Overview

Robust event study results require careful validation and professional presentation. The R package v0.40.0 provides comprehensive diagnostic tools and multiple export formats.

Data Validation

Before running an event study, validate your input data:

library(EventStudy)

task <- EventStudyTask$new(
  data = stock_data,
  event_date = "2020-03-15",
  estimation_window = c(-250, -11),
  event_window = c(-5, 5)
)

# Validate task setup
validate_task(task)

validate_task() checks for:

Missing values in the event or estimation window
Sufficient data points in the estimation window
Correct date format and ordering
Matching firm and market data availability

Model Diagnostics

After fitting a return model, assess its quality:

model <- MarketModel$new()
model$fit(task)

# Run comprehensive diagnostics
diagnostics <- model_diagnostics(task)
print(diagnostics)

Shapiro-Wilk Normality Test

Tests whether estimation window residuals follow a normal distribution. If rejected, consider non-parametric test statistics.

Durbin-Watson Test

Tests for first-order autocorrelation in residuals. Values near 2 indicate no autocorrelation. If rejected, standard errors may be understated.

Ljung-Box Test

Tests for higher-order serial correlation, complementing the Durbin-Watson test for higher-order patterns.

Visual Diagnostics

# Plot diagnostic charts
plot_diagnostics(task)

This produces:

Residual time series plot
Residual histogram with normal overlay
Q-Q plot
ACF (autocorrelation function) plot

Pre-Trend Testing

Pre-trend tests check whether abnormal returns existed before the event:

# Test for pre-event abnormal returns
pretrend <- pretrend_test(task, window = c(-10, -1))
print(pretrend)

A significant pre-trend suggests information leakage, confounding events, or a misspecified event date.

Diagnostic Checklist

Check	Tool	What to look for
Data quality	`validate_task()`	Missing values, date issues
Normality	Shapiro-Wilk	p > 0.05 for parametric tests
Autocorrelation	Durbin-Watson	Values near 2.0
Serial correlation	Ljung-Box	p > 0.05 for independence
Pre-trends	`pretrend_test()`	No significant pre-event ARs
Model fit	R-squared	Reasonable explanatory power

Exporting Results

The R package provides multiple export formats to fit your workflow.

CSV Export

export_results(task, format = "csv", path = "results/")

Creates separate CSV files for AR, CAR, AAR, CAAR, and test statistics.

Excel Export

export_results(task, format = "excel", path = "results/event_study.xlsx")

Multi-sheet Excel workbook with formatted tables. Requires the openxlsx package.

LaTeX Export

export_results(task, format = "latex", path = "results/tables/")

Publication-ready LaTeX tables with significance stars, standard errors in parentheses, and table notes.

Broom Integration

library(broom)

# Tidy output of event study results
tidy_results <- tidy(task)
print(tidy_results)

Returns a tibble compatible with the tidyverse, making it easy to pipe into ggplot2 or other tools.

Result Extraction Helpers

# Get specific result components
ar  <- get_ar(task)           # Abnormal returns
car <- get_car(task)          # Cumulative abnormal returns
aar <- get_aar(task)          # Average abnormal returns
model_stats <- get_model_stats(task)  # Model statistics

Example Workflow

library(EventStudy)

# 1. Run event study
task <- EventStudyTask$new(data = stock_data, ...)
model <- MarketModel$new()
model$fit(task)

# 2. Validate
validate_task(task)
model_diagnostics(task)

# 3. Run test statistics
PatellZ$new()$compute(task)
RankTest$new()$compute(task)

# 4. Export everything
export_results(task, format = "excel", path = "output/results.xlsx")
export_results(task, format = "latex", path = "output/tables/")

# 5. Visualize
plot_event_study(task)
plot_diagnostics(task)