Skip to content

"Vibe-Trading: Your Personal Trading Agent"

feat(data): weekly and monthly bars, built from daily ones (#1479)

Haozhe Wu6ca364f75744
diff --git a/README.md b/README.md
index 93d22d1d5..3ac5e502f 100644
--- a/README.md
+++ b/README.md
@@ -694,7 +694,7 @@ Run `vibe-trading alpha list` to browse, `vibe-trading alpha show <id>` for form
| **Composite** | cross-market | one shared capital pool across markets (`source="auto"`) |
| **options_portfolio** | options | multi-leg, Greeks, payoff/scenario |
-Intraday bars: 1m / 5m / 15m / 30m / 1H / 4H / 1D. 15 metrics + benchmark comparison, **5 portfolio optimizers** (equal-volatility / risk-parity / mean-variance / max-diversification / turnover-aware), and 3 validation tools (Monte Carlo / Bootstrap / Walk-Forward).
+Bars: 1m / 5m / 15m / 30m / 1H / 4H / 1D, plus weekly / monthly (1W / 1M, built from daily bars). 15 metrics + benchmark comparison, **5 portfolio optimizers** (equal-volatility / risk-parity / mean-variance / max-diversification / turnover-aware), and 3 validation tools (Monte Carlo / Bootstrap / Walk-Forward).
</details>
diff --git a/README_ar.md b/README_ar.md
index c5bdfb7ce..d67d7e01c 100644
--- a/README_ar.md
+++ b/README_ar.md
@@ -653,7 +653,7 @@ LONGBRIDGE_ACCESS_TOKEN=...
| **Composite** | عبر الأسواق | تجمّع رأس مال مشترك واحد عبر الأسواق (`source="auto"`) |
| **options_portfolio** | الخيارات | متعدد الأرجل، وGreeks، وpayoff/scenario |
-أشرطة داخل اليوم: 1m / 5m / 15m / 30m / 1H / 4H / 1D. 15 مقياساً + مقارنة معيارية، و**5 محسّنات محفظة** (equal-volatility / risk-parity / mean-variance / max-diversification / turnover-aware)، و3 أدوات تحقق (Monte Carlo / Bootstrap / Walk-Forward).
+الأشرطة: 1m / 5m / 15m / 30m / 1H / 4H / 1D، إضافةً إلى الأسبوعية / الشهرية (1W / 1M، مبنية من الأشرطة اليومية). 15 مقياساً + مقارنة معيارية، و**5 محسّنات محفظة** (equal-volatility / risk-parity / mean-variance / max-diversification / turnover-aware)، و3 أدوات تحقق (Monte Carlo / Bootstrap / Walk-Forward).
</details>
diff --git a/README_es.md b/README_es.md
index ad38dba1b..d7ee3af54 100644
--- a/README_es.md
+++ b/README_es.md
@@ -658,7 +658,7 @@ Ejecuta `vibe-trading alpha list` para explorar, `vibe-trading alpha show <id>`
| **Composite** | cross-market | un único pool de capital compartido entre mercados (`source="auto"`) |
| **options_portfolio** | opciones | multi-leg, greeks, payoff/escenario |
-Barras intradía: 1m / 5m / 15m / 30m / 1H / 4H / 1D. 15 métricas + comparación con benchmark, **5 optimizadores de cartera** (volatilidad-igual / risk-parity / media-varianza / máxima diversificación / con conciencia de turnover), y 3 herramientas de validación (Monte Carlo / Bootstrap / Walk-Forward).
+Barras: 1m / 5m / 15m / 30m / 1H / 4H / 1D, más semanales / mensuales (1W / 1M, construidas a partir de barras diarias). 15 métricas + comparación con benchmark, **5 optimizadores de cartera** (volatilidad-igual / risk-parity / media-varianza / máxima diversificación / con conciencia de turnover), y 3 herramientas de validación (Monte Carlo / Bootstrap / Walk-Forward).
</details>
diff --git a/README_id.md b/README_id.md
index 808dbb069..c25120439 100644
--- a/README_id.md
+++ b/README_id.md
@@ -819,7 +819,7 @@ Jalankan `vibe-trading alpha list` untuk menjelajah, `vibe-trading alpha show <i
| **Composite** | cross-market | one shared capital pool across markets (`source="auto"`) |
| **options_portfolio** | options | multi-leg, Greeks, payoff/scenario |
-Bar intraday: 1m / 5m / 15m / 30m / 1H / 4H / 1D. 15 metrik + perbandingan benchmark, **5 optimizer portofolio** (equal-volatility / risk-parity / mean-variance / max-diversification / turnover-aware), dan 3 tool validasi (Monte Carlo / Bootstrap / Walk-Forward).
+Bar: 1m / 5m / 15m / 30m / 1H / 4H / 1D, plus mingguan / bulanan (1W / 1M, dibangun dari bar harian). 15 metrik + perbandingan benchmark, **5 optimizer portofolio** (equal-volatility / risk-parity / mean-variance / max-diversification / turnover-aware), dan 3 tool validasi (Monte Carlo / Bootstrap / Walk-Forward).
</details>
diff --git a/README_ja.md b/README_ja.md
index 759fe7269..a04352ea2 100644
--- a/README_ja.md
+++ b/README_ja.md
@@ -653,7 +653,7 @@ Paper-vs-live is a **structural per-broker runtime guard** (account-id format, h
| **Composite** | cross-market | one shared capital pool across markets (`source="auto"`) |
| **options_portfolio** | options | multi-leg, Greeks, payoff/scenario |
-Intraday bars: 1m / 5m / 15m / 30m / 1H / 4H / 1D. 15 metrics + benchmark comparison, **5 portfolio optimizers** (equal-volatility / risk-parity / mean-variance / max-diversification / turnover-aware), and 3 validation tools (Monte Carlo / Bootstrap / Walk-Forward).
+Bars: 1m / 5m / 15m / 30m / 1H / 4H / 1D, plus weekly / monthly (1W / 1M, built from daily bars). 15 metrics + benchmark comparison, **5 portfolio optimizers** (equal-volatility / risk-parity / mean-variance / max-diversification / turnover-aware), and 3 validation tools (Monte Carlo / Bootstrap / Walk-Forward).
</details>
diff --git a/README_ko.md b/README_ko.md
index bdc63aed7..7592c333e 100644
--- a/README_ko.md
+++ b/README_ko.md
@@ -653,7 +653,7 @@ Paper-vs-live는 **구조적 브로커별 런타임 가드**(account-id 형식,
| **Composite** | cross-market | one shared capital pool across markets (`source="auto"`) |
| **options_portfolio** | options | multi-leg, Greeks, payoff/scenario |
-Intraday bars: 1m / 5m / 15m / 30m / 1H / 4H / 1D. 15 metrics + benchmark comparison, **5 portfolio optimizers** (equal-volatility / risk-parity / mean-variance / max-diversification / turnover-aware), and 3 validation tools (Monte Carlo / Bootstrap / Walk-Forward).
+Bars: 1m / 5m / 15m / 30m / 1H / 4H / 1D, plus weekly / monthly (1W / 1M, built from daily bars). 15 metrics + benchmark comparison, **5 portfolio optimizers** (equal-volatility / risk-parity / mean-variance / max-diversification / turnover-aware), and 3 validation tools (Monte Carlo / Bootstrap / Walk-Forward).
</details>
diff --git a/README_zh.md b/README_zh.md
index bb420d1d0..931f22749 100644
--- a/README_zh.md
+++ b/README_zh.md
@@ -666,7 +666,7 @@ LONGBRIDGE_ACCESS_TOKEN=...
| **Composite** | 跨市场 | 跨市场共享单一资金池(`source="auto"`) |
| **options_portfolio** | 期权 | 多腿、Greeks、payoff/scenario |
-日内 bar:1m / 5m / 15m / 30m / 1H / 4H / 1D。15 项指标 + benchmark 对比,**5 个组合优化器**(equal-volatility / risk-parity / mean-variance / max-diversification / turnover-aware),以及 3 个验证工具(Monte Carlo / Bootstrap / Walk-Forward)。
+Bar 周期:1m / 5m / 15m / 30m / 1H / 4H / 1D,以及由日线合成的周线 / 月线(1W / 1M)。15 项指标 + benchmark 对比,**5 个组合优化器**(equal-volatility / risk-parity / mean-variance / max-diversification / turnover-aware),以及 3 个验证工具(Monte Carlo / Bootstrap / Walk-Forward)。
</details>
diff --git a/agent/backtest/benchmark.py b/agent/backtest/benchmark.py
index dd236f88d..8658c5d35 100644
--- a/agent/backtest/benchmark.py
+++ b/agent/backtest/benchmark.py
@@ -11,6 +11,7 @@ from typing import Any, Optional
import pandas as pd
+from backtest.loaders.base import resample_bars, source_interval
from backtest.loaders.yfinance_loader import DataLoader as YfinanceLoader
from backtest.metrics import bar_returns, buy_and_hold_return
@@ -79,7 +80,7 @@ def resolve_benchmark(
source: Data source name (tushare / yfinance / okx / akshare / ccxt).
start_date: Backtest start date.
end_date: Backtest end date.
- interval: Bar interval (1m / 5m / 15m / 30m / 1H / 4H / 1D).
+ interval: Bar interval (1m / 5m / 15m / 30m / 1H / 4H / 1D / 1W / 1M).
explicit: Override ticker (e.g. "SPY" passed via config).
loader: Loader of the configured data source. When given, the
benchmark is fetched through it first, falling back to
@@ -200,24 +201,26 @@ def _fetch_benchmark(
Tries the configured source's loader first (when given). Falls back to
yfinance (single symbol, no auth) when no loader is given or it yields
no data — unless ``allow_fallback`` is False (offline sources fail
- closed instead of making a network request).
+ closed instead of making a network request). Weekly and monthly bars are
+ built from daily ones the way the strategy's are (#1479).
"""
+ fetch_as = source_interval(interval)
if loader is not None:
try:
df = _extract_frame(
- loader.fetch([ticker], start_date, end_date, interval=interval),
+ loader.fetch([ticker], start_date, end_date, interval=fetch_as),
ticker,
)
except Exception:
df = pd.DataFrame()
if not df.empty:
- return df
+ return resample_bars(df, interval)
if not allow_fallback:
return pd.DataFrame()
- result = YfinanceLoader().fetch([ticker], start_date, end_date, interval=interval)
- return _extract_frame(result, ticker)
+ result = YfinanceLoader().fetch([ticker], start_date, end_date, interval=fetch_as)
+ return resample_bars(_extract_frame(result, ticker), interval)
def _extract_frame(result: Any, ticker: str) -> pd.DataFrame:
diff --git a/agent/backtest/engines/_market_hooks.py b/agent/backtest/engines/_market_hooks.py
index 071baefb4..e1b9239de 100644
--- a/agent/backtest/engines/_market_hooks.py
+++ b/agent/backtest/engines/_market_hooks.py
@@ -316,13 +316,21 @@ _TIER_TABLE = [
FUNDING_HOURS = {0, 8, 16}
+#: Spans of the calendar-period bars the runner builds from daily ones
+#: (#1479). A month is the mean one, 365.25 / 12 days, so a monthly bar settles
+#: 91 funding periods whether the month has 28 days or 31.
+_PERIOD_SPAN_HOURS = {"1W": 168.0, "1M": 730.5}
+
+
def _interval_span_hours(interval: str) -> float | None:
"""Bar span in hours for a runner interval token, ``None`` when unknown.
- The runner accepts only 1m/5m/15m/30m/1H/4H/1D, so ``m`` is minutes here
- (there is no monthly token to confuse it with).
+ ``1M`` is a month and ``1m`` a minute; the period tokens are looked up
+ whole before any suffix is read, so the two never meet.
"""
token = str(interval).strip()
+ if token in _PERIOD_SPAN_HOURS:
+ return _PERIOD_SPAN_HOURS[token]
for suffix, scale in (("m", 1 / 60), ("H", 1.0), ("D", 24.0)):
if token.endswith(suffix) and token[: -len(suffix)].isdigit():
return int(token[: -len(suffix)]) * scale
diff --git a/agent/backtest/engines/crypto.py b/agent/backtest/engines/crypto.py
index 49fc75605..8bfcf6b40 100644
--- a/agent/backtest/engines/crypto.py
+++ b/agent/backtest/engines/crypto.py
@@ -87,8 +87,9 @@ class CryptoEngine(BaseEngine):
def _validate_strict_resolution(self, config: dict[str, Any]) -> None:
if not self.perpetual_strict or self.default_leverage < 100:
return
- interval = str(config.get("interval", "1D")).strip().lower()
- if interval not in {"1m", "5m", "15m", "30m", "1h"}:
+ # Compared case-exact: lowercasing would read the monthly "1M" as "1m".
+ interval = str(config.get("interval", "1D")).strip()
+ if interval not in {"1m", "5m", "15m", "30m", "1H", "1h"}:
raise ValueError(
"strict 100x requires a supported interval <= 1H "
"(1m/5m/15m/30m/1H); 1H is only a resolution boundary, "
diff --git a/agent/backtest/loaders/base.py b/agent/backtest/loaders/base.py
index ce3f5abb5..25482d9b8 100644
--- a/agent/backtest/loaders/base.py
+++ b/agent/backtest/loaders/base.py
@@ -208,6 +208,83 @@ DEFAULT_BACKOFF: tuple[float, ...] = (0.5, 1.5, 4.0)
DEFAULT_MAX_RETRIES = 3
+#: Bar sizes built from daily bars instead of asked of a loader (#1479). Every
+#: source serves daily bars; few serve weekly or monthly, and those that do
+#: disagree on where a week ends and how its bar is adjusted, so one resample
+#: here keeps a weekly bar the same thing whichever source served the days. A
+#: bar covers one calendar week (Monday to Sunday) or month.
+RESAMPLED_INTERVALS = {"1W": "W-SUN", "1M": "M"}
+
+#: Columns a period bar sums. ``vwap`` is weighted by volume (NaN without a
+#: volume column), ``funding_rate`` (a per-settlement rate) is averaged,
+#: ``open``/``high``/``low`` take their own rule, and every other column is a
+#: level and takes the period's last value.
+_SUMMED_COLUMNS = frozenset({"volume", "amount"})
+
+
+def source_interval(interval: str) -> str:
+ """Return the bar size to request from a loader when serving ``interval``.
+
+ Args:
+ interval: The bar size the caller asked for.
+
+ Returns:
+ ``"1D"`` for a resampled interval, otherwise ``interval`` unchanged.
+ """
+ return "1D" if interval in RESAMPLED_INTERVALS else interval
+
+
+def resample_bars(frame: pd.DataFrame, interval: str) -> pd.DataFrame:
+ """Aggregate daily bars into ``interval`` bars.
+
+ Each bar is stamped with the last trading day inside its period, which is
+ when its close becomes known, so a signal on it never sees a later day and
+ a partial first or last period stays a bar dated inside that period.
+
+ Args:
+ frame: Daily bars on a ``DatetimeIndex``.
+ interval: A key of :data:`RESAMPLED_INTERVALS`; any other interval
+ returns ``frame`` unchanged.
+
+ Returns:
+ One row per period: first open, highest high, lowest low, last close,
+ summed volume and amount, volume-weighted vwap, mean funding rate, last
+ value of any other column. Frame attributes (quote currency) are kept.
+ """
+ if interval not in RESAMPLED_INTERVALS or frame.empty:
+ return frame
+ index = pd.DatetimeIndex(frame.index)
+ naive = index.tz_localize(None) if index.tz is not None else index
+ periods = naive.to_period(RESAMPLED_INTERVALS[interval])
+ grouped = frame.groupby(periods)
+ columns = {}
+ for column in frame.columns:
+ values = grouped[column]
+ if column == "open":
+ columns[column] = values.first()
+ elif column == "high":
+ columns[column] = values.max()
+ elif column == "low":
+ columns[column] = values.min()
+ elif column in _SUMMED_COLUMNS:
+ columns[column] = values.sum(min_count=1)
+ elif column == "funding_rate":
+ columns[column] = values.mean()
+ elif column == "vwap":
+ if "volume" in frame.columns:
+ traded = (frame["vwap"] * frame["volume"]).groupby(periods).sum(min_count=1)
+ columns[column] = traded / grouped["volume"].sum(min_count=1)
+ else:
+ columns[column] = values.first() * float("nan")
+ else:
+ columns[column] = values.last()
+ out = pd.DataFrame(columns, columns=frame.columns)
+ stamps = pd.Series(index, index=periods).groupby(level=0).max()
+ out.index = pd.DatetimeIndex(stamps.loc[out.index], name=frame.index.name)
+ out.attrs = dict(frame.attrs)
+ return out
+
+
def positive_env_int(name: str, default: int) -> int:
"""Read a positive integer env var, warning and falling back on invalid values."""
raw = os.getenv(name) # noqa: env-gate — generic env var helper
diff --git a/agent/backtest/metrics.py b/agent/backtest/metrics.py
index e84ebf11b..9f50d2ffa 100644
--- a/agent/backtest/metrics.py
+++ b/agent/backtest/metrics.py
@@ -140,6 +140,10 @@ _BARS_PER_DAY = {
},
}
+# Weekly and monthly bars count calendar periods, whatever the market's
+# trading days: every week and every month holds one bar (#1479).
+_CALENDAR_BARS_PER_YEAR = {"1W": 52, "1M": 12}
+
# Runner/loaders also emit these aliases; map them onto the table keys above.
_SOURCE_ALIASES = {"yahoo": "yfinance", "binance": "ccxt"}
@@ -149,10 +153,15 @@ def _normalize_interval(interval: str) -> str:
Minute bars stay lowercase (``1m``); hour/day use the uppercase keys the
table already stores (``1H`` / ``4H`` / ``1D``). Loaders accept both cases
- after the interval-map fixes; annualisation must too.
+ after the interval-map fixes; annualisation must too. ``1M`` is a month and
+ is matched before the case fold, which would read it as a minute (#1479).
"""
token = str(interval or "1D").strip()
+ if token == "1M":
+ return token
lower = token.lower()
+ if lower == "1w":
+ return "1W"
if lower in ("1m", "5m", "15m", "30m"):
return lower
if lower in ("1h", "4h", "1d"):
@@ -164,8 +173,9 @@ def calc_bars_per_year(interval: str = "1D", source: str = "tushare") -> int:
"""Number of bars per year for annualisation.
Args:
- interval: Bar size (1m / 5m / 15m / 30m / 1H / 4H / 1D), case-insensitive
- like loaders accept (``1h`` → ``1H``, ``4h`` → ``4H``, ``1d`` → ``1D``).
+ interval: Bar size (1m / 5m / 15m / 30m / 1H / 4H / 1D / 1W / 1M),
+ case-insensitive like loaders accept (``1h`` → ``1H``, ``4h`` →
+ ``4H``, ``1d`` → ``1D``) except ``1M``, which is a month.
source: Data source (any VALID_SOURCES entry). Defaults to 252 days, 1 bar/day
when source is missing from the table.
@@ -173,6 +183,8 @@ def calc_bars_per_year(interval: str = "1D", source: str = "tushare") -> int:
Bars per year.
"""
interval_key = _normalize_interval(interval)
+ if interval_key in _CALENDAR_BARS_PER_YEAR:
+ return _CALENDAR_BARS_PER_YEAR[interval_key]
source_key = str(source or "").strip().lower()
source_key = _SOURCE_ALIASES.get(source_key, source_key)
trading_days = _TRADING_DAYS.get(source_key, 252)
diff --git a/agent/backtest/runner.py b/agent/backtest/runner.py
index a62d14b3a..b151f0446 100644
--- a/agent/backtest/runner.py
+++ b/agent/backtest/runner.py
@@ -1,7 +1,8 @@
"""Fixed backtest entrypoint: read config.json, select loader by source, import signal_engine, run engine.
Supports ``source="auto"`` to route codes to loaders by symbol format.
-Supports ``interval`` for bar size (1m/5m/15m/30m/1H/4H/1D, default 1D).
+Supports ``interval`` for bar size (1m/5m/15m/30m/1H/4H/1D/1W/1M, default 1D;
+1W and 1M are built from daily bars, see ``loaders.base.resample_bars``).
Supports ``engine`` for backtest engine (daily/options, default daily).
Usage: ``python -m backtest.runner <run_dir>``
@@ -32,7 +33,12 @@ from backtest.loaders.registry import (
price_caliber,
resolve_loader,
)
-from backtest.loaders.base import NoAvailableSourceError, validate_ohlc
+from backtest.loaders.base import (
+ NoAvailableSourceError,
+ resample_bars,
+ source_interval,
+ validate_ohlc,
+)
# Symbol classification lives in ``_market_hooks`` so runner.py and
# composite.py share a single source of truth (audit-2026-05-18 B1+C1+C2).
# ``_detect_market`` is also re-exported here for back-compat with
@@ -48,7 +54,7 @@ from backtest.rebalance_mask import RebalanceMask, validate_rebalance_mask
logger = logging.getLogger(__name__)
-_VALID_INTERVALS = {"1m", "5m", "15m", "30m", "1H", "4H", "1D"}
+_VALID_INTERVALS = {"1m", "5m", "15m", "30m", "1H", "4H", "1D", "1W", "1M"}
_VALID_ENGINES = {"daily", "options"}
_PRICE_PANEL_COLUMNS = ("open", "high", "low", "close", "volume", "vwap", "amount")
_FUND_PREFIX = "fund:"
@@ -1320,6 +1326,10 @@ _INTERVAL_SECONDS: dict[str, float] = {
"1H": 3_600.0,
"4H": 14_400.0,
"1D": 86_400.0,
+ "1W": 604_800.0,
+ # A mean month (365.25 / 12 days): a calendar month runs 28 to 31, so a
+ # monthly series' median spacing sits within 1.1x of it either way.
+ "1M": 2_629_800.0,
}
#: How far the served bar spacing may sit from the declared interval's spacing
@@ -1334,10 +1344,11 @@ _SPACING_MISMATCH_RATIO = 1.5
#: (``[1, 1, 3]`` -> 1 day); two do not (``[1, 3]`` -> 2 days).
_MIN_BARS_FOR_SPACING = 4
-#: Spacing at least this many times a day is a weekly or monthly series, not a
-#: daily one with gaps: a daily index over a holiday week measures a median of
-#: two or three days, a weekly file seven. Such a series has no trading-day
-#: table to look up and needs none -- its bars per year is the calendar's.
+#: Spacing at least this many times a day that matches no supported interval
+#: (a fortnightly or quarterly file) is a coarse series, not a daily one with
+#: gaps: a daily index over a holiday week measures a median of two or three
+#: days. Such a series has no trading-day table to look up and needs none --
+#: its bars per year is the calendar's.
_WIDER_THAN_DAILY_RATIO = 4.0
_CALENDAR_YEAR_SECONDS = 365.25 * 86_400.0
@@ -1394,8 +1405,9 @@ def _annualisation_bars(
:func:`~backtest.metrics.calc_bars_per_year` -- looked up with the interval
the spacing actually matches -- so the per-source trading-day table keeps
producing the number and a run card never picks up a window-dependent one.
- Bars spaced wider than any supported interval (a weekly or monthly file)
- are annualised from the calendar instead, 52 for a weekly series; a
+ A weekly or monthly file declared ``1D`` matches ``1W`` / ``1M`` (52 / 12).
+ Bars spaced wider than daily that match no supported interval (a quarterly
+ file) are annualised from the calendar instead, 4 for a quarterly series; a
spacing that matches nothing in either direction keeps the declaration.
Args:
@@ -1723,7 +1735,10 @@ def fetch_data_map(config: dict) -> DataFetchResult:
config = copy.deepcopy(config)
source = str(config.get("source") or "tushare")
codes = list(config.get("codes") or [])
- interval = str(config.get("interval") or "1D")
+ # Weekly and monthly bars are built from daily ones after the fetch, so
+ # every loader below is asked for daily bars (#1479).
+ requested_interval = str(config.get("interval") or "1D")
+ interval = source_interval(requested_interval)
# ``local:`` picks the loader; the instrument is the bare symbol. Everything
# downstream (engine, signals, artifacts, run card) sees the bare symbol, so
@@ -1854,7 +1869,10 @@ def fetch_data_map(config: dict) -> DataFetchResult:
f"incomplete data for source={primary_source}; missing symbols: {missing}"
)
- data_map = _sanitize_data_map(data_map)
+ data_map = {
+ code: resample_bars(frame, requested_interval)
+ for code, frame in _sanitize_data_map(data_map).items()
+ }
caliber_stamps = {
code: stamp for code, stamp in caliber_stamps.items() if code in data_map
}
diff --git a/agent/mcp_server.py b/agent/mcp_server.py
index 3a3e5fed0..92b31f7d8 100644
--- a/agent/mcp_server.py
+++ b/agent/mcp_server.py
@@ -1809,7 +1809,7 @@ def get_market_data(
end_date: End date (YYYY-MM-DD).
source: Data source. Prefer ``auto``; ``yahoo``/``yfinance`` serve
Canada, US, and HK equities.
- interval: Bar size (1m/5m/15m/30m/1H/4H/1D, default "1D").
+ interval: Bar size (1m/5m/15m/30m/1H/4H/1D/1W/1M, default "1D"; 1W and 1M are built from daily bars).
max_rows: Per-symbol row cap (default 250) so the response stays
within the MCP token budget. A symbol exceeding it returns an
even-stride downsample (every step-th bar, last bar pinned)
diff --git a/agent/src/api/attribution_core.py b/agent/src/api/attribution_core.py
index f4b3995a1..fe2093dc7 100644
--- a/agent/src/api/attribution_core.py
+++ b/agent/src/api/attribution_core.py
@@ -51,6 +51,7 @@ _ATTRIBUTION_BARS_PER_YEAR_BY_INTERVAL = {
"1d": 252,
"d": 252,
"daily": 252,
+ "1w": 52,
"1wk": 52,
"w": 52,
"weekly": 52,
@@ -95,9 +96,11 @@ def _attribution_bars_per_year(interval: Any) -> int:
"""Infer the annualization factor from the run's bar interval label."""
if not interval:
return _ATTRIBUTION_DEFAULT_BARS_PER_YEAR
- return _ATTRIBUTION_BARS_PER_YEAR_BY_INTERVAL.get(
- str(interval).strip().lower(), _ATTRIBUTION_DEFAULT_BARS_PER_YEAR
- )
+ label = str(interval).strip()
+ # The runner's monthly token; lower-cased it would read as one minute.
+ if label == "1M":
+ return 12
+ return _ATTRIBUTION_BARS_PER_YEAR_BY_INTERVAL.get(label.lower(), _ATTRIBUTION_DEFAULT_BARS_PER_YEAR)
def _attribution_downsample(
diff --git a/agent/src/market_data.py b/agent/src/market_data.py
index 3aa1cd9b6..ec05d88ef 100644
--- a/agent/src/market_data.py
+++ b/agent/src/market_data.py
@@ -203,7 +203,7 @@ def fetch_market_data(
consumers must read this field instead of assuming a unit.
"""
from backtest.engines._market_hooks import _detect_market
- from backtest.loaders.base import NoAvailableSourceError
+ from backtest.loaders.base import NoAvailableSourceError, resample_bars, source_interval
from backtest.loaders.registry import (
FALLBACK_CHAINS,
_NO_NETWORK_FALLBACK_SOURCES,
@@ -315,7 +315,14 @@ def fetch_market_data(
continue
try:
loader = loader_cls()
- partial = loader.fetch(remaining, start_date, end_date, interval=interval)
+ # Weekly and monthly bars are built from daily ones (#1479).
+ partial = loader.fetch(
+ remaining, start_date, end_date, interval=source_interval(interval)
+ )
+ partial = {
+ code: resample_bars(frame, interval) if hasattr(frame, "groupby") else frame
+ for code, frame in (partial or {}).items()
+ }
except Exception as exc: # noqa: BLE001 — contained per-symbol fallback
logger.error(
"market-data loader %r failed for %s; trying next source in chain: %s",
diff --git a/agent/src/skills/strategy-generate/SKILL.md b/agent/src/skills/strategy-generate/SKILL.md
index e4829cc03..fd1bb8132 100644
--- a/agent/src/skills/strategy-generate/SKILL.md
+++ b/agent/src/skills/strategy-generate/SKILL.md
@@ -138,7 +138,7 @@ Self-check after writing `signal_engine.py`:
- `source`: `"auto"` (recommended, auto-select by code format) / `"tushare"` / `"yfinance"` / `"okx"` / `"akshare"` / `"ccxt"`
- `"auto"` supports mixed instruments. For example, `["000001.SZ", "BTC-USDT"]` will be automatically routed to `tushare` and `okx`
- Futures codes (e.g. `"IF2406.CFFEX"`, `"ESZ4"`) and forex pairs (e.g. `"EUR/USD"`) are also auto-routed
-- `interval`: candlestick interval, default `"1D"`. Supported values: `"1m"` / `"5m"` / `"15m"` / `"30m"` / `"1H"` / `"4H"` / `"1D"`
+- `interval`: candlestick interval, default `"1D"`. Supported values: `"1m"` / `"5m"` / `"15m"` / `"30m"` / `"1H"` / `"4H"` / `"1D"` / `"1W"` / `"1M"` (weekly and monthly bars are built from daily ones; `"1M"` is a month, `"1m"` a minute)
- The annualization factor for minute backtests is inferred automatically from `source` (252 trading days for China A-shares, 365 calendar days for crypto)
- Minute backtests can be very data-heavy. Recommended limits are no more than 30 days for `1m`, or 1 year for `1H`
- `warmup_bars`: how many leading bars exist only to prime the indicators. They are loaded and fed to `SignalEngine.generate()`, then excluded from trades, the equity curve, the benchmark and every metric. Default `0` grades the whole loaded window.
diff --git a/agent/src/tools/market_data_tool.py b/agent/src/tools/market_data_tool.py
index d3504e9cf..3a07edd18 100644
--- a/agent/src/tools/market_data_tool.py
+++ b/agent/src/tools/market_data_tool.py
@@ -14,34 +14,26 @@ from backtest.loaders.registry import VALID_SOURCES
from backtest.runner import _VALID_INTERVALS
# Canonical-case lookup. ``_VALID_INTERVALS`` mixes cases ("1m" minutes vs
-# "1H" hours), so a plain ``.upper()`` fold is unsafe: it maps the universal
-# monthly spelling "1M" onto "1m" (one minute), silently answering a monthly
-# question with minute bars (#1480). ``_canonicalize_interval`` resolves a
-# case-exact match first and only then falls back to this fold for the
-# unambiguous case variants ("1d" -> "1D", "30M" -> "30m").
-_INTERVAL_CANON = {v.upper(): v for v in _VALID_INTERVALS}
-
-# Spellings that read as a request for a bar size this tool does not serve and
-# must never be case-folded onto a minute interval. "1M" is the near-universal
-# monthly convention (pandas ``resample('1M')``, ccxt / TradingView "1M"); "1W"
-# is weekly. Folding "1M" -> "1m" (one minute) is the #1480 footgun: the caller
-# receives minute bars labelled as the answer to a monthly question. These are
-# rejected with a pointed message instead of being silently mis-served. The
-# minute spellings "5M" / "15M" / "30M" deliberately keep folding to "5m" /
-# "15m" / "30m" — multi-month bars are not a convention anyone requests, and
-# that fold is the documented behaviour pinned by the regression tests.
-_UNSUPPORTED_PERIOD_SPELLINGS = {"1M": "one month", "1W": "one week"}
+# "1M" months, "1H" hours), so a plain ``.upper()`` fold is unsafe: it maps
+# "1m" and "1M" onto one key, and before monthly bars existed it answered a
+# monthly question with minute bars (#1480). A case-exact match wins first in
+# ``_canonicalize_interval``; this fold only serves the unambiguous variants
+# ("1d" -> "1D", "30M" -> "30m", "1w" -> "1W"), and a minute token whose upper
+# case is itself an interval ("1m" -> "1M") is left out of it.
+_INTERVAL_CANON = {
+ v.upper(): v for v in _VALID_INTERVALS if v.upper() == v or v.upper() not in _VALID_INTERVALS
+}
+# The yfinance spellings the technical-indicator tool has always documented.
+_INTERVAL_CANON.update({"1WK": "1W", "1MO": "1M"})
def _canonicalize_interval(raw: str) -> str | None:
"""Resolve a user interval spelling to the canonical loader form.
A case-exact match against ``_VALID_INTERVALS`` wins first, so the minute
- interval ``"1m"`` and the monthly spelling ``"1M"`` are never conflated by
- the case fold (#1480). Falls back to ``_INTERVAL_CANON`` for unambiguous
- case variants (``"1d"`` -> ``"1D"``, ``"30M"`` -> ``"30m"``). Month/week
- spellings that are not served resolve to ``None`` rather than folding onto
- a minute interval.
+ interval ``"1m"`` and the monthly ``"1M"`` are never conflated by the case
+ fold (#1480). Falls back to ``_INTERVAL_CANON`` for unambiguous case
+ variants (``"1d"`` -> ``"1D"``, ``"30M"`` -> ``"30m"``, ``"1wk"`` -> ``"1W"``).
Args:
raw: Interval string as supplied by the caller.
@@ -52,10 +44,7 @@ def _canonicalize_interval(raw: str) -> str | None:
token = raw.strip()
if token in _VALID_INTERVALS:
return token
- folded = token.upper()
- if folded in _UNSUPPORTED_PERIOD_SPELLINGS:
- return None
- return _INTERVAL_CANON.get(folded)
+ return _INTERVAL_CANON.get(token.upper())
# Source allow-list derived from the shared loader registry (the same set the
@@ -148,7 +137,10 @@ class MarketDataTool(BaseTool):
},
"interval": {
"type": "string",
- "description": "Bar size, e.g. 1D, 1H, 4H, 30m.",
+ "description": (
+ "Bar size: 1m/5m/15m/30m/1H/4H/1D, or 1W/1M for weekly/monthly "
+ "bars built from daily ones (1M is a month, 1m a minute)."
+ ),
"default": "1D",
},
"max_rows": {
@@ -204,26 +196,11 @@ class MarketDataTool(BaseTool):
if not isinstance(interval, str):
return _error("interval must be a string like '1D', '1H', '4H', '30m'")
interval_token = interval.strip()
- # Reject the month/week spellings the case fold would otherwise map onto
- # a minute interval: "1M" (one month) must never silently become "1m"
- # (one minute) (#1480). A case-exact served spelling is allowed through
- # first, so this stays correct if monthly/weekly bars are ever added to
- # ``_VALID_INTERVALS`` (#1479).
- folded = interval_token.upper()
- if (
- interval_token not in _VALID_INTERVALS
- and folded in _UNSUPPORTED_PERIOD_SPELLINGS
- ):
- return _error(
- f"interval {interval_token!r} ({_UNSUPPORTED_PERIOD_SPELLINGS[folded]}) "
- f"is not supported; supported: {sorted(_VALID_INTERVALS)}. "
- f"Note '1m' (lowercase) is one minute."
- )
normalized_interval = _canonicalize_interval(interval_token)
if normalized_interval is None:
return _error(
- f"interval must be one of {sorted(_VALID_INTERVALS)} "
- f"(case-insensitive); got {interval!r}"
+ f"interval must be one of {sorted(_VALID_INTERVALS)} (case-insensitive, "
+ f"except that 1m is one minute and 1M one month); got {interval!r}"
)
max_rows = kwargs.get("max_rows", DEFAULT_MAX_ROWS)
diff --git a/agent/src/tools/technical_indicator_tool.py b/agent/src/tools/technical_indicator_tool.py
index baf97f281..e63d85211 100644
--- a/agent/src/tools/technical_indicator_tool.py
+++ b/agent/src/tools/technical_indicator_tool.py
@@ -16,6 +16,7 @@ import pandas as pd
from src.agent.tools import BaseTool
from src.market_data import fetch_market_data
+from src.tools.market_data_tool import _canonicalize_interval
logger = logging.getLogger(__name__)
@@ -30,6 +31,10 @@ _SMA_PERIODS = (20, 50, 200)
_EMA_PERIOD = 20
_DEFAULT_LOOKBACK = 200
_MAX_LOOKBACK = 500
+#: Calendar days fetched per requested bar. Two covers weekends and holidays
+#: for daily bars; a weekly or monthly bar needs its whole period, or 200
+#: weekly bars would be asked for over 400 days and come back as 57 (#1479).
+_CALENDAR_DAYS_PER_BAR = {"1W": 7, "1M": 31}
_CLOSE_KEYS = ("close", "Close", "CLOSE", "adj_close")
@@ -206,11 +211,18 @@ class TechnicalIndicatorTool(BaseTool):
def execute(self, **kwargs: Any) -> str:
symbol = str(kwargs.get("symbol", "")).strip()
- interval = str(kwargs.get("interval", "1d")).strip()
+ requested_interval = str(kwargs.get("interval", "1d")).strip()
lookback_raw = kwargs.get("lookback", _DEFAULT_LOOKBACK)
if not symbol:
return json.dumps({"ok": False, "error": "symbol is required"})
+ # The documented 1d / 1wk / 1mo spellings used to reach the loaders
+ # verbatim, and most of them know neither weekly spelling.
+ interval = _canonicalize_interval(requested_interval)
+ if interval is None:
+ return json.dumps(
+ {"ok": False, "error": f"unsupported interval {requested_interval!r}; use 1d, 1wk or 1mo"}
+ )
try:
lookback = int(lookback_raw)
@@ -220,7 +232,8 @@ class TechnicalIndicatorTool(BaseTool):
# Fetch enough bars to cover the longest indicator window + buffer.
end_date = datetime.now().strftime("%Y-%m-%d")
- start_date = (datetime.now() - timedelta(days=lookback * 2)).strftime("%Y-%m-%d")
+ days = lookback * _CALENDAR_DAYS_PER_BAR.get(interval, 2)
+ start_date = (datetime.now() - timedelta(days=days)).strftime("%Y-%m-%d")
try:
data = fetch_market_data(
diff --git a/agent/tests/test_cross_market_annualization.py b/agent/tests/test_cross_market_annualization.py
index 27c9557ae..f0b8203c1 100644
--- a/agent/tests/test_cross_market_annualization.py
+++ b/agent/tests/test_cross_market_annualization.py
@@ -251,14 +251,14 @@ class TestSingleMarketAnnualisationChecksTheServedData:
)
assert _annualisation_bars("1D", "tushare", data, ["600519.SH"]) == 252
- # --- spacing wider than any supported interval (review point 2) ---
+ # --- weekly / monthly files, and spacing wider than any supported interval ---
@pytest.mark.parametrize("declared", ["1D", "1H"])
- def test_weekly_bars_are_annualised_from_the_calendar(self, declared, caplog):
- """A weekly file has no trading-day table and needs none: 52 a year.
+ def test_a_weekly_file_is_annualised_as_weekly(self, declared, caplog):
+ """A weekly file declared finer is read as ``1W``: 52 a year.
- Keeping the declaration was the same bug one step coarser -- a weekly
- file declared ``1H`` annualised at 1,764 bars a year.
+ Keeping the declaration was the bug -- a weekly file declared ``1H``
+ annualised at 1,764 bars a year.
"""
from backtest.runner import _annualisation_bars
@@ -267,7 +267,7 @@ class TestSingleMarketAnnualisationChecksTheServedData:
resolved = _annualisation_bars(declared, "tushare", data, ["600519.SH"])
assert resolved == 52
- assert any("wider than any supported interval" in r.getMessage() for r in caplog.records)
+ assert any("annualising as 1W (52 bars/year)" in r.getMessage() for r in caplog.records)
def test_monthly_bars_are_twelve_a_year(self):
from backtest.runner import _annualisation_bars
@@ -275,6 +275,30 @@ class TestSingleMarketAnnualisationChecksTheServedData:
data = self._frame(pd.date_range("2020-01-01", periods=48, freq="MS"))
assert _annualisation_bars("1D", "yahoo", data, ["600519.SH"]) == 12
+ @pytest.mark.parametrize(("declared", "freq"), [("1W", "W-FRI"), ("1M", "BME")])
+ def test_a_declared_weekly_or_monthly_run_keeps_its_count_silently(self, declared, freq, caplog):
+ """Weekly and monthly are intervals now (#1479): no mismatch report."""
+ from backtest.runner import _annualisation_bars
+
+ data = self._frame(pd.date_range("2020-01-01", periods=60, freq=freq))
+ with caplog.at_level("WARNING", logger="backtest.runner"):
+ resolved = _annualisation_bars(declared, "tushare", data, ["600519.SH"])
+
+ assert resolved == {"1W": 52, "1M": 12}[declared]
+ assert caplog.records == []
+
+ def test_a_quarterly_file_is_annualised_from_the_calendar(self, caplog):
+ """Wider than every supported interval: four bars a year, from the
+ spacing, since no interval has a count to look up."""
+ from backtest.runner import _annualisation_bars
+
+ data = self._frame(pd.date_range("2015-03-31", periods=40, freq="QE"))
+ with caplog.at_level("WARNING", logger="backtest.runner"):
+ resolved = _annualisation_bars("1D", "tushare", data, ["600519.SH"])
+
+ assert resolved == 4
+ assert any("wider than any supported interval" in r.getMessage() for r in caplog.records)
+
def test_a_daily_series_over_a_holiday_week_is_not_read_as_weekly(self, caplog):
"""Five daily bars around Christmas measure a two-day median: the
declaration stands, the report states the spacings, and the count is
diff --git a/agent/tests/test_crypto_engine.py b/agent/tests/test_crypto_engine.py
index 888d29ec7..0fd8adeae 100644
--- a/agent/tests/test_crypto_engine.py
+++ b/agent/tests/test_crypto_engine.py
@@ -557,7 +557,8 @@ class TestHistoricalFundingRate:
class TestStrictPerpetualLifecycle:
- @pytest.mark.parametrize("interval", ["3m", "60m", "4H", "1D"])
+ # "1M" is a month: the old lower-cased comparison read it as "1m" (#1479).
+ @pytest.mark.parametrize("interval", ["3m", "60m", "4H", "1D", "1W", "1M"])
def test_strict_100x_rejects_unsupported_or_coarse_intervals(
self, interval: str
) -> None:
diff --git a/agent/tests/test_market_data_tool.py b/agent/tests/test_market_data_tool.py
index 0ffdb83b8..3b3de01da 100644
--- a/agent/tests/test_market_data_tool.py
+++ b/agent/tests/test_market_data_tool.py
@@ -465,36 +465,35 @@ def test_market_data_tool_accepts_minute_intervals():
assert [c["interval"] for c in calls] == ["1m", "5m", "15m", "30m", "30m", "1H", "1D"]
-def test_market_data_tool_rejects_month_and_week_spellings():
- """'1M' (one month) must never case-fold to '1m' (one minute) (#1480)."""
+def test_market_data_tool_serves_month_and_week_spellings():
+ """'1M' is one month and '1m' one minute; neither folds onto the other
+ (#1480), and weekly/monthly bars reach the fetch as 1W / 1M (#1479)."""
import src.tools.market_data_tool as mod
from unittest import mock
with mock.patch.object(mod, "fetch_market_data_json", return_value="{}") as fetch:
- for interval, meaning in (("1M", "one month"), ("1W", "one week")):
- out = json.loads(
- mod.MarketDataTool().execute(
- codes=["AAPL.US"],
- start_date="2026-08-20",
- end_date="2026-08-21",
- interval=interval,
- )
+ for interval, canonical in (
+ ("1M", "1M"), ("1mo", "1M"), ("1W", "1W"), ("1w", "1W"), ("1wk", "1W"), ("1m", "1m")
+ ):
+ fetch.reset_mock()
+ mod.MarketDataTool().execute(
+ codes=["AAPL.US"],
+ start_date="2026-08-20",
+ end_date="2026-08-21",
+ interval=interval,
)
- assert out["ok"] is False
- assert meaning in out["error"]
- assert "one minute" in out["error"]
- assert interval in out["error"]
- # Silence is the failure mode (#1480): the fold must never reach a loader.
- fetch.assert_not_called()
+ assert fetch.call_args.kwargs["interval"] == canonical, interval
-def test_canonicalize_interval_keeps_minute_and_refuses_month():
- """_canonicalize_interval resolves '1m' but refuses '1M'/'1W' (#1480)."""
- from src.tools.market_data_tool import _canonicalize_interval
+def test_canonicalize_interval_keeps_minute_and_month_apart():
+ """_canonicalize_interval never maps '1m' and '1M' onto each other (#1480)."""
+ from src.tools.market_data_tool import _INTERVAL_CANON, _canonicalize_interval
assert _canonicalize_interval("1m") == "1m"
- assert _canonicalize_interval("1M") is None
- assert _canonicalize_interval("1W") is None
+ assert _canonicalize_interval("1M") == "1M"
+ assert _canonicalize_interval("1W") == "1W"
assert _canonicalize_interval("1d") == "1D"
assert _canonicalize_interval("30M") == "30m"
assert _canonicalize_interval("1h") == "1H"
+ assert _canonicalize_interval("2W") is None
+ assert _INTERVAL_CANON["1M"] == "1M"
diff --git a/agent/tests/test_weekly_monthly_bars.py b/agent/tests/test_weekly_monthly_bars.py
new file mode 100644
index 000000000..b9836e7d7
--- /dev/null
+++ b/agent/tests/test_weekly_monthly_bars.py
@@ -0,0 +1,211 @@
+"""Weekly and monthly bars, built from daily ones at the fetch boundary (#1479)."""
+
+from __future__ import annotations
+
+import json
+from datetime import datetime, timedelta
+
+import numpy as np
+import pandas as pd
+import pytest
+
+from backtest import runner
+from backtest.benchmark import _fetch_benchmark
+from backtest.engines._market_hooks import _interval_span_hours
+from backtest.loaders.base import resample_bars, source_interval
+from backtest.metrics import calc_bars_per_year
+from src.api.attribution_core import _attribution_bars_per_year
+from src.market_data import fetch_market_data
+
+
+def _daily(start: str, end: str, *, freq: str = "B") -> pd.DataFrame:
+ index = pd.DatetimeIndex(pd.date_range(start, end, freq=freq), name="trade_date")
+ step = np.arange(len(index), dtype=float)
+ return pd.DataFrame(
+ {
+ "open": 100.0 + step,
+ "high": 101.0 + step,
+ "low": 99.0 + step,
+ "close": 100.5 + step,
+ "volume": 10.0 + step,
+ },
+ index=index,
+ )
+
+
+class TestResampleBars:
+ def test_a_week_runs_monday_to_sunday_and_is_stamped_with_its_last_trading_day(self):
+ frame = _daily("2024-01-03", "2024-01-19") # Wed 3rd .. Fri 19th
+ weekly = resample_bars(frame, "1W")
+
+ assert list(weekly.index.strftime("%Y-%m-%d")) == ["2024-01-05", "2024-01-12", "2024-01-19"]
+ first = frame.loc["2024-01-03":"2024-01-05"]
+ assert weekly.iloc[0]["open"] == first["open"].iloc[0]
+ assert weekly.iloc[0]["high"] == first["high"].max()
+ assert weekly.iloc[0]["low"] == first["low"].min()
+ assert weekly.iloc[0]["close"] == first["close"].iloc[-1]
+ assert weekly.iloc[0]["volume"] == first["volume"].sum()
+
+ def test_a_seven_day_market_closes_its_week_on_sunday(self):
+ weekly = resample_bars(_daily("2024-01-01", "2024-01-14", freq="D"), "1W")
+ assert list(weekly.index.strftime("%a %Y-%m-%d")) == ["Sun 2024-01-07", "Sun 2024-01-14"]
+
+ def test_a_month_is_a_calendar_month_and_a_partial_last_one_stays_inside_it(self):
+ monthly = resample_bars(_daily("2024-01-15", "2024-03-06"), "1M")
+
+ assert list(monthly.index.strftime("%Y-%m-%d")) == ["2024-01-31", "2024-02-29", "2024-03-06"]
+ assert monthly.loc["2024-02-29", "open"] == _daily("2024-01-15", "2024-03-06").loc["2024-02-01", "open"]
+
+ def test_vwap_is_weighted_by_volume_and_funding_is_averaged(self):
+ frame = _daily("2024-01-01", "2024-01-02")
+ frame["vwap"] = [10.0, 20.0]
+ frame["volume"] = [1.0, 3.0]
+ frame["funding_rate"] = [0.0001, 0.0003]
+ frame["pe"] = [5.0, 6.0]
+
+ bar = resample_bars(frame, "1W").iloc[0]
+
+ assert bar["vwap"] == pytest.approx(17.5)
+ assert bar["funding_rate"] == pytest.approx(0.0002)
+ assert bar["pe"] == 6.0
+
+ def test_a_vwap_without_volume_is_unknown_not_the_last_day(self):
+ frame = _daily("2024-01-01", "2024-01-02").drop(columns="volume")
+ frame["vwap"] = [10.0, 20.0]
+ assert np.isnan(resample_bars(frame, "1W").iloc[0]["vwap"])
+
+ def test_missing_volume_stays_missing(self):
+ frame = _daily("2024-01-01", "2024-01-05")
+ frame["volume"] = np.nan
+ assert np.isnan(resample_bars(frame, "1W").iloc[0]["volume"])
+
+ def test_timezone_and_attributes_are_kept(self):
+ frame = _daily("2024-01-01", "2024-01-10", freq="D")
+ frame.index = frame.index.tz_localize("Asia/Shanghai")
+ frame.attrs["quote_currency"] = "CNY"
+
+ weekly = resample_bars(frame, "1W")
+
+ assert str(weekly.index.tz) == "Asia/Shanghai"
+ assert weekly.attrs == {"quote_currency": "CNY"}
+
+ def test_resampling_twice_changes_nothing(self):
+ weekly = resample_bars(_daily("2024-01-01", "2024-03-01"), "1W")
+ pd.testing.assert_frame_equal(resample_bars(weekly, "1W"), weekly)
+
+ @pytest.mark.parametrize("interval", ["1D", "1H", "1m"])
+ def test_other_intervals_pass_through(self, interval):
+ frame = _daily("2024-01-01", "2024-01-10")
+ assert resample_bars(frame, interval) is frame
+ assert source_interval(interval) == interval
+
+ def test_the_period_intervals_are_fetched_daily(self):
+ assert source_interval("1W") == source_interval("1M") == "1D"
+
+
+class _RecordingLoader:
+ name = "yahoo"
+ asked: list[str] = []
+
+ def is_available(self) -> bool:
+ return True
+
+ def fetch(self, codes, start_date, end_date, fields=None, interval="1D"): # noqa: ANN001, ANN201
+ type(self).asked.append(interval)
+ return {code: _daily("2024-01-01", "2024-03-29") for code in codes}
+
+
+def recording_loader():
+ _RecordingLoader.asked = []
+ return _RecordingLoader
+
+
+class TestFetchPaths:
+ @pytest.mark.parametrize(("interval", "bars"), [("1W", 13), ("1M", 3)])
+ def test_a_backtest_asks_for_daily_bars_and_gets_period_bars(
+ self, monkeypatch, recording_loader, interval, bars
+ ):
+ monkeypatch.setattr(runner, "_get_loader", lambda source: recording_loader)
+ result = runner.fetch_data_map(
+ {"source": "yahoo", "codes": ["AAPL.US"], "start_date": "2024-01-01", "end_date": "2024-03-29", "interval": interval}
+ )
+
+ assert recording_loader.asked == ["1D"]
+ assert len(result.data_map["AAPL.US"]) == bars
+
+ def test_the_auto_route_asks_for_daily_bars_too(self, monkeypatch, recording_loader):
+ monkeypatch.setattr(runner, "resolve_loader", lambda market: recording_loader())
+ monkeypatch.setattr(runner, "LOADER_REGISTRY", {})
+ result = runner.fetch_data_map(
+ {"source": "auto", "codes": ["AAPL.US"], "start_date": "2024-01-01", "end_date": "2024-03-29", "interval": "1M"}
+ )
+
+ assert recording_loader.asked == ["1D"]
+ assert list(result.data_map["AAPL.US"].index.strftime("%Y-%m-%d")) == ["2024-01-31", "2024-02-29", "2024-03-29"]
+
+ def test_get_market_data_asks_for_daily_bars(self, recording_loader):
+ out = fetch_market_data(
+ codes=["AAPL.US"],
+ start_date="2024-01-01",
+ end_date="2024-03-29",
+ source="yahoo",
+ interval="1M",
+ max_rows=0,
+ loader_resolver=lambda source: recording_loader,
+ )
+
+ assert recording_loader.asked == ["1D"]
+ rows = out["AAPL.US"]
+ assert [row["trade_date"][:10] for row in rows] == ["2024-01-31", "2024-02-29", "2024-03-29"]
+
+ def test_the_benchmark_is_built_the_same_way(self, recording_loader):
+ frame = _fetch_benchmark("SPY", "2024-01-01", "2024-03-29", "1W", loader=recording_loader())
+
+ assert recording_loader.asked == ["1D"]
+ assert len(frame) == 13
+
+ def test_the_config_accepts_both(self):
+ for interval in ("1W", "1M"):
+ runner.BacktestConfigSchema(codes=["AAPL.US"], start_date="2024-01-01", end_date="2024-03-29", interval=interval)
+ with pytest.raises(ValueError):
+ runner.BacktestConfigSchema(codes=["AAPL.US"], start_date="2024-01-01", end_date="2024-03-29", interval="1w")
+
+
+class TestAnnualisation:
+ @pytest.mark.parametrize("source", ["tushare", "yahoo", "okx", "mt5"])
+ def test_a_week_and_a_month_are_calendar_counts(self, source):
+ assert calc_bars_per_year("1W", source) == 52
+ assert calc_bars_per_year("1w", source) == 52
+ assert calc_bars_per_year("1M", source) == 12
+
+ def test_a_minute_is_still_a_minute(self):
+ assert calc_bars_per_year("1m", "yahoo") == 252 * 390
+
+ def test_the_funding_span_of_a_period_bar(self):
+ assert _interval_span_hours("1W") == 168.0
+ assert _interval_span_hours("1M") == 730.5
+ assert _interval_span_hours("1m") == pytest.approx(1 / 60)
+
+ def test_attribution_reads_1M_as_a_month(self):
+ assert _attribution_bars_per_year("1M") == 12
+ assert _attribution_bars_per_year("1m") == 98280
+ assert _attribution_bars_per_year("1W") == 52
+
+
+def test_the_indicator_tool_fetches_whole_periods(monkeypatch):
+ import src.tools.technical_indicator_tool as mod
+
+ seen = {}
+
+ def _fake_fetch(**kwargs):
+ seen.update(kwargs)
+ return {kwargs["codes"][0]: _daily("2020-01-01", "2024-03-29")}
+
+ monkeypatch.setattr(mod, "fetch_market_data", _fake_fetch)
+ out = json.loads(mod.TechnicalIndicatorTool().execute(symbol="AAPL.US", interval="1wk", lookback=100))
+
+ assert seen["interval"] == "1W"
+ start = datetime.strptime(seen["start_date"], "%Y-%m-%d")
+ assert datetime.now() - start >= timedelta(days=699)
+ assert out["interval"] == "1W"