#!/usr/bin/env python3
"""
Leverage Decay Study — reproducibility script
AEA Capital Research · Aydin Ali

Recomputes every number in /research/leverage-decay.html from a CSV of
daily adjusted closes.

DATA SOURCE
-----------
Daily adjusted closes were pulled from Massive Market Data (the same market-data
provider cited across the AEA site) for the trailing ~12 months and stored, then
aggregated with SQL. This script reproduces the identical arithmetic from a flat
CSV so the analysis is auditable without an API key.

INPUT
-----
A CSV with columns:  date,ticker,close
(date = ISO or any pandas-parseable date; close = adjusted close)

PAIRS (etf : underlying : stated daily leverage)
------------------------------------------------
NBIL:NBIS:2   TSLL:TSLA:2   NVDL:NVDA:2   AAPU:AAPL:2
MSFU:MSFT:2   TQQQ:QQQ:3    SOXL:SOXX:3

METHOD
------
For each pair, over the pair's common date window:
  und_ret   = und_last / und_first - 1
  target    = leverage * und_ret
  etf_ret   = etf_last / etf_first - 1
  decay_gap = (etf_ret - target)                      # in return space
  und_vol   = std(daily log returns) * sqrt(252)      # annualized

NBIL began trading in Oct 2025, so the NBIL/NBIS pair is measured over NBIL's
shorter common window; every other pair spans the full ~12 months.

USAGE
-----
    python leverage_decay.py closes.csv
"""

import sys
import numpy as np
import pandas as pd

PAIRS = [
    ("NBIL", "NBIS", 2),
    ("TSLL", "TSLA", 2),
    ("NVDL", "NVDA", 2),
    ("AAPU", "AAPL", 2),
    ("MSFU", "MSFT", 2),
    ("TQQQ", "QQQ", 3),
    ("SOXL", "SOXX", 3),
]


def load(path):
    df = pd.read_csv(path, parse_dates=["date"])
    df = df.sort_values("date")
    return df.pivot(index="date", columns="ticker", values="close")


def annualized_vol(series):
    lr = np.log(series / series.shift(1)).dropna()
    return lr.std(ddof=0) * np.sqrt(252) * 100


def analyze(px):
    rows = []
    for etf, und, lev in PAIRS:
        common = px[[etf, und]].dropna()
        if len(common) < 2:
            continue
        e, u = common[etf], common[und]
        und_ret = (u.iloc[-1] / u.iloc[0] - 1) * 100
        etf_ret = (e.iloc[-1] / e.iloc[0] - 1) * 100
        target = lev * und_ret
        rows.append({
            "etf": etf, "lev": f"{lev}x", "underlying": und,
            "und_ret_%": round(und_ret, 2),
            "target_%": round(target, 2),
            "etf_ret_%": round(etf_ret, 2),
            "decay_gap_pp": round(etf_ret - target, 2),
            "und_ann_vol_%": round(annualized_vol(u), 1),
            "days": len(common),
        })
    return pd.DataFrame(rows)


if __name__ == "__main__":
    if len(sys.argv) != 2:
        print(__doc__)
        sys.exit(1)
    result = analyze(load(sys.argv[1]))
    pd.set_option("display.width", 160)
    print(result.to_string(index=False))
    single = result[result["lev"] == "2x"]["decay_gap_pp"].mean()
    print(f"\nAvg decay gap, 2x single-stock funds: {single:.2f} pp")
    print(f"Avg decay gap, all {len(result)} funds:        {result['decay_gap_pp'].mean():.2f} pp")