Datasets:

SarthakVishnu
/

dissertation-dataset

Tags:

License:

Dataset card Files Files and versions

xet

Community

dissertation-dataset / scripts /postprocess_form4.py

SarthakVishnu

Upload scripts/postprocess_form4.py with huggingface_hub

1a5e56b verified 5 months ago

Raw History Blame

8.97 kB

# postprocess_form4.py
# Cleans up form4_transactions.csv:
#   1. Drops footnote columns (*_fn)
#   2. Converts date format from 30-MAR-2006 → 2006-03-30
#   3. Re-fetches is_officer / is_director / officer_title from one sample
#      quarter to confirm correct column names, then re-downloads all quarters
#      to add these columns to the existing data
#   4. Renames columns for clarity
#   5. Overwrites form4_transactions.csv with the clean version
#
# Run: python scripts/postprocess_form4.py

import io
import os
import time
import zipfile
import requests
import pandas as pd
from tqdm import tqdm

# ── Paths ──────────────────────────────────────────────────────────────────────
OUT_PATH  = r"D:\UoE AI\Dissertation\IPP Draft\datasets\form4_transactions.csv"
HEADERS   = {'User-Agent': 'S2880814 University of Edinburgh s.g.vishnu@sms.ed.ac.uk'}
SLEEP     = 0.5
TIMEOUT   = 120

# ── 1. Load existing data ──────────────────────────────────────────────────────
print("Loading form4_transactions.csv...")
df = pd.read_csv(OUT_PATH, low_memory=False)
print(f"  {len(df):,} rows, {len(df.columns)} columns")

# ── 2. Drop footnote columns ───────────────────────────────────────────────────
fn_cols = [c for c in df.columns if c.endswith('_fn')]
df.drop(columns=fn_cols, inplace=True)
print(f"  Dropped {len(fn_cols)} footnote columns → {len(df.columns)} columns remain")

# ── 3. Fix date format: 30-MAR-2006 → 2006-03-30 ─────────────────────────────
print("  Converting dates...")
df['transaction_date'] = pd.to_datetime(
    df['transaction_date'], format='mixed', dayfirst=True, errors='coerce'
).dt.strftime('%Y-%m-%d')
print(f"  Sample dates after fix: {df['transaction_date'].dropna().head(3).tolist()}")

# ── 4. Discover correct REPORTINGOWNER column names from one sample quarter ────
print("\nChecking REPORTINGOWNER column names from 2024 Q1...")
sample_url = "https://www.sec.gov/files/structureddata/data/insider-transactions-data-sets/2024q1_form345.zip"
r = requests.get(sample_url, headers=HEADERS, timeout=TIMEOUT)
zf = zipfile.ZipFile(io.BytesIO(r.content))
owner_file = next((n for n in zf.namelist() if 'reportingowner' in n.lower()), None)
sample_own = pd.read_csv(io.BytesIO(zf.read(owner_file)), sep='\t', low_memory=False, nrows=5)
sample_own.columns = sample_own.columns.str.lower().str.strip()
print(f"  REPORTINGOWNER columns: {sample_own.columns.tolist()}")

# Identify the correct column names
col_map = {}
for col in sample_own.columns:
    if col in ('isdir', 'isdirector'):
        col_map['is_director'] = col
    elif col == 'isofficer':
        col_map['is_officer'] = col
    elif col == 'istenpercentowner':
        col_map['is_ten_pct_owner'] = col
    elif col == 'officertitle':
        col_map['officer_title'] = col
    elif 'accession' in col:
        col_map['accession_col'] = col

print(f"  Mapped columns: {col_map}")

# ── 5. Re-download all quarters to get owner role columns ─────────────────────
def quarter_urls():
    urls = []
    year, qtr = 2006, 1
    while (year, qtr) <= (2026, 1):
        url = f"https://www.sec.gov/files/structureddata/data/insider-transactions-data-sets/{year}q{qtr}_form345.zip"
        urls.append((year, qtr, url))
        qtr += 1
        if qtr > 4:
            qtr, year = 1, year + 1
    return urls

# Get accession numbers already in our dataset
our_accessions = set(df['accession_number'].dropna())
print(f"\nRe-fetching owner role columns for {len(our_accessions):,} accessions...")

acc_col   = col_map.get('accession_col', 'accession_number')
dir_col   = col_map.get('is_director')
off_col   = col_map.get('is_officer')
pct_col   = col_map.get('is_ten_pct_owner')
title_col = col_map.get('officer_title')
keep_own  = [c for c in [acc_col, 'rptownercik', dir_col, off_col, pct_col, title_col] if c]

owner_frames = []

for year, qtr, url in tqdm(quarter_urls(), desc="Re-fetching owner data"):
    time.sleep(SLEEP)
    try:
        r = requests.get(url, headers=HEADERS, timeout=TIMEOUT)
        if r.status_code != 200:
            continue
        zf  = zipfile.ZipFile(io.BytesIO(r.content))
        own_file = next((n for n in zf.namelist() if 'reportingowner' in n.lower()), None)
        if not own_file:
            continue
        own = pd.read_csv(io.BytesIO(zf.read(own_file)), sep='\t', low_memory=False, on_bad_lines='skip')
        own.columns = own.columns.str.lower().str.strip()

        acc_col_q = next((c for c in own.columns if 'accession' in c), None)
        if not acc_col_q:
            continue

        # Only keep rows for our accessions
        own = own[own[acc_col_q].isin(our_accessions)]
        if own.empty:
            continue

        cols_to_keep = [c for c in [acc_col_q, 'rptownercik', dir_col, off_col, pct_col, title_col] if c and c in own.columns]
        owner_frames.append(own[cols_to_keep].copy())

    except Exception as e:
        tqdm.write(f"  [WARN] {year} Q{qtr}: {e}")
        continue

if owner_frames:
    owner_df = pd.concat(owner_frames, ignore_index=True)
    owner_df.columns = owner_df.columns.str.lower().str.strip()

    # Rename to standard names
    rename_own = {}
    if dir_col   and dir_col   in owner_df.columns: rename_own[dir_col]   = 'is_director'
    if off_col   and off_col   in owner_df.columns: rename_own[off_col]   = 'is_officer'
    if pct_col   and pct_col   in owner_df.columns: rename_own[pct_col]   = 'is_ten_pct_owner'
    if title_col and title_col in owner_df.columns: rename_own[title_col] = 'officer_title'
    owner_df.rename(columns=rename_own, inplace=True)

    # Deduplicate: one row per accession + rptownercik
    acc_col_own = next((c for c in owner_df.columns if 'accession' in c), None)
    if acc_col_own:
        owner_df = owner_df.drop_duplicates(subset=[acc_col_own, 'rptownercik'])

        print(f"  Owner rows fetched: {len(owner_df):,}")
        print(f"  Owner columns: {owner_df.columns.tolist()}")

        # Merge onto main df
        df = df.merge(
            owner_df,
            left_on=['accession_number', 'rptownercik'],
            right_on=[acc_col_own, 'rptownercik'],
            how='left'
        )
        # Drop duplicate accession col if created
        if acc_col_own != 'accession_number' and acc_col_own in df.columns:
            df.drop(columns=[acc_col_own], inplace=True)

        print(f"  is_officer populated: {df['is_officer'].notna().sum():,} rows" if 'is_officer' in df.columns else "  [WARN] is_officer not in merged result")
else:
    print("  [WARN] No owner role data retrieved — skipping merge")

# ── 6. Final column rename + reorder ──────────────────────────────────────────
rename_final = {
    'shrs_ownd_folwng_trans': 'shares_after_txn',
    'valu_ownd_folwng_trans': 'value_after_txn',
    'direct_indirect_ownership': 'ownership_type',
    'trans_form_type': 'form_type',
    'nonderiv_trans_sk': 'trans_sk',
}
df.rename(columns={k: v for k, v in rename_final.items() if k in df.columns}, inplace=True)

# Preferred column order
priority = ['ticker', 'issuer_cik', 'accession_number', 'filing_date' if 'filing_date' in df.columns else None,
            'owner_name', 'rptownercik', 'is_officer', 'is_director', 'is_ten_pct_owner', 'officer_title',
            'transaction_date', 'transaction_code', 'acquired_disposed',
            'shares', 'price_per_share', 'shares_after_txn', 'value_after_txn',
            'security_title', 'ownership_type', 'form_type']
priority = [c for c in priority if c and c in df.columns]
rest     = [c for c in df.columns if c not in priority]
df       = df[priority + rest]

# ── 7. Save ────────────────────────────────────────────────────────────────────
df.to_csv(OUT_PATH, index=False)

print(f"\n{'='*60}")
print(f"Post-processing complete.")
print(f"  Final shape    : {df.shape}")
print(f"  Size on disk   : {os.path.getsize(OUT_PATH)/1024/1024:.1f} MB")
print(f"  Columns        : {df.columns.tolist()}")
print(f"\nSample (first 5 rows):")
preview = [c for c in ['ticker','owner_name','officer_title','is_officer',
                        'transaction_date','transaction_code','acquired_disposed',
                        'shares','price_per_share'] if c in df.columns]
print(df[preview].head().to_string(index=False))
print(f"\nSaved -> {OUT_PATH}")

	# postprocess_form4.py
	# Cleans up form4_transactions.csv:
	# 1. Drops footnote columns (*_fn)
	# 2. Converts date format from 30-MAR-2006 → 2006-03-30
	# 3. Re-fetches is_officer / is_director / officer_title from one sample
	# quarter to confirm correct column names, then re-downloads all quarters
	# to add these columns to the existing data
	# 4. Renames columns for clarity
	# 5. Overwrites form4_transactions.csv with the clean version
	#
	# Run: python scripts/postprocess_form4.py

	import io
	import os
	import time
	import zipfile
	import requests
	import pandas as pd
	from tqdm import tqdm

	# ── Paths ──────────────────────────────────────────────────────────────────────
	OUT_PATH = r"D:\UoE AI\Dissertation\IPP Draft\datasets\form4_transactions.csv"
	HEADERS = {'User-Agent': 'S2880814 University of Edinburgh s.g.vishnu@sms.ed.ac.uk'}
	SLEEP = 0.5
	TIMEOUT = 120

	# ── 1. Load existing data ──────────────────────────────────────────────────────
	print("Loading form4_transactions.csv...")
	df = pd.read_csv(OUT_PATH, low_memory=False)
	print(f" {len(df):,} rows, {len(df.columns)} columns")

	# ── 2. Drop footnote columns ───────────────────────────────────────────────────
	fn_cols = [c for c in df.columns if c.endswith('_fn')]
	df.drop(columns=fn_cols, inplace=True)
	print(f" Dropped {len(fn_cols)} footnote columns → {len(df.columns)} columns remain")

	# ── 3. Fix date format: 30-MAR-2006 → 2006-03-30 ─────────────────────────────
	print(" Converting dates...")
	df['transaction_date'] = pd.to_datetime(
	df['transaction_date'], format='mixed', dayfirst=True, errors='coerce'
	).dt.strftime('%Y-%m-%d')
	print(f" Sample dates after fix: {df['transaction_date'].dropna().head(3).tolist()}")

	# ── 4. Discover correct REPORTINGOWNER column names from one sample quarter ────
	print("\nChecking REPORTINGOWNER column names from 2024 Q1...")
	sample_url = "https://www.sec.gov/files/structureddata/data/insider-transactions-data-sets/2024q1_form345.zip"
	r = requests.get(sample_url, headers=HEADERS, timeout=TIMEOUT)
	zf = zipfile.ZipFile(io.BytesIO(r.content))
	owner_file = next((n for n in zf.namelist() if 'reportingowner' in n.lower()), None)
	sample_own = pd.read_csv(io.BytesIO(zf.read(owner_file)), sep='\t', low_memory=False, nrows=5)
	sample_own.columns = sample_own.columns.str.lower().str.strip()
	print(f" REPORTINGOWNER columns: {sample_own.columns.tolist()}")

	# Identify the correct column names
	col_map = {}
	for col in sample_own.columns:
	if col in ('isdir', 'isdirector'):
	col_map['is_director'] = col
	elif col == 'isofficer':
	col_map['is_officer'] = col
	elif col == 'istenpercentowner':
	col_map['is_ten_pct_owner'] = col
	elif col == 'officertitle':
	col_map['officer_title'] = col
	elif 'accession' in col:
	col_map['accession_col'] = col

	print(f" Mapped columns: {col_map}")

	# ── 5. Re-download all quarters to get owner role columns ─────────────────────
	def quarter_urls():
	urls = []
	year, qtr = 2006, 1
	while (year, qtr) <= (2026, 1):
	url = f"https://www.sec.gov/files/structureddata/data/insider-transactions-data-sets/{year}q{qtr}_form345.zip"
	urls.append((year, qtr, url))
	qtr += 1
	if qtr > 4:
	qtr, year = 1, year + 1
	return urls

	# Get accession numbers already in our dataset
	our_accessions = set(df['accession_number'].dropna())
	print(f"\nRe-fetching owner role columns for {len(our_accessions):,} accessions...")

	acc_col = col_map.get('accession_col', 'accession_number')
	dir_col = col_map.get('is_director')
	off_col = col_map.get('is_officer')
	pct_col = col_map.get('is_ten_pct_owner')
	title_col = col_map.get('officer_title')
	keep_own = [c for c in [acc_col, 'rptownercik', dir_col, off_col, pct_col, title_col] if c]

	owner_frames = []

	for year, qtr, url in tqdm(quarter_urls(), desc="Re-fetching owner data"):
	time.sleep(SLEEP)
	try:
	r = requests.get(url, headers=HEADERS, timeout=TIMEOUT)
	if r.status_code != 200:
	continue
	zf = zipfile.ZipFile(io.BytesIO(r.content))
	own_file = next((n for n in zf.namelist() if 'reportingowner' in n.lower()), None)
	if not own_file:
	continue
	own = pd.read_csv(io.BytesIO(zf.read(own_file)), sep='\t', low_memory=False, on_bad_lines='skip')
	own.columns = own.columns.str.lower().str.strip()

	acc_col_q = next((c for c in own.columns if 'accession' in c), None)
	if not acc_col_q:
	continue

	# Only keep rows for our accessions
	own = own[own[acc_col_q].isin(our_accessions)]
	if own.empty:
	continue

	cols_to_keep = [c for c in [acc_col_q, 'rptownercik', dir_col, off_col, pct_col, title_col] if c and c in own.columns]
	owner_frames.append(own[cols_to_keep].copy())

	except Exception as e:
	tqdm.write(f" [WARN] {year} Q{qtr}: {e}")
	continue

	if owner_frames:
	owner_df = pd.concat(owner_frames, ignore_index=True)
	owner_df.columns = owner_df.columns.str.lower().str.strip()

	# Rename to standard names
	rename_own = {}
	if dir_col and dir_col in owner_df.columns: rename_own[dir_col] = 'is_director'
	if off_col and off_col in owner_df.columns: rename_own[off_col] = 'is_officer'
	if pct_col and pct_col in owner_df.columns: rename_own[pct_col] = 'is_ten_pct_owner'
	if title_col and title_col in owner_df.columns: rename_own[title_col] = 'officer_title'
	owner_df.rename(columns=rename_own, inplace=True)

	# Deduplicate: one row per accession + rptownercik
	acc_col_own = next((c for c in owner_df.columns if 'accession' in c), None)
	if acc_col_own:
	owner_df = owner_df.drop_duplicates(subset=[acc_col_own, 'rptownercik'])

	print(f" Owner rows fetched: {len(owner_df):,}")
	print(f" Owner columns: {owner_df.columns.tolist()}")

	# Merge onto main df
	df = df.merge(
	owner_df,
	left_on=['accession_number', 'rptownercik'],
	right_on=[acc_col_own, 'rptownercik'],
	how='left'
	)
	# Drop duplicate accession col if created
	if acc_col_own != 'accession_number' and acc_col_own in df.columns:
	df.drop(columns=[acc_col_own], inplace=True)

	print(f" is_officer populated: {df['is_officer'].notna().sum():,} rows" if 'is_officer' in df.columns else " [WARN] is_officer not in merged result")
	else:
	print(" [WARN] No owner role data retrieved — skipping merge")

	# ── 6. Final column rename + reorder ──────────────────────────────────────────
	rename_final = {
	'shrs_ownd_folwng_trans': 'shares_after_txn',
	'valu_ownd_folwng_trans': 'value_after_txn',
	'direct_indirect_ownership': 'ownership_type',
	'trans_form_type': 'form_type',
	'nonderiv_trans_sk': 'trans_sk',
	}
	df.rename(columns={k: v for k, v in rename_final.items() if k in df.columns}, inplace=True)

	# Preferred column order
	priority = ['ticker', 'issuer_cik', 'accession_number', 'filing_date' if 'filing_date' in df.columns else None,
	'owner_name', 'rptownercik', 'is_officer', 'is_director', 'is_ten_pct_owner', 'officer_title',
	'transaction_date', 'transaction_code', 'acquired_disposed',
	'shares', 'price_per_share', 'shares_after_txn', 'value_after_txn',
	'security_title', 'ownership_type', 'form_type']
	priority = [c for c in priority if c and c in df.columns]
	rest = [c for c in df.columns if c not in priority]
	df = df[priority + rest]

	# ── 7. Save ────────────────────────────────────────────────────────────────────
	df.to_csv(OUT_PATH, index=False)

	print(f"\n{'='*60}")
	print(f"Post-processing complete.")
	print(f" Final shape : {df.shape}")
	print(f" Size on disk : {os.path.getsize(OUT_PATH)/1024/1024:.1f} MB")
	print(f" Columns : {df.columns.tolist()}")
	print(f"\nSample (first 5 rows):")
	preview = [c for c in ['ticker','owner_name','officer_title','is_officer',
	'transaction_date','transaction_code','acquired_disposed',
	'shares','price_per_share'] if c in df.columns]
	print(df[preview].head().to_string(index=False))
	print(f"\nSaved -> {OUT_PATH}")