JJRESEARCH
ARCHIVE
RECORD / IR–001CLOSE PAPER ×
EMPIRICAL FINANCE / EVENT STUDYREVISED PROTOCOL / VERSION 0.2SEPTEMBER 2026
ANALYSIS IN PROGRESS

CPI SURPRISES
AND S&P 500
RETURNS

How do the size and direction of CPI surprises affect S&P 500 returns across 1-, 5-, and 10-trading-day windows—and does the market environment change that response?

AUTHORJordan JohnsonFIELDMacroeconomics × Empirical FinanceOUTPUTProtocol before inference
01PRIMARY INFORMATION
SIGNAL
03RETURN
HORIZONS
PRERELEASE REGIME
MEASUREMENT
FINAL ESTIMATES
PENDING

This project studies information that arrives differently from expectation—not inflation in isolation.

This research examines how the size and direction of U.S. CPI surprises relate to subsequent S&P 500 returns and whether that relationship changes across market environments. The response is evaluated over 1-, 5-, and 10-trading-day windows.

The objective is not to prove that CPI predicts the market. It is to test whether a relationship is detectable, economically meaningful, stable across defensible specifications, and robust to attempts to break it.

RESEARCH PRINCIPLEA null result is a result. An unstable coefficient is a result. A relationship confined to one unusual inflation regime is also a result.
How do the size and direction of CPI surprises affect S&P 500 returns over 1, 5, and 10 trading days—and does the market environment change that response?
INFORMATIONCPI surprise

Actual CPI minus the consensus expectation: what the announcement added beyond the forecast.

CONDITIONPre-release environment

Volatility and, where defensible, inflation or monetary-policy regimes measured without post-release information.

RESPONSEForward S&P 500 return

Simple returns over clearly aligned 1D, 5D, and 10D event windows—not the index price level.

CLAIM BOUNDARY

The models estimate associations around scheduled announcements. Coefficients will not be described as causal effects, trading signals, or proof of prediction.

MARKETS REACT TO
WHAT CHANGED.

Financial markets are forward-looking. Investors form expectations before the Bureau of Labor Statistics publishes CPI, so the reported inflation rate alone is not necessarily new information.

If CPI is close to consensus, much of the information may already be reflected in prices. A meaningful departure from consensus introduces a forecast error that can prompt repricing.

RAW SURPRISESurpriset = Actualt − Consensust
01 / RELEASECPI reportActual headline and core measures
02 / EXPECTATIONActual vs. consensusWhat was new to the market?
03 / INFORMATIONCPI surpriseRaw and standardized forecast error
04 / RESPONSEMarket repricingS&P 500 return after release
CONDITIONED BYPre-release volatilityInflation / policy environmentMEASURED AT1D5D10D
01

Headline CPI YoY

Reported 12-month change in the all-items index.

02

Headline CPI MoM

Reported one-month change in the all-items index.

03

Core CPI YoY

Reported 12-month change excluding food and energy.

04

Core CPI MoM

Reported one-month change excluding food and energy.

05

Headline surprise

Headline actual minus the matched headline consensus.

06

Core surprise

Core actual minus the matched core consensus.

07

Standardized surprise

Raw forecast error scaled by historical surprise dispersion available before event t.

08

Surprise direction

Hotter, approximately in-line, or cooler than expected under a documented rule.

09

Pre-release VIX

Last observed VIX value available before the CPI announcement.

10

Economic regime

Inflation or policy state used only if its definition is fixed and reproducible.

RAW SURPRISESt = Actualt − Consensust

Kept in the original percentage-point units so the economic size remains interpretable.

STANDARDIZED SURPRISEZt = St / σhistorical,t−1

Scales the raw error by dispersion estimated from prior surprises only. The historical window and dispersion estimator remain a flagged protocol decision.

HOTTER THAN EXPECTEDActual > consensus

Positive surprise outside the eventual in-line band.

APPROXIMATELY IN LINEActual ≈ consensus

The numerical tolerance has not been chosen and will not be invented after viewing returns.

COOLER THAN EXPECTEDActual < consensus

Negative surprise outside the eventual in-line band.

DEPENDENT VARIABLES

R₁ / R₅ / R₁₀

Rk = (Pt+k−1 / Ppre) − 1

Proposed primary simple-return anchor: the last close before the 8:30 a.m. ET release through the close of the kth trading session. This definition is displayed as a protocol decision and must be locked before calculation; no silent switch to log returns is permitted.

H1

SURPRISE & RETURNS

H₀: βsurprise = 0
H₁: βsurprise ≠ 0
Is a CPI surprise statistically associated with the subsequent return?

TWO-SIDED / PRIMARY
H2

MARKET ENVIRONMENT

H₀: βinteraction = 0
H₁: βinteraction ≠ 0
Does the association change under the predefined volatility regime?

INTERACTION / PRIMARY
H3

RETURN HORIZON

H₀: β₁D = β₅D = β₁₀D
H₁: Not all horizon estimates are equal.
Persistence is tested; it is not assumed.

HORIZON / PRIMARY
H4

ASYMMETRY

H₀:hot| = |βcool|
H₁:hot| ≠ |βcool|
Neither direction is presumed to dominate.

NONLINEAR / PRIMARY
CONFIRMATORY

Primary variables, horizons, hypotheses, and model specifications are fixed before final inference.

EXPLORATORY

Relationships discovered after inspecting the data will be labeled exploratory and will not be rewritten as prior hypotheses.

THE ASSUMPTIONS
STAY VISIBLE.

BASELINE SPECIFICATION / k ∈ {1, 5, 10}Rt→t+k = α + β₁Surpriset + β₂HighVIXt + β₃(Surpriset × HighVIXt) + εt
α

Baseline return

Expected return under the reference surprise and volatility conditions.

β₁

Surprise relationship

Estimated change in return associated with a one-unit CPI surprise.

β₂

Regime difference

Average difference associated with the high-volatility regime.

β₃

Interaction

Whether an equally sized surprise is associated with a different response when pre-release volatility is high.

ε

Unexplained variation

Other influences on returns not represented by this specification.

LOOK-AHEAD GUARD

Condition must exist before the release.

HighVIXt may use only VIX information available before the CPI announcement. The exact high-versus-normal criterion is not yet locked, so no regime observations are classified in the published artifact.

OPEN PROTOCOL DECISIONSThese are exposed—not silently assumed.
DECISION REQUIRED

Consensus provider

Provider and forecast-snapshot timestamp

DECISION REQUIRED

Return anchor

Confirm proposed prior-close-to-kth-session convention

DECISION REQUIRED

In-line band

Numerical tolerance for hot / in-line / cool

DECISION REQUIRED

Standardization

Historical window and dispersion estimator

DECISION REQUIRED

VIX regime

Predefined high-volatility criterion

DECISION REQUIRED

SPX provider

Exact price-series vendor and adjustment policy

LAYER 01

RAW / MASTER CPI RELEASE DATASET

Reference month · release date · release time · headline CPI YoY / MoM · core CPI YoY / MoM · headline consensus · core consensus · raw surprises · standardized surprise · source · source timestamp

LAYER 02

EVENT-STUDY DATASET

Release date · surprise measures · direction · S&P 500 price anchors · 1D / 5D / 10D returns · pre-release VIX · volatility regime · optional inflation / policy regime · event flags · audit notes

SERIESEXACT SOURCE STATUSROLEPUBLICATION RULE
01CPI actualsU.S. Bureau of Labor StatisticsRelease informationReference month and release timestamp retained
02ConsensusProvider decision requiredExpectation baselineProvider and snapshot time required before use
03S&P 500Price provider decision requiredReturn responseAdjustment and holiday alignment documented
04VIXCboe historical VIX / plannedPre-release conditionLast observation available before announcement
EVENT AUDIT INTERFACE

No publication-ready event rows loaded

0 VALIDATED RECORDS
EMPIRICAL INTEGRITY GATE

Sorting, filtering, and event-detail views will activate only after release dates, expectations, returns, regimes, and sources pass validation. Empty is more accurate than publishing invented records.

EVENT DETAIL WILL SHOW

Actual · consensus · raw surprise · standardized surprise · pre-release VIX · SPX 1D / 5D / 10D · regime labels · sources · overlapping-event notes

RESULTS WILL EARN
THEIR PLACE.

NO EMPIRICAL ESTIMATES PUBLISHED
Immediate response

1D

Captures the release session from the last pre-release close through the event-day close.

Lowest window-contamination risk
PLANNED ANALYSISPENDING VALIDATED ESTIMATES

CPI Surprise vs. 1-Day SPX Return

X
Standardized CPI surprise
Y
SPX 1D return

Scatterplot and model estimate publish only after source and event-date validation.

PLANNED ANALYSISPENDING VALIDATED ESTIMATES

CPI Surprise vs. 5-Day SPX Return

X
Standardized CPI surprise
Y
SPX 5D return

Longer window; contamination risk is reported beside the estimate.

PLANNED ANALYSISPENDING VALIDATED ESTIMATES

CPI Surprise vs. 10-Day SPX Return

X
Standardized CPI surprise
Y
SPX 10D return

No persistence is assumed; sign changes and null results remain visible.

PLANNED ANALYSISPENDING VALIDATED ESTIMATES

Hot vs. In-Line vs. Cool

X
Predefined surprise class
Y
Return distribution

Classification does not activate until the in-line threshold is locked.

PLANNED ANALYSISPENDING VALIDATED ESTIMATES

Normal vs. High VIX

X
Pre-release volatility regime
Y
CPI-surprise sensitivity

Uses only information available before the release.

PLANNED ANALYSISPENDING VALIDATED ESTIMATES

Rolling CPI Sensitivity

X
Event time
Y
Estimated coefficient

Publishes only if sample support and window length are statistically defensible.

STATISTICAL SIGNIFICANCE

Is the estimate distinguishable from zero?

Report coefficient, standard error, confidence interval, and p-value. A p-value above the chosen threshold means insufficient evidence to reject the null under that specification—not proof of no relationship.

ECONOMIC SIGNIFICANCE

Is the estimated response large enough to matter?

Translate a one-standard-deviation surprise into basis points only after the estimate exists. Statistical significance alone does not establish economic importance.

ROBUSTNESS & SENSITIVITY

Try to break the finding.

Alternative specifications challenge the primary result; they are not a search for a preferred p-value.

  1. 01Raw surprise vs. standardized surprise
  2. 02Headline vs. core CPI
  3. 03Month-over-month vs. year-over-year measures
  4. 041D vs. 5D vs. 10D return windows
  5. 05Alternative pre-release volatility rules
  6. 06Full sample vs. defensible economic subperiods
  7. 07Models with and without influential observations
  8. 08Heteroskedasticity-robust standard errors
  9. 09Autocorrelation-robust inference when warranted
Primary variables, event windows, hypotheses, and model specifications are documented before final inference. Statistically insignificant results will be retained and reported. Alternative specifications will test robustness rather than search for a preferred result.

The gap is not whether anyone has studied CPI surprises. The question is whether documented relationships remain stable outside one unusual period.

FEDERAL RESERVE / FEDS 2025

How Markets Process Macro News: The Importance of Investor Attention

T. Niklas Kroner

Kroner documents a sharp increase in financial-market reactions to CPI surprises during the 2021–2023 inflation surge and links stronger responses partly to elevated pre-announcement investor attention. The study motivates time-varying CPI sensitivity.

OPEN FEDERAL RESERVE PAPER ↗
APPLIED ECONOMICS LETTERS / 2026

Asymmetric S&P 500 Reactions to CPI Surprises in a High-Inflation Environment

Leoš Šafár · Jakub Sopko · Michal Mešťan

The paper studies asymmetric S&P 500 responses to CPI surprises in the high-inflation environment of the early 2020s. It motivates testing asymmetry without assuming which direction must dominate.

OPEN PUBLISHER RECORD ↗
01

STABILITY

Does the relationship persist across broader inflation and policy environments?

02

MAGNITUDE

Is the response economically meaningful, not merely statistically detectable?

03

ASYMMETRY

Do hot and cool surprises produce different absolute responses?

04

PERSISTENCE

Does the estimated response remain across 1D, 5D, and 10D windows?

05

STATE DEPENDENCE

Does the pre-release market environment change the association?

01

CAUSALITY

An observed association does not establish that CPI caused the full market move. The design measures response around an announcement, not an isolated structural shock.

02

OVERLAPPING INFORMATION

Federal Reserve communication, labor data, earnings, geopolitics, and other news can enter prices during the same window.

03

SAMPLE SIZE

CPI is released roughly monthly, so even a long sample contains far fewer events than a daily-return dataset. Statistical power and model complexity are limited.

04

CHANGING REGIMES

A surprise of the same numerical size may carry different information across inflation, growth, and monetary-policy environments.

05

CONSENSUS QUALITY

Consensus is only a proxy for market expectations. Provider coverage and the forecast snapshot time can change the measured surprise.

06

WINDOW CONTAMINATION

Unrelated information accumulates in the 5D and 10D windows, weakening any causal interpretation as the horizon expands.

07

MODEL UNCERTAINTY

Results may depend on the horizon, surprise definition, sample, regime rule, extreme-event treatment, and model specification.

WHAT WOULD CHANGE MY MIND?

EVIDENCE THAT WEAKENS THE HYPOTHESIS.

  • Coefficients close to zero or confidence intervals too wide to support a useful conclusion
  • Failure to reject the null under the confirmatory specification
  • Effects disappearing across reasonable alternative specifications
  • Opposite or unstable signs across return horizons
  • Results driven by a small number of extreme CPI events
  • Effects disappearing outside the 2021–2023 inflation surge
  • Unstable Surprise × HighVIX interaction estimates
01

Research question

The information–condition–response framework is defined.

COMPLETE
02

Literature review

Core papers are mapped; broader synthesis is still expanding.

IN PROGRESS
03

Hypotheses

Four confirmatory questions are specified without requiring significance.

COMPLETE
04

Data collection

Historical releases exist; consensus provenance and provider coverage still require audit.

IN PROGRESS
05

Data cleaning

Begins after provider, timestamp, and release-date checks are locked.

PENDING
06

Exploratory analysis

No published visual estimates until the event dataset passes validation.

PENDING
07

Baseline models

Regression estimates have not been published.

PENDING
08

Regime analysis

The pre-release VIX rule must be specified first.

PENDING
09

Robustness

Runs only after the confirmatory specification is frozen.

PENDING
10

Interpretation

Statistical and economic significance will be reported separately.

PENDING
11

Paper

The current public artifact is a methodology-first research protocol.

PROTOCOL
METHODOLOGY REVISION LOG

THE PROJECT EVOLVED
AS THE QUESTION SHARPENED.

01ORIGINAL FRAME

CPI levels / broad CPI–SPX relationship

Established the topic and assembled the initial historical research base.

02EVENT ALIGNMENT

CPI announcement event study

Shifted market timing from the CPI reference month to the actual release date.

03EXPECTATION ADJUSTMENT

CPI surprise analysis

Separated the published value from what investors expected before the release.

04CURRENT PROTOCOL

Regime-conditioned, robustness-tested design

Tests stability, asymmetry, persistence, and sensitivity without requiring significance.

U.S. Bureau of Labor Statistics. Consumer Price Index releases, historical tables, release dates, and methodology. Source ↗

Kroner, T. Niklas. “How Markets Process Macro News: The Importance of Investor Attention.” Finance and Economics Discussion Series, Board of Governors of the Federal Reserve System, 2025. DOI ↗

Šafár, Leoš, Jakub Sopko, and Michal Mešťan. “Asymmetric S&P 500 Reactions to CPI Surprises in a High-Inflation Environment.” Applied Economics Letters, 2026. DOI ↗

Cboe Global Markets. Historical daily closing values for the Cboe Volatility Index. Source ↗

MacKinlay, A. Craig. “Event Studies in Economics and Finance.” Journal of Economic Literature, 1997.

Market-data and consensus sources. Final provider, coverage, timestamp, and transformation documentation remain open protocol decisions and will be added before empirical publication.

DOWNLOADS

REVISED RESEARCH PROTOCOL / v0.2

Use the browser print dialog to save the complete methodology, decision register, and status record as a PDF.