Event Study Analysis in R

A comprehensive, modular R package for financial event study analysis. Implements the classical methodology (MacKinlay 1997) and extends it with modern multi-factor models, long-horizon methods, panel (DiD) event studies, and more.

Ecosystem

EventStudy is part of the eventstudy.de toolkit -- "Event Study Analysis Made Simple":

Installation

GitHub

install.packages("devtools")
# devtools::install_github("sipemu/eventstudy")
install.packages("EventStudy", repos = c('https://sipemu.r-universe.dev', 'https://cloud.r-project.org'))

🤖 Claude Code Skills (AI pair-programming)

Two Claude Code Agent Skills ship with this repo as an installable plugin, giving your AI assistant grounded knowledge of the package API:

Install via the built-in plugin marketplace:

/plugin marketplace add sipemu/eventstudy
/plugin install eventstudy@eventstudy

The skills are then auto-invoked from their descriptions, or callable as /eventstudy:es-capabilities and /eventstudy:es-advisor.

Features

Quick Start

library(EventStudy)

# Create task from your data
task <- EventStudyTask$new(firm_tbl, index_tbl, request_tbl)

# Run with default settings (Market Model, simple returns, all test statistics)
task <- run_event_study(task)

# Inspect results
print(task)
summary(task)

AI Advisor

New in 0.60.0. Get grounded advice about your fitted study — fully offline, no API key, and it never invents numbers (it interprets only what the package computed):

diag   <- es_diagnostics(task)
advice <- es_advise(diag, task_type = "recommend_stat")
print(advice)

Runs offline out of the box. To use an LLM for prose interpretation, pass an optional provider() (Anthropic, OpenAI-compatible, or custom).

The best entry point for understanding the advisor is the "The Grounded AI Advisor" vignette (vignette("ai-advisor")), which walks through the bundled dieselgate dataset (data(dieselgate) — Volkswagen vs. the DAX around the 2015 EPA disclosure), runs the deterministic advisor layer live, and shows the LLM layer as a captured static example.

Example: Dieselgate

Analyzing the 2015 Volkswagen emissions scandal and its impact on automotive stock prices.

1. Load Data

library(tidyquant)
library(dplyr)
library(EventStudy)

index_symbol <- "^GDAXI"
firm_symbols <- c("VOW.DE", "PAH3.DE", "BMW.DE", "MBG.DE")

request_tbl <- tibble(
  event_id = 1:4,
  firm_symbol = firm_symbols,
  index_symbol = rep(index_symbol, 4),
  event_date = rep("18.09.2015", 4),
  group = c(rep("VW Group", 2), rep("Other", 2)),
  event_window_start = rep(-10, 4),
  event_window_end = rep(10, 4),
  shift_estimation_window = rep(-11, 4),
  estimation_window_length = rep(250, 4)
)

firm_tbl <- firm_symbols %>%
  tq_get(from = "2014-06-01", to = "2015-11-01") %>%
  mutate(date = format(date, "%d.%m.%Y")) %>%
  select(symbol, date, adjusted)

index_tbl <- index_symbol %>%
  tq_get(from = "2014-06-01", to = "2015-11-01") %>%
  mutate(date = format(date, "%d.%m.%Y")) %>%
  select(symbol, date, adjusted)

2. Run the Event Study

One-liner with defaults:

task <- EventStudyTask$new(firm_tbl, index_tbl, request_tbl)
task <- run_event_study(task)

Step-by-step with custom parameters:

task <- EventStudyTask$new(firm_tbl, index_tbl, request_tbl)

est_params <- ParameterSet$new(
  return_calculation = LogReturn$new(),
  return_model = MarketModel$new(),
  single_event_statistics = SingleEventStatisticsSet$new(),
  multi_event_statistics = MultiEventStatisticsSet$new()
)

task <- prepare_event_study(task, est_params)
task <- fit_model(task, est_params)
task <- calculate_statistics(task, est_params)

Pipe-friendly:

task <- EventStudyTask$new(firm_tbl, index_tbl, request_tbl)
est_params <- ParameterSet$new(return_calculation = LogReturn$new())

task |>
  prepare_event_study(est_params) |>
  fit_model(est_params) |>
  calculate_statistics(est_params)

3. Extract Results

# Overview
print(task)
summary(task)

# Single-event results
task$get_ar(event_id = 1)           # Abnormal returns
task$get_car(event_id = 1)          # Cumulative abnormal returns
task$get_model_stats(event_id = 1)  # Model fit statistics

# Multi-event results
task$get_aar(stat_name = "CSectT")  # AAR/CAAR table

# Broom-compatible tidy output
tidy.EventStudyTask(task, type = "ar")
tidy.EventStudyTask(task, type = "car")
tidy.EventStudyTask(task, type = "aar")
tidy.EventStudyTask(task, type = "model")

4. Export Results

# CSV (one file per result type)
export_results(task, "results.csv")

# Excel (each result type on a separate sheet)
export_results(task, "results.xlsx")

# LaTeX (publication-ready tables)
export_results(task, "results.tex", which = c("model", "aar"))

5. Visualize

# Event study plots with confidence bands
plot_event_study(task, type = "car", event_id = 1)
plot_event_study(task, type = "caar", group = "VW Group")

# Model diagnostics
plot_diagnostics(task, event_id = 1)

# CAR distribution across events
plot_car_distribution(task, by_group = TRUE)

# Raw price plots
plot_stocks(task, add_event_date = TRUE)

6. Diagnostics and Validation

validate_task(task)
model_diagnostics(task)
pretrend_test(task)

Available Models

Model Class Description
Market Model MarketModel OLS: $R_i = \alpha + \beta R_m + \epsilon$
Market Adjusted MarketAdjustedModel $AR_i = R_i - R_m$
Mean Adjusted ComparisonPeriodMeanAdjustedModel $AR_i = R_i - \bar{R}_i^{est}$
Fama-French 3-Factor FamaFrench3FactorModel $R_i - R_f = \alpha + \beta_m(R_m - R_f) + \beta_s SMB + \beta_h HML + \epsilon$
Fama-French 5-Factor FamaFrench5FactorModel 3-factor + RMW + CMA
Carhart 4-Factor Carhart4FactorModel 3-factor + MOM
GARCH(1,1) GARCHModel Time-varying volatility (requires rugarch)
BHAR BHARModel Buy-and-hold: $\prod(1+R_i) - \prod(1+R_m)$
Volume VolumeModel Abnormal trading volume
Volatility VolatilityModel Abnormal volatility (variance ratio)
Custom CustomModel Extend MarketModel with custom AR logic
Linear Factor LinearFactorModel Base class for all OLS factor models

Using Factor Models

Factor models require a factor_tbl with Fama-French data:

factor_data <- tibble(
  date = dates,             # matching format as firm/index data
  smb = smb_returns,
  hml = hml_returns,
  risk_free_rate = rf_rate  # triggers automatic excess return computation
)

task <- EventStudyTask$new(firm_tbl, index_tbl, request_tbl,
                            factor_tbl = factor_data)
params <- ParameterSet$new(return_model = FamaFrench3FactorModel$new())
task <- run_event_study(task, params)

Available Test Statistics

Single Event (per-firm AR/CAR)

Test Class Description
AR T-Test ARTTest $t = AR_{i,t} / \sigma_i$
CAR T-Test CARTTest $t = CAR_i / (\sqrt{L} \cdot \sigma_i)$
BHAR T-Test BHARTTest t-test on buy-and-hold abnormal returns

Multiple Events (cross-sectional AAR/CAAR)

Test Class Description
Cross-Sectional T CSectTTest Standard cross-sectional t-test
Patell Z PatellZTest Standardized residual test (Patell 1976)
BMP Test BMPTest Standardized cross-sectional (Boehmer et al. 1991)
Sign Test SignTest Proportion of positive ARs vs 50%
Generalized Sign GeneralizedSignTest Adjusted for asymmetry (Cowan 1992)
Rank Test RankTest Non-parametric rank test (Corrado 1989)
Calendar-Time Portfolio CalendarTimePortfolioTest Portfolio approach for clustered events

Cross-Sectional Regression

Explain why some firms are affected more than others:

firm_chars <- tibble(
  event_id = 1:4,
  log_market_cap = c(10, 11, 12, 10.5),
  leverage = c(0.3, 0.5, 0.2, 0.4)
)

result <- cross_sectional_regression(
  task, formula = ~ log_market_cap + leverage,
  data = firm_chars
)
print(result)

# Group comparisons
car_by_group(task)
car_quantiles(task)

Intraday Event Studies

For high-frequency studies around earnings announcements, Fed decisions, etc.:

task <- IntradayEventStudyTask$new(
  firm_intraday_tbl,    # symbol, timestamp (POSIXct), price
  index_intraday_tbl,
  request_tbl           # event_timestamp, windows in observation counts
)
task <- prepare_intraday_event_study(task, ParameterSet$new())
task <- fit_model(task, ParameterSet$new())

Non-parametric significance test (Rinaudo & Saha 2014): compares event-day CARs to the empirical distribution of estimation-day CARs to build confidence bands without distributional assumptions.

results <- nonparametric_intraday_test(
  estimation_window = est_data,   # day, time, abnormalReturn
  event_window      = event_data, # time, abnormalReturn
  event_times       = c("10:00", "14:30"),
  p = 0.05
)

See vignette("intraday-event-study") for details.

Panel Event Studies (DiD)

For difference-in-differences style analysis with staggered treatment:

task <- PanelEventStudyTask$new(
  panel_data,
  unit_id = "firm_id",
  time_id = "year",
  outcome = "revenue",
  treatment = "policy_adopted",
  treatment_time = "adoption_year"
)

# Dynamic TWFE with event study plot
task <- estimate_panel_event_study(task, method = "dynamic_twfe",
                                    leads = 5, lags = 5)
plot_panel_event_study(task)

# Sun & Abraham for staggered treatment
task <- estimate_panel_event_study(task, method = "sun_abraham",
                                    leads = 5, lags = 5)

Extending the Package

Custom Models

MyModel <- R6::R6Class("MyModel",
  inherit = ModelBase,
  public = list(
    model_name = "MyModel",
    fit = function(data_tbl) {
      # Estimate on estimation window, set private$.is_fitted = TRUE
      # Populate private$.statistics$sigma and $degree_of_freedom
    },
    abnormal_returns = function(data_tbl) {
      # Return data_tbl with an abnormal_returns column
    }
  )
)

Custom Test Statistics

MyTest <- R6::R6Class("MyTest",
  inherit = TestStatisticBase,
  public = list(
    name = "MyTest",
    compute = function(data_tbl, model) {
      # Return a tibble with your test results
    }
  )
)

multi_tests <- MultiEventStatisticsSet$new()
multi_tests$add_test(MyTest$new())

See vignette("custom-models") and vignette("custom-test-statistics") for detailed tutorials.

Vignettes

Vignette Topic
ai-advisor Grounded AI advisor two-layer walkthrough on the bundled dieselgate dataset
introduction Package overview with Dieselgate example
custom-models Creating and using custom return models
custom-test-statistics Creating custom test statistics
result-extraction Extracting and exporting results
factor-models-bhar Factor models (FF3, FF5, Carhart), GARCH, and BHAR
cross-sectional-analysis Cross-sectional regression, group comparisons, CAR quantiles
diagnostics-validation Data validation, model diagnostics, pre-trend testing
volume-volatility-event-study Volume and volatility event studies
intraday-event-study Non-parametric intraday event studies
panel-event-study Panel DiD event studies

Advisor Pro (Waitlist)

Advisor Pro is a planned premium add-on that will provide retrieval-grounded, evidence-backed AI advice powered by a curated academic knowledge base. It will integrate seamlessly with the existing es_advise() / es_diagnostics() pipeline.

Planned features:

Current free tier already includes offline KB-based advice via recommend_stat() and flag_robustness() — no API key required.

Join the waitlist: https://github.com/sipemu/eventstudy#advisor-pro-waitlist

Opt-in footer reminder (silent by default):

options(eventstudy.advisor_pro_footer = TRUE)
# Appends a static waitlist URL to advice print output — no network call.

See ?advisor_pro for details.

References

License

AGPL-3