"""
Quick check: is the volatility risk premium (short VIX futures, out when the
VIX term structure inverts) a stream with a Sharpe high enough to lever toward
50% a year, and does it survive 2008, 2018-02 and 2020-03?

    PYTHONPATH=. uv run python -m research.lab.short_vol [--no-proxy]

Data (Yahoo):
  ^SHORTVOL  Cboe short VIX futures index from 2005-12: daily -1x short-term
             VIX futures (daily corr with -VIXY 0.987); treated as an excess
             return, so collateral earns the T-bill
  ^VIX, ^VIX3M  spot VIX and 3-month VIX (from 2006-07)
Executable as SVXY (-0.5x, 0.95%/yr) or short VX / VXM futures.

Strategy (Cooper 2013 "Easy Volatility Investing", published 2013-11):
  signal    short vol for day t+1 when VIX/VIX3M at the close of t < 1 (contango)
  size      weight = min(1, 20% / EWMA-21 vol of ^SHORTVOL through t)
  return    rf + w x shortvol - 0.95%/yr x w - 10 bp x |dw|
Reported only: no filter; filter at 0.95; 30% vol target; unlevered -0.5x
(SVXY-like, no sizing).
Periods: 2006-07..2013-12 (before publication, includes 2008) and
2014-01..2026-09 (after).

The rule is written to prereg.json before the run; the script refuses to
overwrite results.json. The project has not tested volatility selling before.

Result (2026-10-08): STOP. C2 and C4 pass; C1 and C3 fail.
  - vrp_ts_vt20, full span: 12.9% / Sharpe 0.61 / MaxDD -36.5%.
    - Before publication (2006-13): Sharpe 1.05.
    - After (2014-26): 7.8% / Sharpe 0.37.
    - Worst day -15.7% (2007-02-27). Bad years: 2018 -24%, 2024 -10%, 2020 -8%.
  - The term-structure filter holds MaxDD to -36% (no filter: -52%) but does not
    lift the post-2014 Sharpe (0.33-0.37 in every variant).
  - SVXY-like -0.5x held throughout: 2014-26 Sharpe 0.52, MaxDD -64%, worst day
    -21% (2024-08-05).
  - Monthly corr: SPY 0.58, TrendRider 0.33. It is equity beta with a short
    crash option, not a diversifier.
Interpretation: the volatility premium shrank after 2013. What is left does
not pay for the tails it carries, at any leverage.
"""

from __future__ import annotations

import argparse
import json
import os

import numpy as np
import pandas as pd

from research.lab.factor_scan import RESULTS
from research.lab.levered_book import stats, tbill

OUT = os.path.join(RESULTS, "short_vol")
PRE, POST = ("2006-08-01", "2013-12-31"), ("2014-01-01", "2026-09-24")
FEE, COST = 0.0095, 0.001

PREREG = {
    "question": "Is term-structure-filtered short VIX futures a high-Sharpe stream that survives its crashes?",
    "primary": "vrp_ts_vt20: short ^SHORTVOL when VIX/VIX3M < 1 at the prior close, sized to 20% vol (cap 1x)",
    "continue_if_all": [
        "C1 2014-2026 (post-publication) excess Sharpe >= 1.0",
        "C2 2006-2013 excess Sharpe >= 0.5",
        "C3 full-span MaxDD >= -35% and worst single day >= -10%",
        "C4 monthly corr with TrendRider V10 <= 0.5",
    ],
    "decision": {"CONTINUE": "preregister a VRP sleeve in the book (instrument SVXY / VXM, leverage, tail limits)",
                 "STOP": "drop short volatility"},
}


def strategy(sv: pd.Series, ratio: pd.Series, rf: pd.Series, thr: float | None, target: float | None,
             fixed: float | None = None) -> tuple[pd.Series, pd.Series]:
    vol = sv.ewm(span=21, min_periods=21).std() * np.sqrt(252)
    if fixed is not None:
        w = pd.Series(fixed, index=sv.index)
    else:
        w = (target / vol).clip(upper=1.0)
    if thr is not None:
        w = w.where(ratio < thr, 0.0)
    w = w.shift(1).fillna(0.0)
    ret = rf / 252 + w * sv.fillna(0.0) - w * FEE / 252 - w.diff().abs().fillna(0.0) * COST
    return ret, w


def main() -> None:
    ap = argparse.ArgumentParser()
    ap.add_argument("--no-proxy", action="store_true")
    if not ap.parse_args().no_proxy:
        from research.lab.fetch_us import _proxy_env
        _proxy_env()
    os.makedirs(OUT, exist_ok=True)
    res_path = os.path.join(OUT, "results.json")
    if os.path.exists(res_path):
        raise SystemExit(f"{res_path} exists; this study does not overwrite results")
    with open(os.path.join(OUT, "prereg.json"), "w") as fh:
        json.dump(PREREG, fh, indent=2)

    from research.lab.fetch_long_history import fetch_yahoo
    c, _ = fetch_yahoo(["^SHORTVOL", "^VIX", "^VIX3M"], start="2005-01-01")
    c = c.loc[:POST[1]].dropna(subset=["^SHORTVOL"])
    sv = c["^SHORTVOL"].pct_change(fill_method=None)
    ratio = c["^VIX"] / c["^VIX3M"]
    rf = tbill(c.index).fillna(0.0)
    variants = {
        "vrp_ts_vt20": (1.0, 0.20, None),
        "vrp_ts95_vt20": (0.95, 0.20, None),
        "vrp_ts_vt30": (1.0, 0.30, None),
        "vrp_nofilter_vt20": (None, 0.20, None),
        "svxy_like_half_nofilter": (None, None, 0.5),
    }
    R, res = {}, {"variants": {}}
    for k, (thr, tgt, fixed) in variants.items():
        r, w = strategy(sv, ratio, rf, thr, tgt, fixed)
        r = r.loc[PRE[0]:POST[1]]
        R[k] = r
        row = {}
        for name, (a, b) in {"full": (PRE[0], POST[1]), "pre": PRE, "post": POST}.items():
            s = stats(r.loc[a:b], rf.loc[a:b])
            row[name] = {x: float(s[x]) for x in ("cagr", "sharpe", "vol", "mdd", "worst_yr")}
        row["worst_day"] = float(r.min())
        row["worst_days"] = {str(d.date()): float(v) for d, v in r.nsmallest(5).items()}
        row["time_in"] = float((w.loc[PRE[0]:] > 0).mean())
        row["avg_w"] = float(w.loc[PRE[0]:].mean())
        res["variants"][k] = row
        f, p, q = row["full"], row["pre"], row["post"]
        print(f"{k:24s} full {f['cagr']*100:6.1f}% xsh {f['sharpe']:5.2f} mdd {f['mdd']*100:6.1f}% | pre xsh "
              f"{p['sharpe']:5.2f} mdd {p['mdd']*100:6.1f}% | post xsh {q['sharpe']:5.2f} cagr {q['cagr']*100:5.1f}% "
              f"mdd {q['mdd']*100:6.1f}% | worst day {row['worst_day']*100:5.1f}% in {row['time_in']:.2f} "
              f"avg w {row['avg_w']:.2f}")
        print("     worst days:", {d: round(v * 100, 1) for d, v in row["worst_days"].items()})
    yr = (1 + R["vrp_ts_vt20"]).groupby(R["vrp_ts_vt20"].index.year).prod() - 1
    print("vrp_ts_vt20 yearly %:", (yr * 100).round(1).to_dict())
    res["yearly_primary"] = {int(k): float(v) for k, v in yr.items()}
    S = pd.read_pickle(os.path.join(RESULTS, "stream_portfolio", "streams.pkl"))
    m = lambda x: (1 + x.dropna()).resample("ME").prod() - 1
    res["corr_trend"] = float(m(R["vrp_ts_vt20"]).corr(m(S["trend"])))
    res["corr_spy"] = float(m(R["vrp_ts_vt20"]).corr(m(S["spy"])))
    print(f"monthly corr: TrendRider {res['corr_trend']:.2f}, SPY {res['corr_spy']:.2f}")

    p = res["variants"]["vrp_ts_vt20"]
    checks = {"C1": p["post"]["sharpe"] >= 1.0, "C2": p["pre"]["sharpe"] >= 0.5,
              "C3": p["full"]["mdd"] >= -0.35 and p["worst_day"] >= -0.10, "C4": res["corr_trend"] <= 0.5}
    res["verdict"] = {"checks": checks, "decision": "CONTINUE" if all(checks.values()) else "STOP"}
    print(f"checks {checks} -> {res['verdict']['decision']}")
    with open(res_path, "w") as fh:
        json.dump(res, fh, indent=2)
    pd.DataFrame(R).to_pickle(os.path.join(OUT, "returns.pkl"))
    from research.lab.trials import append_trials
    append_trials(os.path.join(RESULTS, "trials_us.jsonl"),
                  [{"name": f"short_vol:{k}", **v["full"]} for k, v in res["variants"].items()],
                  {"study": "short_vol", "role": "preregistered quick check"})


if __name__ == "__main__":
    main()