How practitioners run this

This is not only an academic construction. The Investment Clock has a continuous buy-side lineage: Trevor Greetham co-authored it as Director of Asset Allocation at Merrill Lynch, ran it at Fidelity, and since 2015 has run it as Head of Multi Asset at Royal London Asset Management, where it governs allocation across a large multi-asset book and a current clock position is published each quarter as a public, independent regime read. The growth-by-inflation quadrant logic is genuinely industry-standard rather than a single-firm curio: the same four-box thinking underlies Bridgewater's All Weather approach and is taught across the asset-management industry. On the input side, practitioners do not hand-build every surprise. The Citigroup Economic Surprise Index (CESI, since 2003) is the off-the-shelf aggregate version of exactly this method, a daily, time-decayed, standardized (actual minus consensus) surprise gauge that macro and FX desks watch as a real-time "are the data beating or missing" weather reading; the high-frequency FOMC surprise series of Bauer and Swanson (2023) and the FRBSF database of Acosta et al. (2025) supply the monetary-policy inputs. Framed this way, our Abnormal Return Calculator is the firm-specific, asset-by-asset cross-section that CESI produces in aggregate: where CESI gives one scalar, ARC gives you the full cross-section of who reacted and by how much, which is what locates the regime.

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. Build event files for asset-class proxies and the GICS sector portfolios with the release date as the event date, choose a short event window and an estimation window that ends before the event, and select a benchmark suited to economy-wide events (a comparison-period mean, or a market model used to construct sector-minus-market abnormal returns). Run the full battery of parametric and non-parametric tests designed for clustered, cross-correlated samples, then export the per-portfolio CARs and regress them on the standardized surprise, interacted with a regime indicator, to read the cross-section. The worked CPI example above is a direct ARC application.

• News Analytics (CATA) turns the macro narrative into measurable inputs: build proprietary sentiment or tone series around releases and screen the surrounding news flow for confounding announcements that would contaminate a daily window.

• Abnormal Volume Calculator (AVC) and Abnormal Volatility Calculator (AVyC) corroborate that information actually arrived: a genuine surprise shows up as abnormal volume and an event-window volatility spike, which also helps separate anticipated from surprise releases.

• Event Date Identifier (EDI) helps assemble and align clean release-date series across many assets to a common timestamp, the prerequisite for a comparable cross-section.

Common misconceptions and pitfalls

• "Strong data means stocks go up" is wrong. Only the surprise moves prices, and its sign is regime-dependent: in a hot economy good real-activity news can sell stocks through the discount-rate channel (Kuttner, 2001; Boyd, Hu and Jagannathan, 2005).
• The clock is a map, not a GPS. The framework does not tell you which phase you are in. The event-study cross-section is the instrument that locates you; the clock just names the destinations.
• Do not pool across regimes. Sign-flipping reactions average to near zero, hiding the signal (McQueen and Roley, 1993). Split or interact by regime first.
• Do not use a market-model benchmark for economy-wide events. The index is the shock, so a market model mechanically absorbs part of it. Use mean-adjusted or sector-minus-market abnormal returns.
• Sector rotation is the weak link. Even the original report's own ANOVA was significant for only about half of broad sectors, and out-of-sample sector alpha is fragile after costs (Molchanov and Stangl, 2024). Treat rotation as diagnosis, not turnkey alpha.
• Phase averages can be episode-driven. The +28.6% Stagflation commodity return is mostly the 1970s oil shocks, with non-oil commodities falling. Small samples per regime mean a single episode can dominate an average.
• Do not bracket only the FOMC statement. The 14:30 ET press conference is now the dominant policy-news channel; a window that closes at 14:00 truncates the move (Acosta et al., 2025).
• Do not assume a documented pattern persists. The pre-FOMC drift disappeared after about 2015 once publicized (Kurov, Wolfe and Gilbert, 2021). Always re-test out of sample.

Frequently asked questions

Why do stocks sometimes fall on good economic news?

Because markets price the surprise, not the headline, and the sign of the equity reaction depends on the cycle. In a strong economy, good real-activity news raises expected interest rates and the discount rate faster than it raises expected cash flows, so equities can fall. The same surprise that lifts stocks in a weak economy can sink them in a hot one (McQueen and Roley, 1993; Boyd, Hu and Jagannathan, 2005).

What are the four phases of the Investment Clock, and which asset wins each?

Reflation (growth down, inflation down) favors bonds; Recovery (growth up, inflation down) favors stocks; Overheat (growth up, inflation up) favors commodities; Stagflation (growth down, inflation up) favors cash. In the original 1973 to 2004 back-test the phase winners were bonds at 9.8% real in Reflation, stocks at 19.9% in Recovery, commodities at 19.7% in Overheat, and cash at -0.3% (best of a bad bunch) in Stagflation (Greetham and Hartnett, 2004).

Is the Investment Clock still used today?

Yes. Trevor Greetham, co-author of the original Merrill Lynch report, has run the clock as a live multi-asset process for roughly two decades and now does so at Royal London Asset Management, which publishes a current clock position each quarter. The growth-by-inflation quadrant logic is industry-standard (it parallels Bridgewater's All Weather framing).

Does sector rotation actually work?

Modestly and fragilely. The direction is real (cyclicals in expansions, defensives in contractions), but the original report's own ANOVA was significant for only about half of broad sectors, and Molchanov and Stangl (2024) find no systematic sector outperformance after transaction costs and realistic cycle-timing error. Use rotation as a diagnostic, not as a guaranteed strategy.

What is the difference between the Investment Clock and the Investment Weather?

They use the same two axes, the direction of growth and the direction of inflation relative to trend, and differ only in the visual metaphor: the Clock arranges the four phases on a clock face, while the Investment Weather renders the same growth-by-inflation space as a 2x2 forecast map.

Which macro release moves markets the most?

Roughly: FOMC and monetary policy first, then nonfarm payrolls, then CPI, GDP, and the ISM/PMI surveys. But the dominant release rotates with the regime and differs by asset class: CPI dominated in the 2022 to 2024 inflation regime and drives interest-rate volatility, while employment reports often drive equity reactions. A release's impact is driven mainly by its timing and intrinsic value (Gilbert, Scotti, Strasser and Vega, 2017).

Why does the event window matter so much?

Because the macro risk premium is concentrated there. The roughly 13% of days carrying a scheduled CPI, employment, or FOMC release earn the bulk of the equity premium (more than 60% in Savor and Wilson, 2013; more than 71% over 1961 to 2023 in Ai, Bansal and Guo, 2023), and the CAPM beta-return relation holds on those days across stocks, bonds and currencies. The information and the compensation both live in the window.

Related use cases

Reading the macro weather sits next to several sibling applications. See Tactical Asset Allocation Signals for turning abnormal-effect output into asset-class tilts, Investment Strategies and Bottom Fishing for trading applications, and Stock Market Responses to Economy-Wide Events for the broader category of macro and market-wide event studies. The Comparative Event-Type Analyses page covers pooling and comparing across event types. For the full catalogue, return to the Practical Applications overview.