Data Preparation for Event Studies

library(EventStudy)
library(tidyverse)

# Load price data (e.g., from Yahoo Finance via tidyquant)
# library(tidyquant)
# raw_prices <- tq_get(c("AAPL", "MSFT", "GOOGL"),
#                       from = "2018-01-01", to = "2023-12-31")

# Format for the EventStudy package
firm_data <- raw_prices |>
  transmute(
    symbol = symbol,
    date = format(date, "%d.%m.%Y"),
    adjusted = adjusted
  )

head(firm_data)
# # A tibble: 6 x 3
#   symbol date       adjusted
#   <chr>  <chr>         <dbl>
# 1 AAPL   02.01.2018     172.
# 2 AAPL   03.01.2018     172.
# 3 AAPL   04.01.2018     173.
# 4 AAPL   05.01.2018     175.
# 5 AAPL   08.01.2018     174.
# 6 AAPL   09.01.2018     174.

# Market index data (same structure as firm data)
index_data <- sp500_prices |>
  transmute(
    symbol = "SP500",
    date = format(date, "%d.%m.%Y"),
    adjusted = adjusted
  )

# Ensure the index covers the same date range as firm data
stopifnot(min(index_data$date) <= min(firm_data$date))
stopifnot(max(index_data$date) >= max(firm_data$date))

# Define events
request <- tibble(
  event_id = 1:3,
  firm_symbol = c("AAPL", "MSFT", "GOOGL"),
  index_symbol = "SP500",
  event_date = c("15.03.2022", "20.04.2022", "10.05.2022"),
  group = c("Tech", "Tech", "Tech"),
  event_window_start = -5,
  event_window_end = 5,
  shift_estimation_window = -10,
  estimation_window_length = 250
)

# The estimation window will be:
# [event_date + shift - estimation_window_length,
#  event_date + shift]
# i.e., 250 days ending 10 days before the event

# Convert from Date object to DD.MM.YYYY string
date_obj <- as.Date("2022-03-15")
formatted <- format(date_obj, "%d.%m.%Y")
# "15.03.2022"

# Convert from common formats
# From YYYY-MM-DD (ISO)
as.Date("2022-03-15") |> format("%d.%m.%Y")

# From MM/DD/YYYY (US)
as.Date("03/15/2022", format = "%m/%d/%Y") |> format("%d.%m.%Y")

# From YYYYMMDD (compact)
as.Date("20220315", format = "%Y%m%d") |> format("%d.%m.%Y")

# Batch convert a data frame column
df$date <- format(as.Date(df$date), "%d.%m.%Y")

# Check for missing values
sum(is.na(firm_data$adjusted))

# Check for gaps in trading dates per firm
firm_data |>
  group_by(symbol) |>
  summarise(
    n_days = n(),
    first_date = min(date),
    last_date = max(date),
    n_missing = sum(is.na(adjusted))
  )

# Remove firms with insufficient data
min_required_days <- 250 + 10 + 5  # estimation + gap + event window
firm_data <- firm_data |>
  group_by(symbol) |>
  filter(n() >= min_required_days) |>
  ungroup()

# Optional: forward-fill isolated missing prices
# (use with caution -- this assumes zero return on missing days)
firm_data <- firm_data |>
  group_by(symbol) |>
  arrange(date) |>
  fill(adjusted, .direction = "down") |>
  ungroup()

# 1. Remove duplicates
firm_data <- firm_data |> distinct(symbol, date, .keep_all = TRUE)
index_data <- index_data |> distinct(symbol, date, .keep_all = TRUE)

# 2. Check for zero or negative prices
stopifnot(all(firm_data$adjusted > 0, na.rm = TRUE))
stopifnot(all(index_data$adjusted > 0, na.rm = TRUE))

# 3. Check for extreme returns (flag but don't auto-remove)
firm_returns <- firm_data |>
  group_by(symbol) |>
  arrange(date) |>
  mutate(ret = adjusted / lag(adjusted) - 1) |>
  ungroup()

extreme <- firm_returns |> filter(abs(ret) > 1)
if (nrow(extreme) > 0) {
  warning(paste(nrow(extreme), "extreme returns detected. Verify data."))
  print(extreme)
}

# 4. Check that event dates exist in firm data
event_dates_valid <- request |>
  rowwise() |>
  mutate(
    date_in_firm = event_date %in%
      (firm_data |> filter(symbol == firm_symbol) |> pull(date)),
    date_in_index = event_date %in%
      (index_data |> filter(symbol == index_symbol) |> pull(date))
  )

invalid_events <- event_dates_valid |>
  filter(!date_in_firm | !date_in_index)

if (nrow(invalid_events) > 0) {
  warning("Some event dates are not trading days. Adjust or remove:")
  print(invalid_events)
}

library(EventStudy)
library(tidyverse)

# --- Step 1: Prepare firm prices ---
firm_data <- tibble(
  symbol = rep(c("AAPL", "MSFT"), each = 300),
  date = rep(format(
    seq(as.Date("2021-01-04"), by = "day", length.out = 430) |>
      ((d) d[!weekdays(d) %in% c("Saturday", "Sunday")])() |>
      head(300),
    "%d.%m.%Y"
  ), 2),
  adjusted = c(
    cumprod(1 + rnorm(300, 0.0005, 0.015)) * 130,
    cumprod(1 + rnorm(300, 0.0004, 0.012)) * 240
  )
)

# --- Step 2: Prepare market index ---
index_data <- tibble(
  symbol = "SP500",
  date = unique(firm_data$date),
  adjusted = cumprod(1 + rnorm(300, 0.0003, 0.010)) * 4200
)

# --- Step 3: Create event request ---
request <- tibble(
  event_id = 1:2,
  firm_symbol = c("AAPL", "MSFT"),
  index_symbol = "SP500",
  event_date = firm_data |>
    filter(symbol == "AAPL") |>
    slice(260) |>
    pull(date),
  group = "Tech",
  event_window_start = -5,
  event_window_end = 5,
  shift_estimation_window = -10,
  estimation_window_length = 240
)

# --- Step 4: Quality checks ---
stopifnot(all(firm_data$adjusted > 0))
stopifnot(nrow(distinct(firm_data, symbol, date)) == nrow(firm_data))

# --- Step 5: Run event study ---
task <- EventStudyTask$new(firm_data, index_data, request)
ps <- ParameterSet$new()
task <- run_event_study(task, ps)

# --- Step 6: View results ---
task$get_aar()