Upload scripts/collect_form4.py with huggingface_hub

scripts/collect_form4.py

@@ -0,0 +1,242 @@
+# 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))