Upload scripts/collect_form4_topup.py with huggingface_hub

scripts/collect_form4_topup.py

@@ -0,0 +1,251 @@
+# collect_form4_topup.py
+# Collects Form 4 insider transactions for the 22 CIKs identified by Sean Choi
+# (PhD verifier) that are in the ground-truth S&P 500 list but were missing
+# from the original sp500_union_constituents(1).csv / crsp_dsenames.csv universe.
+#
+# These 22 represent ~4% of the full S&P 500 universe that were not covered in
+# the original 656-CIK collection run.  This script:
+#   1. Downloads all 81 quarters from SEC EDGAR Insider Transactions flat files
+#   2. Filters to the 22 new CIKs only
+#   3. Joins REPORTINGOWNER role columns (is_officer, is_director, is_ten_pct_owner)
+#   4. Applies the same post-processing as the main pipeline
+#   5. APPENDS to the existing datasets/form4_transactions.csv
+#
+# Run AFTER the main pipeline (collect_form4.py → postprocess_form4.py →
+# form4_add_roles.py) has completed successfully.
+#
+# Run: python scripts/collect_form4_topup.py
+
+import io
+import time
+import zipfile
+import requests
+import pandas as pd
+from tqdm import tqdm
+
+# ── Config ─────────────────────────────────────────────────────────────────────
+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     = 0.12   # ~8 req/sec — well under EDGAR 10 req/sec limit
+TIMEOUT   = 120
+
+# ── 22 missing CIK → Ticker pairs (from Sean Choi, 02 Jun 2026) ───────────────
+NEW_UNIVERSE = {
+    1725057: 'DAY',
+    14693:   'BF.B',
+    827099:  'CPWR',
+    1751008: 'APP',
+    1176948: 'ARES',
+    1679788: 'COIN',
+    849395:  'CRH',
+    1690820: 'CVNA',
+    1792789: 'DASH',
+    1561550: 'DDOG',
+    105634:  'EME',
+    895126:  'EXE',
+    1035983: 'FIX',
+    1783869: 'HOOD',
+    1381197: 'IBKR',
+    62709:   'MRSH',
+    2041610: 'PSKY',
+    2064953: 'SOLS',
+    1973266: 'TKO',
+    1671933: 'TTD',
+    719955:  'WSM',
+    1512673: 'XYZ',
+}
+
+# Zero-pad CIKs to 10 digits (EDGAR format)
+new_ciks = {str(cik).zfill(10) for cik in NEW_UNIVERSE}
+cik_to_ticker = {str(cik).zfill(10): tkr for cik, tkr in NEW_UNIVERSE.items()}
+
+print(f"Top-up target: {len(new_ciks)} CIKs")
+print(f"CIKs: {sorted(new_ciks)}")
+
+
+# ── Quarter URL generator (2006 Q1 → 2026 Q1) ─────────────────────────────────
+def quarter_urls():
+    urls, year, qtr = [], 2006, 1
+    while (year, qtr) <= (2026, 1):
+        urls.append((year, qtr,
+            f"https://www.sec.gov/files/structureddata/data/insider-transactions-data-sets/{year}q{qtr}_form345.zip"))
+        qtr += 1
+        if qtr > 4:
+            qtr, year = 1, year + 1
+    return urls
+
+
+# ── Pass 1: Collect SUBMISSION + NONDERIV_TRANS ───────────────────────────────
+print("\n" + "=" * 60)
+print("Pass 1 of 2: SUBMISSION + NONDERIV_TRANS (transactions)")
+print("=" * 60)
+
+trans_frames = []
+found_accessions = set()
+
+for year, qtr, url in tqdm(quarter_urls(), desc="Quarters (transactions)"):
+    time.sleep(SLEEP)
+    try:
+        r = requests.get(url, headers=HEADERS, timeout=TIMEOUT)
+        if r.status_code != 200:
+            tqdm.write(f"  [SKIP] {year} Q{qtr} — HTTP {r.status_code}")
+            continue
+
+        zf = zipfile.ZipFile(io.BytesIO(r.content))
+        names_lc = {n.lower(): n for n in zf.namelist()}
+
+        # ── SUBMISSION ────────────────────────────────────────────────────────
+        sub_file = next((v for k, v in names_lc.items() if 'submission' in k), None)
+        if not sub_file:
+            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()
+
+        # issuer_cik in SUBMISSION is stored as an integer — zero-pad it
+        if 'issuercik' in sub.columns:
+            sub['issuercik'] = sub['issuercik'].astype(str).str.strip().str.zfill(10)
+        sub_match = sub[sub['issuercik'].isin(new_ciks)]
+        if sub_match.empty:
+            continue
+
+        matched_acc = set(sub_match['accession_number'].dropna())
+
+        # ── NONDERIV_TRANS ────────────────────────────────────────────────────
+        nd_file = next((v for k, v in names_lc.items() if 'nonderiv_trans' in k), None)
+        if not nd_file:
+            continue
+        nd = pd.read_csv(io.BytesIO(zf.read(nd_file)), sep='\t',
+                         low_memory=False, on_bad_lines='skip')
+        nd.columns = nd.columns.str.lower().str.strip()
+        nd = nd[nd['accession_number'].isin(matched_acc)]
+        if nd.empty:
+            continue
+
+        # Join submission metadata
+        sub_keep = [c for c in ['accession_number', 'issuercik', 'rptownercik',
+                                 'rptowner_name', 'period_of_report'] if c in sub_match.columns]
+        merged = nd.merge(sub_match[sub_keep].drop_duplicates('accession_number'),
+                          on='accession_number', how='left')
+
+        # Map ticker
+        merged['ticker'] = merged['issuercik'].map(cik_to_ticker)
+        merged.rename(columns={'issuercik': 'issuer_cik',
+                                'rptowner_name': 'owner_name'}, inplace=True)
+
+        trans_frames.append(merged)
+        found_accessions.update(matched_acc)
+        tqdm.write(f"  {year} Q{qtr}: {len(nd):,} transactions, {len(matched_acc)} filings")
+
+    except Exception as e:
+        tqdm.write(f"  [WARN] {year} Q{qtr}: {e}")
+
+if not trans_frames:
+    print("\n[ERROR] No transactions found for any of the 22 new CIKs.")
+    raise SystemExit(1)
+
+df_new = pd.concat(trans_frames, ignore_index=True)
+print(f"\nPass 1 complete: {len(df_new):,} transactions across {df_new['issuer_cik'].nunique()} companies")
+print(f"Unique accession numbers: {len(found_accessions):,}")
+
+
+# ── Pass 2: Collect REPORTINGOWNER for role flags ─────────────────────────────
+print("\n" + "=" * 60)
+print("Pass 2 of 2: REPORTINGOWNER (is_officer / is_director / is_ten_pct_owner)")
+print("=" * 60)
+
+owner_frames = []
+
+for year, qtr, url in tqdm(quarter_urls(), desc="Quarters (owner roles)"):
+    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))
+        names_lc = {n.lower(): n for n in zf.namelist()}
+
+        own_file = next((v for k, v in names_lc.items() if 'reportingowner' in k), 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()
+
+        own = own[own['accession_number'].isin(found_accessions)]
+        if own.empty:
+            continue
+
+        keep = [c for c in ['accession_number', 'rptownercik',
+                             'rptowner_relationship', 'rptowner_title'] if c in own.columns]
+        owner_frames.append(own[keep].copy())
+
+    except Exception as e:
+        tqdm.write(f"  [WARN] {year} Q{qtr}: {e}")
+
+# ── Build roles lookup ────────────────────────────────────────────────────────
+if owner_frames:
+    owner_df = pd.concat(owner_frames, ignore_index=True)
+    owner_df = owner_df.drop_duplicates(subset=['accession_number', 'rptownercik'])
+
+    rel = owner_df['rptowner_relationship'].fillna('')
+    owner_df['is_officer']       = rel.str.contains('Officer',        case=False).astype(int)
+    owner_df['is_director']      = rel.str.contains('Director',       case=False).astype(int)
+    owner_df['is_ten_pct_owner'] = rel.str.contains('TenPercentOwner',case=False).astype(int)
+    if 'rptowner_title' in owner_df.columns:
+        owner_df.rename(columns={'rptowner_title': 'officer_title'}, inplace=True)
+    owner_df.drop(columns=['rptowner_relationship'], inplace=True)
+
+    print(f"  Owner rows: {len(owner_df):,}")
+    print(f"  Officers  : {owner_df['is_officer'].sum():,}")
+    print(f"  Directors : {owner_df['is_director'].sum():,}")
+
+    df_new = df_new.merge(owner_df, on=['accession_number', 'rptownercik'], how='left')
+else:
+    print("  [WARN] No owner role data retrieved — role columns will be NaN")
+    for col in ['is_officer', 'is_director', 'is_ten_pct_owner', 'officer_title']:
+        df_new[col] = pd.NA
+
+
+# ── Post-process dates ────────────────────────────────────────────────────────
+print("\nPost-processing dates...")
+for date_col in ['transaction_date', 'deemed_execution_date']:
+    if date_col in df_new.columns:
+        df_new[date_col] = pd.to_datetime(
+            df_new[date_col], format='mixed', dayfirst=True, errors='coerce'
+        ).dt.strftime('%Y-%m-%d')
+
+# Drop footnote columns
+fn_cols = [c for c in df_new.columns if c.endswith('_fn')]
+df_new.drop(columns=fn_cols, inplace=True)
+
+
+# ── Load existing and append ──────────────────────────────────────────────────
+print("\nLoading existing form4_transactions.csv...")
+df_existing = pd.read_csv(OUT_PATH, low_memory=False)
+print(f"  Existing: {len(df_existing):,} rows, {len(df_existing.columns)} columns")
+
+# Align columns — add any missing cols as NaN in either direction
+all_cols = list(df_existing.columns)
+for c in all_cols:
+    if c not in df_new.columns:
+        df_new[c] = pd.NA
+df_new = df_new[all_cols]   # reorder to match existing
+
+# Guard against duplicate accessions (e.g., companies that were already in the 656)
+existing_acc = set(df_existing['accession_number'].dropna())
+df_new = df_new[~df_new['accession_number'].isin(existing_acc)]
+print(f"  New rows after dedup: {len(df_new):,}")
+
+df_combined = pd.concat([df_existing, df_new], ignore_index=True)
+df_combined.to_csv(OUT_PATH, index=False)
+
+print(f"\n{'='*60}")
+print(f"Top-up complete.")
+print(f"  Previous rows : {len(df_existing):,}")
+print(f"  Added rows    : {len(df_new):,}")
+print(f"  Total rows    : {len(df_combined):,}")
+print(f"  Companies     : {df_combined['ticker'].nunique()}")
+print(f"  Date range    : {df_combined['transaction_date'].min()} → {df_combined['transaction_date'].max()}")
+print(f"\nSaved → {OUT_PATH}")