PaPP v2: cartella Analisi (script, risultati, metodologia incroci)

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
2026-06-17 07:40:41 +00:00
parent 2070d44ce6
commit 19c2e0ac72
11 changed files with 639 additions and 0 deletions
@@ -0,0 +1,69 @@
+"""Fase 2 raffinata: extra-rendimento vs baseline + significativita'.
+
+Per ogni (coppia, direzione) calcola:
+  - rendimento medio grezzo a 1/3/5/10/20 g
+  - EXTRA-rendimento = evento - baseline nello stesso regime (toglie il bias di periodo)
+  - effetto in sigma del movimento tipico a h giorni
+  - quota di eventi che battono la baseline
+  - p-value (block bootstrap) e q-value (Benjamini-Hochberg)
+
+Uso:
+  python excess_analysis.py [SYMBOL] [YEAR_MIN]
+Output:
+  ../results/summary_<SYM>_<YEAR>.csv  + stampa dei top a 5/10/20 g
+"""
+import os, sys, numpy as np, pandas as pd
+from _common import load_pair, add_regime, regime_edges, block_bootstrap_p, benjamini_hochberg, HZ, HERE
+
+SYM = sys.argv[1] if len(sys.argv) > 1 else "EURUSD"
+YEAR = int(sys.argv[2]) if len(sys.argv) > 2 else 1999
+RESULTS = os.path.normpath(os.path.join(HERE, "..", "results"))
+
+cr, ba = load_pair(SYM, YEAR)
+print(f"[{SYM} {YEAR}+] incroci={len(cr)}  baseline={len(ba)}")
+
+edges = regime_edges(ba)
+cr = add_regime(cr, edges)
+ba = add_regime(ba, edges)
+
+base_mean = {h: ba.groupby("regime")[f"cret_{h}"].mean() for h in HZ}
+base_glob = {h: ba[f"cret_{h}"].mean() for h in HZ}
+base_sd   = {h: ba[f"cret_{h}"].std()  for h in HZ}
+
+rows = []
+for (pair, d), g in cr.groupby(["pair", "dir"]):
+    n = len(g)
+    if n < 30:
+        continue
+    rec = {"pair": pair, "dir": int(d), "n": n}
+    for h in HZ:
+        ev = g[f"cret_{h}"].values
+        bexp = g["regime"].map(base_mean[h]).fillna(base_glob[h]).values
+        exc = ev - bexp
+        rec[f"raw_{h}"] = np.nanmean(ev)
+        rec[f"exc_{h}"] = np.nanmean(exc)
+        rec[f"eff_{h}"] = np.nanmean(exc) / base_sd[h]
+        rec[f"pos_{h}"] = (exc > 0).mean()
+        rec[f"p_{h}"]   = block_bootstrap_p(exc)
+    rows.append(rec)
+
+res = pd.DataFrame(rows)
+for h in HZ:
+    res[f"q_{h}"] = benjamini_hochberg(res[f"p_{h}"].values)
+
+os.makedirs(RESULTS, exist_ok=True)
+outp = os.path.join(RESULTS, f"summary_{SYM}_{YEAR}.csv")
+res.to_csv(outp, index=False)
+
+print(f"vol giornaliera prezzo (sd cret_1) = {ba['cret_1'].std():.3f}%")
+print(f"baseline drift 5/10/20g = {base_glob[5]:.3f}% / {base_glob[10]:.3f}% / {base_glob[20]:.3f}%")
+for h in [5, 10, 20]:
+    sig = res[res[f"q_{h}"] < 0.10].copy()
+    sig["ae"] = sig[f"eff_{h}"].abs()
+    sig = sig.sort_values("ae", ascending=False).head(12)
+    print(f"\n=== TOP {h}g (q<0.10) ===")
+    for _, r in sig.iterrows():
+        print(f"  {r['pair']:13s} dir={int(r['dir']):+d} n={int(r['n']):5d} "
+              f"exc={r[f'exc_{h}']:+.3f}% eff={r[f'eff_{h}']:+.2f}sd "
+              f"pos={r[f'pos_{h}']*100:.0f}% q={r[f'q_{h}']:.3f}")
+print(f"\nsalvato: {outp}")