EventStudy R Package v0.50.0: Synthetic Control, Modern DiD, and More

What's New in v0.50.0

EventStudy v0.50.0 is the latest major release of the R package for event study analysis, expanding the toolkit to 15 return models, 12 test statistics, and 6 panel DiD estimators. This release adds synthetic control methods, wild bootstrap inference, Monte Carlo power analysis, and modern difference-in-differences estimators that address the negative weighting problem identified by Goodman-Bacon (2021).

Building on the foundation of v0.40.0, this release adds time-varying parameter models, modern DiD estimators, synthetic control methods, and robust inference tools. Here is everything new.

EventStudy v0.50.0 delivers 6 panel DiD estimators that address the negative weighting bias identified by Goodman-Bacon (2021) in traditional TWFE regressions. The release also introduces Monte Carlo power analysis, enabling researchers to estimate detection rates before data collection -- a critical capability given that, as documented by Kothari and Warner (2007), event studies with fewer than 30 events often fail to detect abnormal returns of 1% or less with adequate power.

2 New Return Models (15 total)

• Rolling Window Model (RollingWindowModel) -- time-varying $\alpha$ and $\beta$ via rolling OLS. Captures structural breaks without parametric assumptions.
• DCC-GARCH Model (DCCGARCHModel) -- Dynamic Conditional Correlation GARCH for time-varying beta and volatility. Based on Engle (2002).

# Rolling window with 60-day window
ps <- ParameterSet$new(return_model = RollingWindowModel$new(window_size = 60))

# DCC-GARCH for time-varying correlation
ps <- ParameterSet$new(return_model = DCCGARCHModel$new())

Kolari-Pynnonen Test (12 test statistics total)

The new KolariPynnonenTest adjusts the BMP test for cross-sectional correlation caused by event clustering. Essential when multiple events occur on the same calendar date.

ps <- ParameterSet$new(
  multi_event_statistics = MultiEventStatisticsSet$new(
    tests = list(BMPTest$new(), KolariPynnonenTest$new())
  )
)

3 New Panel DiD Estimators (6 total)

The panel event study module now includes three modern DiD estimators alongside the existing TWFE and Sun & Abraham methods:

• Callaway & Sant'Anna (2021) -- doubly robust group-time ATTs with flexible aggregation. Requires did.
• de Chaisemartin & D'Haultfoeuille (2020) -- robust estimation for switchers, handles treatment reversals. Requires DIDmultiplegt.
• Borusyak, Jaravel & Spiess (2024) -- efficient imputation estimator for staggered adoption. Requires didimputation.

result <- estimate_panel_event_study(
  panel_task,
  method = "callaway_santanna",
  leads = 5, lags = 5
)

Synthetic Control Methods

A completely new module for single-unit causal inference. Construct a weighted combination of donor units to approximate the treated unit's counterfactual, then assess significance via placebo tests.

sc_task <- SyntheticControlTask$new(treated_data, donor_data, treatment_time = 25)
result <- estimate_synthetic_control(sc_task)
placebo <- sc_placebo_test(sc_task)
plot_synthetic_control(placebo, type = "placebo")

Wild Bootstrap Inference

When asymptotic p-values are unreliable (short estimation windows, non-normal residuals, small samples), the wild bootstrap provides more accurate inference:

boot_result <- bootstrap_test(task, n_boot = 999, weight_type = "rademacher")

Multiple Testing Corrections

Control false discovery when testing multiple windows, subgroups, or statistics:

adjusted <- adjust_p_values(task, method = "BH")    # Benjamini-Hochberg
adjusted <- adjust_p_values(task, method = "holm")   # Holm (step-down Bonferroni)

Power Analysis via Monte Carlo Simulation

Design your study with confidence. As emphasized by Kothari and Warner (2007), event studies with fewer than 20-30 events often have power below 50% to detect economically meaningful effects. The new simulation module lets you estimate detection power before collecting data:

sim <- simulate_event_study(
  n_events = 30,
  n_simulations = 1000,
  abnormal_return = 0.01
)
sim$power  # e.g., 0.82 = 82% detection rate

Summary

Synthetic Control

A method that constructs a counterfactual for a treated unit from a weighted combination of untreated donor units. Used for single-unit causal inference when traditional event study methods lack power.

Wild Bootstrap

A resampling method for constructing p-values that is robust to heteroskedasticity and non-normality. Provides more accurate inference than asymptotic tests in small samples.

Getting Started

Install from GitHub:

# install.packages("devtools")
devtools::install_github("sipemu/eventstudy")

Check the GitHub repository for the latest updates and full documentation.

Learn More

• Expected Return Models Guide
• Test Statistics Guide
• Panel Event Study Guide
• Synthetic Control Guide
• Inference & Robustness Guide
• Power Analysis Guide
• What Was New in v0.40.0
• A Beginner's Guide to Event Studies