fcostalonga/Trading
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
f5c668f70fde3bdb3e71ced3cab86c6c797e24d4
Trading/Trading Pattern Recon w Hurst - Forex.py

848 lines
29 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# -*- coding: utf-8 -*-
"""Trading Pattern Recon w Hurst - Forex
VERSIONE CORRETTA (EOD Close-to-Close) – con SHORT, Wavelet denoise (DB) su EstOutcome,
fee 1bp e cap esposizione per singola valuta 35%.
REQUISITI IMPLEMENTATI (come da tue istruzioni):
1) Denoise su EstOutcome (rolling/online, NO look-ahead)
2) Wavelet Daubechies (DB) -> db4
3) Operatività EOD: decisione a close(t), esecuzione a close(t)
4) PnL sempre su close(t+1)/close(t) (cioè Ret_fwd = close[t+1]/close[t]-1)
5) Fee = 1 bp applicata sui cambi di posizione (enter/exit/flip) al close(t)
6) Cap esposizione per singola valuta = 35% (net exposure per currency)
7) Per il resto: mantiene lo schema originale (shared_utils, per-asset backtest, ranking, Top15, EW e Risk Parity)
NOTA:
- Lo script usa SOLO serie AdjClose dal tuo endpoint (stesso JSON di prima).
- Non usa Open/High/Low e non introduce logiche intraday.
Output principali (cartella OUT_DIR):
- forward_bt_signals.csv (per-ticker: Signal, EstOutcomeRaw, EstOutcomeDenoised, Ret_fwd, PnL)
- forward_bt_summary.csv (metriche per ticker)
- ranking_score.csv (ranking + score)
- topn_tickers.csv (TopN dal ranking)
- portfolio_daily.csv (ret/eq line EW e RP)
- weights_eq_active_capped.csv / weights_rp_active_capped.csv (pesi con cash + cap valuta)
- currency_exposure_eq.csv / currency_exposure_rp.csv (diagnostica esposizioni)
Dipendenze:
- numpy, pandas, matplotlib, requests
- PyWavelets (pip install PyWavelets)
"""
from __future__ import annotations
import sys
import types
from pathlib import Path
from urllib.parse import quote
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import requests
# -------------------------
# Wavelets
# -------------------------
try:
import pywt
except ImportError as e:
raise ImportError("Manca PyWavelets (pywt). Installa con: pip install PyWavelets") from e
# ------------------------------------------------------------
# shared_utils import (local file next to this script)
# ------------------------------------------------------------
# Alcune versioni di shared_utils importano moduli opzionali (es. pyodbc). Evitiamo crash.
sys.modules.setdefault("pyodbc", types.SimpleNamespace())
import importlib.util
SHARED_UTILS_PATH = Path(__file__).with_name("shared_utils.py")
if not SHARED_UTILS_PATH.exists():
raise FileNotFoundError(f"shared_utils.py non trovato accanto allo script: {SHARED_UTILS_PATH}")
spec = importlib.util.spec_from_file_location("shared_utils", str(SHARED_UTILS_PATH))
shared_utils = importlib.util.module_from_spec(spec)
sys.modules["shared_utils"] = shared_utils
assert spec.loader is not None
spec.loader.exec_module(shared_utils)
build_pattern_library = shared_utils.build_pattern_library
predict_from_library = shared_utils.predict_from_library
z_norm = shared_utils.z_norm
# ============================================================
# CONFIG
# ============================================================
TICKERS = [
"EURUSD Curncy",
"USDJPY Curncy",
"GBPUSD Curncy",
"USDCHF Curncy",
"USDCAD Curncy",
"AUDUSD Curncy",
"NZDUSD Curncy",
"EURJPY Curncy",
"EURGBP Curncy",
"EURCHF Curncy",
"EURCAD Curncy",
"EURAUD Curncy",
"EURNZD Curncy",
"GBPJPY Curncy",
"GBPCHF Curncy",
"GBPCAD Curncy",
"GBPAUD Curncy",
"GBPNZD Curncy",
"CHFJPY Curncy",
"CADJPY Curncy",
"AUDJPY Curncy",
"NZDJPY Curncy",
"AUDNZD Curncy",
"AUDCAD Curncy",
"AUDCHF Curncy",
"NZDCAD Curncy",
"NZDCHF Curncy",
"CADCHF Curncy",
]
TICKERS = [t.strip() for t in TICKERS if isinstance(t, str) and t.strip()]
BASE_URL = "https://fin.scorer.app/finance/v2/history"
FROM_DATE = "20201224"
# Pattern params (come nello schema originale)
WP = 60
HA = 10
KNN_K = 25
# Short attivo per FX
ALLOW_SHORT = True
# Fee (FIX: 1 bp)
FEE_BPS = 1
# Exit controls (EOD approximation su Ret_fwd)
SL_BPS = 300.0
TP_BPS = 800.0
TRAIL_BPS = 300.0
TIME_STOP_BARS = 20
THETA_EXIT = 0.0
# Theta entry fallback se Hurst non disponibile
THETA_FALLBACK = 0.005
# Portfolio construction
TOP_N = 15
RP_MAX_WEIGHT = 2.0 / TOP_N # 0.1333... (come in legenda che avevi visto)
RANKING_WINDOW_BARS = 252
RP_LOOKBACK = 60
# Currency cap (net exposure per currency)
CURRENCY_CAP = 0.35
# Wavelet denoise su EstOutcome
DENOISE_ENABLED = True
DENOISE_WAVELET = "db4" # DB family (Daubechies)
DENOISE_LEVEL = 3
DENOISE_MIN_LEN = 96
DENOISE_THRESHOLD_MODE = "soft"
DAYS_PER_YEAR = 252
OUT_DIR = Path("./out_shared_utils_url_forex")
OUT_DIR.mkdir(parents=True, exist_ok=True)
# ============================================================
# Helpers: JSON -> DataFrame (AdjClose)
# ============================================================
def _detect_col(cols, candidates):
cols_l = {c.lower(): c for c in cols}
for cand in candidates:
if cand.lower() in cols_l:
return cols_l[cand.lower()]
# fallback: contains
for cand in candidates:
for c in cols:
if cand.lower() in str(c).lower():
return c
return None
def fetch_price_series(ticker: str, from_date: str) -> pd.DataFrame:
"""Scarica il JSON e restituisce DataFrame indicizzato per Date con colonna AdjClose."""
url = f"{BASE_URL}/{quote(ticker)}?fromDate={from_date}"
r = requests.get(url, timeout=30)
r.raise_for_status()
obj = r.json()
# Gestione schemi JSON possibili
if isinstance(obj, list) and len(obj) == 1 and isinstance(obj[0], dict) and "data" in obj[0]:
obj = obj[0]["data"]
if not isinstance(obj, list):
raise ValueError(f"Schema JSON inatteso per {ticker}: {type(obj)}")
df = pd.DataFrame(obj)
if df.empty:
raise ValueError(f"Nessuna riga per {ticker}")
col_date = _detect_col(df.columns, ["date", "datetime", "timestamp", "time"])
if col_date is None:
raise ValueError(f"Colonna date non trovata per {ticker}. Colonne: {df.columns.tolist()}")
# Priorità: AdjClose, altrimenti Close
col_px = _detect_col(df.columns, ["adj_close", "adjclose", "adjusted_close", "Adj Close", "AdjClose"])
if col_px is None:
col_px = _detect_col(df.columns, ["close", "px_last", "last", "price"])
if col_px is None:
raise ValueError(f"Colonna prezzo non trovata per {ticker}. Colonne: {df.columns.tolist()}")
df[col_date] = pd.to_datetime(df[col_date], errors="coerce", utc=True).dt.tz_localize(None)
df[col_px] = pd.to_numeric(df[col_px], errors="coerce")
df = df.dropna(subset=[col_date, col_px]).sort_values(col_date)
df = df.drop_duplicates(subset=[col_date]).set_index(col_date)
# Normalizza indice a date
df.index = pd.to_datetime(df.index).normalize()
df = df[~df.index.duplicated(keep="last")]
out = pd.DataFrame(index=df.index)
out.index.name = "Date"
out["AdjClose"] = df[col_px].astype(float)
return out.sort_index()
# ============================================================
# Hurst (RS + DFA) su returns
# ============================================================
def hurst_rs_returns(r: np.ndarray, win_grid=None, min_seg=1) -> float:
r = pd.Series(r).dropna().astype("float64").values
n = len(r)
if n < 200:
return np.nan
if win_grid is None:
base = np.array([16, 24, 32, 48, 64, 96, 128, 192, 256, 384], dtype=int)
win_grid = [w for w in base if w <= n // 2]
RS_vals, sizes = [], []
for w in win_grid:
m = n // w
if w < 8 or m < min_seg:
continue
rs_list = []
for i in range(m):
seg = r[i * w : (i + 1) * w]
seg = seg - np.mean(seg)
sd = seg.std(ddof=1)
if sd == 0 or not np.isfinite(sd):
continue
y = np.cumsum(seg)
rs = (np.max(y) - np.min(y)) / sd
if np.isfinite(rs) and rs > 0:
rs_list.append(rs)
if rs_list:
RS_vals.append(np.mean(rs_list))
sizes.append(w)
if len(RS_vals) < 3:
return np.nan
sizes = np.array(sizes, float)
RS_vals = np.array(RS_vals, float)
mask = np.isfinite(RS_vals) & (RS_vals > 0)
sizes, RS_vals = sizes[mask], RS_vals[mask]
if sizes.size < 3:
return np.nan
slope, _ = np.polyfit(np.log(sizes), np.log(RS_vals), 1)
return float(np.clip(slope, 0.0, 1.0)) if np.isfinite(slope) else np.nan
def hurst_dfa_returns(r: np.ndarray, win_grid=None) -> float:
r = pd.Series(r).dropna().astype("float64").values
n = len(r)
if n < 200:
return np.nan
y = np.cumsum(r - np.mean(r))
if win_grid is None:
base = np.array([16, 24, 32, 48, 64, 96, 128, 192, 256], dtype=int)
win_grid = [w for w in base if w <= n // 2]
F_vals, sizes = [], []
for s in win_grid:
m = n // s
if s < 8 or m < 2:
continue
rms_list = []
for i in range(m):
seg = y[i * s : (i + 1) * s]
t = np.arange(s, dtype=float)
A = np.vstack([t, np.ones(s)]).T
coeff, *_ = np.linalg.lstsq(A, seg, rcond=None)
detr = seg - (A @ coeff)
rms = np.sqrt(np.mean(detr**2))
if np.isfinite(rms) and rms > 0:
rms_list.append(rms)
if rms_list:
F_vals.append(np.mean(rms_list))
sizes.append(s)
if len(F_vals) < 3:
return np.nan
sizes = np.array(sizes, float)
F_vals = np.array(F_vals, float)
mask = np.isfinite(F_vals) & (F_vals > 0)
sizes, F_vals = sizes[mask], F_vals[mask]
if sizes.size < 3:
return np.nan
slope, _ = np.polyfit(np.log(sizes), np.log(F_vals), 1)
return float(np.clip(slope, 0.0, 1.0)) if np.isfinite(slope) else np.nan
# ============================================================
# Wavelet denoise (ONLINE, NO look-ahead)
# ============================================================
def wavelet_denoise_last_online(x: np.ndarray) -> float:
"""Denoise usando SOLO la storia fino a t. Ritorna l'ultimo punto denoisato."""
x = np.asarray(x, dtype=float)
n = x.size
if n < DENOISE_MIN_LEN:
return float(x[-1])
s = pd.Series(x).ffill().bfill().values
wav = pywt.Wavelet(DENOISE_WAVELET)
max_level = pywt.dwt_max_level(n, wav.dec_len)
level = int(min(DENOISE_LEVEL, max_level)) if max_level > 0 else 1
coeffs = pywt.wavedec(s, DENOISE_WAVELET, level=level)
detail = coeffs[-1]
if detail.size == 0:
return float(s[-1])
sigma = np.median(np.abs(detail - np.median(detail))) / 0.6745
if not np.isfinite(sigma) or sigma <= 0:
return float(s[-1])
thr = sigma * np.sqrt(2 * np.log(n))
coeffs_th = [coeffs[0]]
for d in coeffs[1:]:
coeffs_th.append(pywt.threshold(d, thr, mode=DENOISE_THRESHOLD_MODE))
rec = pywt.waverec(coeffs_th, DENOISE_WAVELET)
rec = rec[:n]
return float(rec[-1])
# ============================================================
# Per-asset forward EOD backtest
# ============================================================
def forward_backtest_one_asset(close: pd.Series, theta_entry: float, allow_short: bool) -> pd.DataFrame:
"""Backtest EOD:
- Segnale calcolato a close(t) (si usa finestra di returns che include il return di t).
- Esecuzione al close(t).
- PnL per step t = Signal(t) * Ret_fwd(t) - fee * abs(diff(Signal)).
dove Ret_fwd(t) = close(t+1)/close(t)-1.
Output per Date=t con Ret_fwd già allineato.
"""
close = pd.to_numeric(close, errors="coerce").dropna()
close = close[~close.index.duplicated(keep="last")].sort_index()
# returns log per pattern matching
r_log = np.log(close / close.shift(1))
r_log = r_log.dropna()
# forward return (close-to-close): Ret_fwd[t] = close[t+1]/close[t]-1
ret_fwd = close.pct_change().shift(-1)
# Allineamento su date comuni
idx = r_log.index.intersection(ret_fwd.index)
r_log = r_log.loc[idx]
ret_fwd = ret_fwd.loc[idx]
pos = 0
entry_t = None
trade_pnl = 0.0
trade_peak = 0.0
est_hist = []
rows = []
# t è indice relativo su r_log/ret_fwd
for t in range(WP, len(r_log) - 1):
dt = r_log.index[t]
past = r_log.iloc[:t]
if past.dropna().shape[0] < (WP + HA):
rows.append((dt, pos, np.nan, np.nan, np.nan, float(ret_fwd.iloc[t])))
continue
lib_wins, lib_out = build_pattern_library(past, WP, HA)
if lib_wins is None or lib_out is None or len(lib_out) == 0:
rows.append((dt, pos, np.nan, np.nan, np.nan, float(ret_fwd.iloc[t])))
continue
win_last = r_log.iloc[t - WP : t].values
curr_zn = z_norm(win_last)
if curr_zn is None:
rows.append((dt, pos, np.nan, np.nan, np.nan, float(ret_fwd.iloc[t])))
continue
est_raw, avg_dist, _ = predict_from_library(curr_zn, lib_wins, lib_out, k=KNN_K)
est_raw = float(est_raw)
avg_dist = float(avg_dist)
est_hist.append(est_raw)
est_use = wavelet_denoise_last_online(np.array(est_hist, dtype=float)) if DENOISE_ENABLED else est_raw
# ===== ENTRY =====
if pos == 0:
if est_use > theta_entry:
pos = 1
entry_t = t
trade_pnl = 0.0
trade_peak = 0.0
elif allow_short and est_use < -theta_entry:
pos = -1
entry_t = t
trade_pnl = 0.0
trade_peak = 0.0
# ===== EXIT (EOD approximation: usa Ret_fwd della prossima barra t per aggiornare pnl) =====
else:
next_ret = float(ret_fwd.iloc[t])
signed_next = next_ret if pos == 1 else (-next_ret)
pnl_if_stay = (1.0 + trade_pnl) * (1.0 + signed_next) - 1.0
peak_if_stay = max(trade_peak, pnl_if_stay)
exit_now = False
if SL_BPS is not None and pnl_if_stay <= -SL_BPS / 10000.0:
exit_now = True
if TP_BPS is not None and pnl_if_stay >= TP_BPS / 10000.0:
exit_now = True
if TRAIL_BPS is not None and (peak_if_stay - pnl_if_stay) >= TRAIL_BPS / 10000.0:
exit_now = True
if TIME_STOP_BARS is not None and entry_t is not None and (t - entry_t + 1) >= TIME_STOP_BARS:
exit_now = True
# Theta exit simmetrico
if THETA_EXIT is not None:
if pos == 1:
is_weak = est_use <= THETA_EXIT
else:
is_weak = est_use >= -THETA_EXIT
if is_weak:
exit_now = True
if exit_now:
pos = 0
entry_t = None
trade_pnl = 0.0
trade_peak = 0.0
else:
trade_pnl = pnl_if_stay
trade_peak = peak_if_stay
rows.append((dt, pos, est_raw, est_use, avg_dist, float(ret_fwd.iloc[t])))
df = pd.DataFrame(
rows,
columns=["Date", "Signal", "EstOutcomeRaw", "EstOutcome", "AvgDist", "Ret_fwd"],
).set_index("Date")
fee = FEE_BPS / 10000.0
trade_chg = df["Signal"].diff().abs().fillna(0.0)
df["PnL"] = df["Signal"] * df["Ret_fwd"] - trade_chg * fee
return df
# ============================================================
# Portfolio utilities
# ============================================================
def equity_from_returns(r: pd.Series) -> pd.Series:
r = pd.to_numeric(r, errors="coerce").fillna(0.0)
return (1 + r).cumprod() * 100
def monthly_returns(r: pd.Series) -> pd.Series:
r = pd.to_numeric(r, errors="coerce").fillna(0.0)
return (1 + r).resample("M").prod() - 1
def plot_heatmap_monthly(r: pd.Series, title: str):
m = monthly_returns(r)
df = m.to_frame("ret")
df["Year"], df["Month"] = df.index.year, df.index.month
pv = df.pivot(index="Year", columns="Month", values="ret")
fig, ax = plt.subplots(figsize=(10, 6))
im = ax.imshow(pv.fillna(0) * 100, aspect="auto")
for i in range(pv.shape[0]):
for j in range(pv.shape[1]):
val = pv.iloc[i, j]
if not np.isnan(val):
ax.text(j, i, f"{val*100:.1f}", ha="center", va="center", fontsize=8)
ax.set_title(title)
ax.set_xlabel("Mese")
ax.set_ylabel("Anno")
ax.set_xticks(range(12))
ax.set_xticklabels(range(1, 13))
fig.colorbar(im, ax=ax, label="%")
plt.tight_layout()
return fig
def inverse_vol_weights(returns_df: pd.DataFrame, window: int, max_weight: float | None) -> pd.DataFrame:
vol = returns_df.rolling(window).std()
inv = 1 / vol.replace(0, np.nan)
w = inv.div(inv.sum(axis=1), axis=0)
w = w.ffill()
if max_weight is not None:
w = w.clip(upper=max_weight)
return w
def make_active_weights(w_target: pd.DataFrame, sig: pd.DataFrame, add_cash: bool = True, cash_label: str = "Cash") -> pd.DataFrame:
# Attivo se Signal != 0 (sia long che short)
w = (w_target * (sig != 0)).fillna(0.0)
if add_cash:
w[cash_label] = (1.0 - w.sum(axis=1)).clip(lower=0.0)
return w
# ============================================================
# FX currency exposure cap (net exposure per currency)
# ============================================================
def parse_fx_pair(ticker: str) -> tuple[str, str] | None:
# "EURUSD Curncy" -> ("EUR","USD")
if not isinstance(ticker, str):
return None
pair = ticker.split()[0].strip().upper()
if len(pair) < 6:
return None
return pair[:3], pair[3:6]
def currency_exposure_from_weights(weights_row: pd.Series, sig_row: pd.Series) -> dict[str, float]:
"""Net exposure per currency, considerando anche la direzione (long/short).
Long EURUSD (sig=+1): +w EUR, -w USD
Short EURUSD (sig=-1): -w EUR, +w USD
"""
exp: dict[str, float] = {}
for tkr, w in weights_row.items():
if tkr == "Cash":
continue
s = float(sig_row.get(tkr, 0.0))
if s == 0 or w == 0 or not np.isfinite(w):
continue
pq = parse_fx_pair(tkr)
if pq is None:
continue
base, quote = pq
exp[base] = exp.get(base, 0.0) + w * s
exp[quote] = exp.get(quote, 0.0) - w * s
return exp
def apply_currency_cap(weights_act: pd.DataFrame, sig: pd.DataFrame, cap: float) -> tuple[pd.DataFrame, pd.DataFrame]:
"""Applica cap |exposure(currency)| <= cap.
Metodo conservativo ma robusto (NO look-ahead):
- Se max_abs_exposure > cap, scala TUTTE le posizioni risky di fattore cap/max_abs_exposure,
e manda il residuo a Cash.
Ritorna:
- weights_capped (con Cash)
- exposure_wide (diagnostica: esposizioni per currency)
"""
w = weights_act.copy()
if "Cash" not in w.columns:
w["Cash"] = (1.0 - w.sum(axis=1)).clip(lower=0.0)
tick_cols = [c for c in w.columns if c != "Cash"]
# raccogliamo anche esposizioni
all_ccy = set()
exp_dicts: dict[pd.Timestamp, dict[str, float]] = {}
for dt in w.index:
sig_row = sig.loc[dt] if dt in sig.index else pd.Series(dtype=float)
wr = w.loc[dt, tick_cols].fillna(0.0)
exp = currency_exposure_from_weights(wr, sig_row)
exp_dicts[pd.to_datetime(dt)] = exp
all_ccy.update(exp.keys())
if not exp:
w.loc[dt, "Cash"] = (1.0 - float(wr.sum())).clip(lower=0.0) if hasattr(float(wr.sum()), 'clip') else max(0.0, 1.0 - float(wr.sum()))
continue
max_abs = max(abs(v) for v in exp.values())
if np.isfinite(max_abs) and max_abs > cap and max_abs > 1e-12:
scale = cap / max_abs
wr2 = wr * scale
w.loc[dt, tick_cols] = wr2.values
w.loc[dt, "Cash"] = max(0.0, 1.0 - float(wr2.sum()))
else:
w.loc[dt, "Cash"] = max(0.0, 1.0 - float(wr.sum()))
# exposure wide
all_ccy = sorted(all_ccy)
exp_wide = pd.DataFrame(0.0, index=w.index, columns=all_ccy)
for dt, ex in exp_dicts.items():
if dt in exp_wide.index:
for c, v in ex.items():
exp_wide.loc[dt, c] = v
return w, exp_wide
# ============================================================
# Ranking / scoring (semplice, robusto)
# ============================================================
def drawdown_stats_simple(ret: pd.Series) -> dict:
ret = pd.to_numeric(ret, errors="coerce").fillna(0.0)
eq = (1 + ret).cumprod()
peak = eq.cummax()
dd = eq / peak - 1.0
cagr = eq.iloc[-1] ** (DAYS_PER_YEAR / max(1, len(eq))) - 1
annvol = ret.std() * np.sqrt(DAYS_PER_YEAR)
sharpe = (ret.mean() / (ret.std() + 1e-12)) * np.sqrt(DAYS_PER_YEAR)
maxdd = dd.min() if len(dd) else np.nan
calmar = (cagr / abs(maxdd)) if (np.isfinite(cagr) and np.isfinite(maxdd) and maxdd < 0) else np.nan
return {
"CAGR_%": float(np.round(cagr * 100, 3)) if np.isfinite(cagr) else np.nan,
"AnnVol_%": float(np.round(annvol * 100, 3)) if np.isfinite(annvol) else np.nan,
"Sharpe": float(np.round(sharpe, 3)) if np.isfinite(sharpe) else np.nan,
"MaxDD_%": float(np.round(maxdd * 100, 3)) if np.isfinite(maxdd) else np.nan,
"Calmar": float(np.round(calmar, 3)) if np.isfinite(calmar) else np.nan,
}
def compute_score(summary_df: pd.DataFrame) -> pd.DataFrame:
"""Score semplice e stabile (senza ottimizzazioni fancy):
- alto Sharpe e CAGR, basso MaxDD
- normalizzazione via rank percentili
Mantiene colonne di diagnostica (mode) per trasparenza.
"""
df = summary_df.copy()
# ranks 0..1
def rnk(s):
s = pd.to_numeric(s, errors="coerce")
n = s.notna().sum()
if n <= 1:
return pd.Series(np.nan, index=s.index)
return s.rank(method="average") / n
sharpe_r = rnk(df.get("Sharpe"))
cagr_r = rnk(df.get("CAGR_%"))
maxdd_r = rnk(-pd.to_numeric(df.get("MaxDD_%"), errors="coerce")) # meno negativo = meglio
df["Score_mode"] = "rank_sharpe_cagr_maxdd"
df["Score"] = (0.5 * sharpe_r.fillna(0) + 0.3 * cagr_r.fillna(0) + 0.2 * maxdd_r.fillna(0))
return df
# ============================================================
# MAIN
# ============================================================
def main():
print("\n=== Trading Pattern Recon w Hurst - Forex (CORRETTO) ===")
print(f"Fee: {FEE_BPS} bp | Short: {ALLOW_SHORT} | Currency cap: {CURRENCY_CAP:.2f}")
print(f"Wavelet denoise: {DENOISE_ENABLED} ({DENOISE_WAVELET}, level={DENOISE_LEVEL}, min_len={DENOISE_MIN_LEN})")
print("Esecuzione: close(t), PnL: close(t+1)/close(t)\n")
# 1) Fetch prices
prices: dict[str, pd.DataFrame] = {}
for tkr in TICKERS:
try:
print(f"Fetching {tkr} ...")
prices[tkr] = fetch_price_series(tkr, FROM_DATE)
except Exception as e:
print(f"[WARN] Skip {tkr}: {e}")
if len(prices) < 5:
raise RuntimeError(f"Pochi ticker validi ({len(prices)}).")
# 2) Per ticker backtest
hurst_rows = []
summary_rows = []
sig_rows = []
for tkr, dfp in prices.items():
if "AdjClose" not in dfp.columns:
continue
close = dfp["AdjClose"].copy()
close = pd.to_numeric(close, errors="coerce").dropna()
if len(close) < (WP + HA + 80):
print(f"[WARN] Serie troppo corta per {tkr}: len={len(close)}")
continue
r_log = np.log(close / close.shift(1)).dropna()
if len(r_log) < 250:
print(f"[WARN] Serie troppo corta per Hurst per {tkr}")
h_rs = hurst_rs_returns(r_log.values)
h_dfa = hurst_dfa_returns(r_log.values)
H = np.nanmedian([h_rs, h_dfa])
H = float(H) if np.isfinite(H) else np.nan
theta_entry = (H / 100.0) if np.isfinite(H) else THETA_FALLBACK
hurst_rows.append({"Ticker": tkr, "Hurst": H, "theta_entry": float(theta_entry)})
bt = forward_backtest_one_asset(close=close, theta_entry=float(theta_entry), allow_short=ALLOW_SHORT)
bt = bt.copy()
bt.insert(0, "Ticker", tkr)
sig_rows.append(bt.reset_index())
stats = drawdown_stats_simple(bt["PnL"])
hit = 100.0 * ((bt["PnL"] > 0).sum() / max(1, bt["PnL"].notna().sum()))
turnover = bt["Signal"].diff().abs().fillna(0.0).mean() * 100.0 # % steps changed (0..200)
stats.update({
"Ticker": tkr,
"HitRate_%": float(np.round(hit, 3)),
"AvgStepRet_bps": float(np.round(bt["PnL"].mean() * 10000, 3)),
"Turnover_%/day": float(np.round(turnover, 3)),
"N_Steps": int(bt.shape[0]),
"theta_entry": float(theta_entry),
"theta_exit": float(THETA_EXIT) if THETA_EXIT is not None else np.nan,
"sl_bps": float(SL_BPS) if SL_BPS is not None else np.nan,
"tp_bps": float(TP_BPS) if TP_BPS is not None else np.nan,
"trail_bps": float(TRAIL_BPS) if TRAIL_BPS is not None else np.nan,
"time_stop_bars": int(TIME_STOP_BARS) if TIME_STOP_BARS is not None else np.nan,
"allow_short": bool(ALLOW_SHORT),
"fee_bps": float(FEE_BPS),
"wavelet": DENOISE_WAVELET if DENOISE_ENABLED else "none",
"currency_cap": float(CURRENCY_CAP),
})
summary_rows.append(stats)
if not sig_rows:
raise RuntimeError("Nessun ticker backtestato con successo")
hurst_df = pd.DataFrame(hurst_rows).sort_values("Ticker").reset_index(drop=True)
summary_df = pd.DataFrame(summary_rows).sort_values("Ticker").reset_index(drop=True)
signals_df = pd.concat(sig_rows, ignore_index=True)
signals_df["Date"] = pd.to_datetime(signals_df["Date"]).dt.normalize()
# 3) Ranking / Score / TopN
ranking_df = compute_score(summary_df).sort_values("Score", ascending=False).reset_index(drop=True)
top_tickers = ranking_df.head(TOP_N)["Ticker"].astype(str).tolist()
print(f"\nTop{TOP_N} tickers (dal ranking): {top_tickers}\n")
# 4) Costruzione matrici wide
wide_pnl = signals_df.pivot_table(index="Date", columns="Ticker", values="PnL", aggfunc="sum").fillna(0.0).sort_index()
wide_sig = signals_df.pivot_table(index="Date", columns="Ticker", values="Signal", aggfunc="last").fillna(0).astype(int).sort_index()
wide_est = signals_df.pivot_table(index="Date", columns="Ticker", values="EstOutcome", aggfunc="last").sort_index()
# 5) Selezione dinamica TopN giornaliera: manteniamo lo schema classico (attivi + ranking statico)
# - Equal Weight: 1/N sui tickers del ranking (TopN statico)
# - Risk Parity: inverse-vol sui rendimenti dei tickers TopN statici
cols = [c for c in top_tickers if c in wide_pnl.columns]
if not cols:
raise RuntimeError("Nessun ticker del TopN presente in wide_pnl")
dates = wide_pnl.index
# target weights
w_eq_target = pd.DataFrame(0.0, index=dates, columns=wide_pnl.columns)
w_eq_target.loc[:, cols] = 1.0 / len(cols)
rp_hist = wide_pnl[cols]
w_rp_hist = inverse_vol_weights(rp_hist, window=RP_LOOKBACK, max_weight=RP_MAX_WEIGHT).reindex(dates).ffill().fillna(0.0)
w_rp_target = pd.DataFrame(0.0, index=dates, columns=wide_pnl.columns)
w_rp_target.loc[:, cols] = w_rp_hist.values
# rinormalizza sui cols
s = w_rp_target[cols].sum(axis=1).replace(0, np.nan)
w_rp_target.loc[:, cols] = w_rp_target[cols].div(s, axis=0).fillna(0.0)
# active weights (mask su Signal != 0) + cash
w_eq_act = make_active_weights(w_eq_target, wide_sig, add_cash=True)
w_rp_act = make_active_weights(w_rp_target, wide_sig, add_cash=True)
# currency cap
w_eq_cap, ccy_eq = apply_currency_cap(w_eq_act, wide_sig, cap=CURRENCY_CAP)
w_rp_cap, ccy_rp = apply_currency_cap(w_rp_act, wide_sig, cap=CURRENCY_CAP)
# portfolio returns (solo risky assets)
ret_eq = (wide_pnl * w_eq_cap.drop(columns=["Cash"], errors="ignore")).sum(axis=1).rename("Ret_EqW_TopN")
ret_rp = (wide_pnl * w_rp_cap.drop(columns=["Cash"], errors="ignore")).sum(axis=1).rename("Ret_RP_TopN")
eq_eq = equity_from_returns(ret_eq).rename("Eq_EqW_TopN")
eq_rp = equity_from_returns(ret_rp).rename("Eq_RP_TopN")
# 6) Plots
plt.figure(figsize=(10, 5))
plt.plot(eq_eq, label=f"Equal Weight (Top{TOP_N}, fee {FEE_BPS}bp)")
plt.plot(eq_rp, label=f"Risk Parity (Top{TOP_N}, cap {RP_MAX_WEIGHT:.4f}, fee {FEE_BPS}bp)")
plt.title(f"Equity line portafogli FX (base 100) – Top{TOP_N} (shared_utils + URL)")
plt.grid(True)
plt.legend()
plt.tight_layout()
plt.show()
plot_heatmap_monthly(ret_eq, f"Heatmap mensile – Equal Weight FX (Top{TOP_N})")
plt.show()
plot_heatmap_monthly(ret_rp, f"Heatmap mensile – Risk Parity FX (Top{TOP_N})")
plt.show()
# 7) Export
hurst_df.to_csv(OUT_DIR / "hurst.csv", index=False)
summary_df.to_csv(OUT_DIR / "forward_bt_summary.csv", index=False)
signals_df.to_csv(OUT_DIR / "forward_bt_signals.csv", index=False)
ranking_df.to_csv(OUT_DIR / "ranking_score.csv", index=False)
pd.Series(top_tickers, name="TopN_Tickers").to_csv(OUT_DIR / "topn_tickers.csv", index=False)
pd.concat([ret_eq, ret_rp, eq_eq, eq_rp], axis=1).to_csv(OUT_DIR / "portfolio_daily.csv")
w_eq_cap.to_csv(OUT_DIR / "weights_eq_active_capped.csv")
w_rp_cap.to_csv(OUT_DIR / "weights_rp_active_capped.csv")
ccy_eq.to_csv(OUT_DIR / "currency_exposure_eq.csv")
ccy_rp.to_csv(OUT_DIR / "currency_exposure_rp.csv")
print(f"\nSaved to: {OUT_DIR.resolve()}\n")
if __name__ == "__main__":
main()
Reference in New Issue View Git Blame Copy Permalink