# collect_form4.py # Collects SEC Form 4 (insider trading) data for all 656 S&P 500 CIKs. # # Source: SEC EDGAR Insider Transactions Data Sets (official quarterly flat files) # https://www.sec.gov/data-research/sec-markets-data/insider-transactions-data-sets # # The SEC pre-parses all Form 3/4/5 XML submissions into quarterly ZIP files. # This is vastly faster than scraping individual XMLs (~20 min vs ~11 hours). # # Method: # 1. Download quarterly ZIP files for 2006 Q1 → 2026 Q1 (~80 files) # 2. From each ZIP, read SUBMISSION.tsv + REPORTINGOWNER.tsv + NONDERIV_TRANS.tsv # 3. Filter SUBMISSION rows where issuer_cik is in our 656-company universe # 4. Join owner and transaction details # 5. Output clean form4_transactions.csv # # Output: # datasets/form4_transactions.csv — one row per non-derivative transaction # # Run: python scripts/collect_form4.py import io import os import time import zipfile import requests import pandas as pd from tqdm import tqdm # ── Paths ────────────────────────────────────────────────────────────────────── UNIVERSE_PATH = r"D:\UoE AI\Dissertation\IPP Draft\datasets\sp500_union_constituents(1).csv" OUT_PATH = r"D:\UoE AI\Dissertation\IPP Draft\datasets\form4_transactions.csv" # ── Config ───────────────────────────────────────────────────────────────────── HEADERS = {'User-Agent': 'S2880814 University of Edinburgh s.g.vishnu@sms.ed.ac.uk'} SLEEP = 0.5 # conservative — these are bulk files, not per-filing requests TIMEOUT = 120 # ZIP files can be large, allow more time # Date range: 2006 Q1 → 2026 Q1 START_YEAR, START_QTR = 2006, 1 END_YEAR, END_QTR = 2026, 1 # ── 1. Load S&P 500 universe ─────────────────────────────────────────────────── print("Loading S&P 500 universe...") universe = pd.read_csv(UNIVERSE_PATH) cik_col = [c for c in universe.columns if 'cik' in c.lower()][0] tkr_col = [c for c in universe.columns if any(k in c.lower() for k in ('ticker', 'symbol', 'tic'))][0] universe[cik_col] = universe[cik_col].astype(int) cik_set = set(universe[cik_col].dropna().astype(int)) cik2tkr = dict(zip(universe[cik_col], universe[tkr_col])) print(f" {len(cik_set)} unique CIKs loaded") # ── 2. Build list of quarterly ZIP URLs ─────────────────────────────────────── # SEC URL pattern: https://www.sec.gov/files/dera/data/insider-transactions/{YYYY}q{Q}_form345.zip def quarter_urls(): urls = [] year, qtr = START_YEAR, START_QTR while (year, qtr) <= (END_YEAR, END_QTR): 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 = 1 year += 1 return urls quarters = quarter_urls() print(f"\n{len(quarters)} quarterly ZIP files to process (2006 Q1 → 2026 Q1)") # ── 3. Process each quarterly ZIP ───────────────────────────────────────────── all_records = [] failed_quarters = [] for year, qtr, url in tqdm(quarters, desc="Quarters"): time.sleep(SLEEP) # Download ZIP into memory try: r = requests.get(url, headers=HEADERS, timeout=TIMEOUT) except requests.RequestException as e: tqdm.write(f" [SKIP] {year} Q{qtr} — request error: {e}") failed_quarters.append((year, qtr)) continue if r.status_code == 404: tqdm.write(f" [SKIP] {year} Q{qtr} — not yet published (404)") continue if r.status_code != 200: tqdm.write(f" [SKIP] {year} Q{qtr} — HTTP {r.status_code}") failed_quarters.append((year, qtr)) continue try: zf = zipfile.ZipFile(io.BytesIO(r.content)) except zipfile.BadZipFile: tqdm.write(f" [SKIP] {year} Q{qtr} — bad ZIP file") failed_quarters.append((year, qtr)) continue zip_names = zf.namelist() # ── Read SUBMISSION table (issuer CIK + filing metadata) ────────────────── sub_file = next((n for n in zip_names if 'submission' in n.lower()), None) if not sub_file: tqdm.write(f" [SKIP] {year} Q{qtr} — SUBMISSION file not found in ZIP") continue sub = pd.read_csv( io.BytesIO(zf.read(sub_file)), sep='\t', low_memory=False, on_bad_lines='skip' ) sub.columns = sub.columns.str.lower().str.strip() # Identify issuer CIK column (varies slightly across years) issuer_col = next((c for c in sub.columns if 'issuer' in c and 'cik' in c), None) \ or next((c for c in sub.columns if c == 'issuercik'), None) if not issuer_col: tqdm.write(f" [SKIP] {year} Q{qtr} — can't find issuer CIK column. Cols: {sub.columns.tolist()[:8]}") continue sub[issuer_col] = pd.to_numeric(sub[issuer_col], errors='coerce') sub_filtered = sub[sub[issuer_col].isin(cik_set)].copy() if sub_filtered.empty: continue # none of our companies filed this quarter accessions = set(sub_filtered['accession_number'] if 'accession_number' in sub_filtered.columns else sub_filtered[sub_filtered.columns[0]]) # ── Read REPORTINGOWNER table (insider details) ──────────────────────────── owner_file = next((n for n in zip_names if 'reportingowner' in n.lower()), None) if owner_file: own = pd.read_csv( io.BytesIO(zf.read(owner_file)), sep='\t', low_memory=False, on_bad_lines='skip' ) own.columns = own.columns.str.lower().str.strip() acc_col_own = next((c for c in own.columns if 'accession' in c), None) if acc_col_own: own = own[own[acc_col_own].isin(accessions)] else: own = pd.DataFrame() # ── Read NONDERIV_TRANS table (actual buy/sell transactions) ─────────────── txn_file = next((n for n in zip_names if 'nonderiv_trans' in n.lower()), None) if not txn_file: continue txn = pd.read_csv( io.BytesIO(zf.read(txn_file)), sep='\t', low_memory=False, on_bad_lines='skip' ) txn.columns = txn.columns.str.lower().str.strip() acc_col_txn = next((c for c in txn.columns if 'accession' in c), None) if not acc_col_txn: continue txn = txn[txn[acc_col_txn].isin(accessions)].copy() if txn.empty: continue # ── Join: txn + submission (issuer info) + owner (insider info) ─────────── # Merge txn with submission on accession number merge_on = acc_col_txn sub_cols = [issuer_col, 'accession_number' if 'accession_number' in sub_filtered.columns else sub_filtered.columns[0], 'periodofreport', 'filingdate'] if 'filingdate' in sub_filtered.columns \ else [issuer_col, sub_filtered.columns[0]] # Keep only useful submission columns that exist useful_sub = [c for c in [issuer_col, 'accession_number', 'periodofreport', 'filingdate', 'reporttype'] if c in sub_filtered.columns] merged = txn.merge( sub_filtered[useful_sub].rename(columns={issuer_col: 'issuer_cik'}), left_on=acc_col_txn, right_on='accession_number', how='left' ) # Merge with owner info if available if not own.empty and acc_col_own in own.columns: useful_own = [c for c in [acc_col_own, 'rptownername', 'rptownercik', 'isofficer', 'isdirector', 'istenpercentowner', 'officertitle'] if c in own.columns] merged = merged.merge( own[useful_own], left_on=acc_col_txn, right_on=acc_col_own, how='left' ) # Add ticker from our universe mapping if 'issuer_cik' in merged.columns: merged['ticker'] = merged['issuer_cik'].map(cik2tkr) all_records.append(merged) tqdm.write(f" {year} Q{qtr}: {len(merged):,} transactions for {sub_filtered[issuer_col].nunique()} companies") # ── 4. Combine and save ──────────────────────────────────────────────────────── if not all_records: print("\nNo records collected — check URL pattern or network connection.") else: df = pd.concat(all_records, ignore_index=True) # Standardise key column names where possible rename_map = { 'trans_date': 'transaction_date', 'transaction_date': 'transaction_date', 'trans_shares': 'shares', 'transaction_shares':'shares', 'trans_pricepershare': 'price_per_share', 'trans_acquired_disp_cd': 'acquired_disposed', 'trans_code': 'transaction_code', 'rptownername': 'owner_name', 'isofficer': 'is_officer', 'isdirector': 'is_director', 'istenpercentowner': 'is_ten_pct_owner', 'officertitle': 'officer_title', 'filingdate': 'filing_date', } df.rename(columns={k: v for k, v in rename_map.items() if k in df.columns}, inplace=True) df.to_csv(OUT_PATH, index=False) print(f"\n{'='*60}") print(f"Form 4 collection complete.") print(f" Total transactions : {len(df):,}") print(f" Unique companies : {df['issuer_cik'].nunique() if 'issuer_cik' in df.columns else 'N/A'}") print(f" Columns : {df.columns.tolist()}") if failed_quarters: print(f" Failed quarters : {failed_quarters}") print(f"\nSaved -> {OUT_PATH}") print("\nSample (first 5 rows):") preview_cols = [c for c in ['ticker', 'owner_name', 'officer_title', 'transaction_date', 'transaction_code', 'acquired_disposed', 'shares', 'price_per_share'] if c in df.columns] print(df[preview_cols].head().to_string(index=False))