VaR_model / data /validation.py
MinsaitEU's picture
Claude Opus 5.5
Load VaR inputs from the Coral zip; add backtesting and stress tests
cbdae8d
Raw History Blame Contribute Delete
2.43 kB
import numpy as np
import pandas as pd
from data.schema import Col
REQUIRED_PORTFOLIO_COLUMNS = {
Col.POSITION,
Col.FACTOR,
Col.QUANTITY,
Col.MULTIPLIER,
Col.PRICE_UNIT,
Col.PRICE_CURRENCY,
}
def validate_inputs(
prices: pd.DataFrame,
portfolio: pd.DataFrame,
return_methods: dict[str, str],
) -> None:
if prices.index.has_duplicates:
raise ValueError(
"Prices contain duplicate dates."
)
missing = (
REQUIRED_PORTFOLIO_COLUMNS
- set(portfolio.columns)
)
if missing:
raise ValueError(
f"Missing portfolio columns: {sorted(missing)}"
)
if portfolio[Col.POSITION].duplicated().any():
raise ValueError(
"Portfolio contains duplicated positions."
)
if not np.isfinite(
portfolio[Col.QUANTITY].astype(float)
).all():
raise ValueError(
"Portfolio contains non-finite quantities."
)
factors = set(portfolio[Col.FACTOR])
missing_prices = factors - set(prices.columns)
if missing_prices:
raise ValueError(
f"No price series found for: "
f"{sorted(missing_prices)}"
)
missing_methods = factors - set(return_methods)
if missing_methods:
raise ValueError(
f"No return method for: "
f"{sorted(missing_methods)}"
)
def align_prices(
prices: pd.DataFrame,
factors: list[str],
lookback: int | None,
) -> pd.DataFrame:
"""
Keep the factors held in the portfolio, on dates where all of
them are quoted, limited to the last `lookback` returns
(None = use the full history).
"""
aligned = (
prices[factors]
.apply(pd.to_numeric, errors="raise")
.dropna(how="any")
)
required = (lookback or 2) + 1
if len(aligned) < required:
raise ValueError(
f"Not enough price observations: "
f"{len(aligned)} available, need at least {required}."
)
if lookback:
aligned = aligned.tail(lookback + 1)
return aligned
def zero_change_share(
prices: pd.DataFrame,
) -> pd.Series:
"""
Share of days on which each price did not move.
High values point to stale (illiquid or manually
rolled) quotes, which understate volatility.
"""
return (
prices.diff().iloc[1:] == 0
).mean()