Market calendars

NYSE sessions, 24/7 crypto, event-resolution windows. Gate order submission on what the venue will accept.

Backtests do not care what time it is. The loop advances bar by bar and positions change whenever data arrives. Live trading cares. An order sent into a closed NYSE session dangles until open. A halted LULD window gets orders rejected. A prediction market after resolution is an error.

horizon.calendars (plural, to avoid Python’s stdlib calendar) provides one Protocol and three implementations. The live loop calls calendar.is_open(ts) before every submission. See Watchdog for the integration.

Protocol

python

class ExchangeCalendar(Protocol):
    name: str
    def is_open(self, ts: datetime) -> bool: ...
    def session_kind(self, ts: datetime) -> SessionKind: ...
    def next_open(self, ts: datetime) -> datetime | None: ...
    def next_close(self, ts: datetime) -> datetime | None: ...
    def previous_close(self, ts: datetime) -> datetime | None: ...
    def current_session(self, ts: datetime) -> MarketSession | None: ...

`NYSECalendar`

US equities, ETFs, equity options.

python

from horizon.calendars import NYSECalendar, SessionKind
from datetime import datetime
from zoneinfo import ZoneInfo

nyse = NYSECalendar()
nyse_ext = NYSECalendar(include_extended_hours=True)   # 4am-9:30am + 4pm-8pm ET

et = ZoneInfo("America/New_York")
ts = datetime(2026, 3, 2, 10, 30, tzinfo=et)   # Monday 10:30 ET
nyse.is_open(ts)                                # True
nyse.session_kind(ts)                           # SessionKind.Regular
nyse.current_session(ts).is_early_close          # False

Holidays and half-days

With pandas_market_calendars (in the optional equity extra) NYSECalendar uses accurate US holiday detection and half-day (13:00 ET close) identification:

bash

pip install pandas-market-calendars
# or
pip install "horizon[equity]"

Without it, the calendar falls back to a Mon-Fri 9:30-16:00 ET rule. Fine for dev, not production.

Next / previous

python

ts = datetime(2026, 3, 6, 18, 0, tzinfo=et)   # Friday evening
next_open  = nyse.next_open(ts)                # Monday 9:30 ET
next_close = nyse.next_close(ts)               # Monday 16:00 ET (or 13:00 early close)
prev_close = nyse.previous_close(ts)           # Friday 16:00 ET

`AlwaysOpenCalendar`

Crypto and perps.

python

from horizon.calendars import AlwaysOpenCalendar

crypto = AlwaysOpenCalendar()
crypto.is_open(any_timestamp)                   # True
crypto.session_kind(ts)                         # SessionKind.Regular
crypto.next_open(ts)                            # returns ts
crypto.next_close(ts)                           # None

Use for Hyperliquid, Binance, Coinbase, Kraken, most perps venues.

`ResolutionCalendar`

Polymarket and Kalshi events trade until the question resolves, then settle to $1 or $0 per Yes-share.

python

from horizon.calendars import ResolutionCalendar
from datetime import datetime, timezone

elections = ResolutionCalendar()
elections.register(
    "will_candidate_x_win_2028",
    resolution_ts=datetime(2028, 11, 8, 23, 59, tzinfo=timezone.utc),
)

# Before resolution
elections.is_open(datetime(2026, 10, 1, tzinfo=timezone.utc),
                  market_id="will_candidate_x_win_2028")   # True

# After resolution
elections.is_open(datetime(2028, 11, 9, tzinfo=timezone.utc),
                  market_id="will_candidate_x_win_2028")   # False

The L1 lifecycle manager listens for resolution_ts and emits AuditCategory.PredictionResolved events that credit winning positions.

Per-venue wiring

python

calendars = {
    "alpaca":      NYSECalendar(include_extended_hours=False),
    "ibkr":        NYSECalendar(include_extended_hours=True),
    "hyperliquid": AlwaysOpenCalendar(),
    "polymarket":  ResolutionCalendar(),
    "kalshi":      ResolutionCalendar(),
}

Each Venue picks up its calendar at construction. The execution loop consults the calendar on the tick timestamp, not wall clock.

Watchdog integration

Attach a calendar to LiveWatchdog. The watchdog then:

Halts new submissions outside the session (halt_outside_session=True).
Fires a one-shot flatten N seconds before the scheduled close (flatten_before_close_seconds > 0).
Respects half-days and holidays (via pandas_market_calendars if installed).

python

from horizon.calendars import NYSECalendar
from horizon.ops import LiveWatchdogConfig

hz.run(
    mode="live",
    feed=my_feed,
    watchdog=LiveWatchdogConfig(
        session_calendar=NYSECalendar(),
        halt_outside_session=True,
        flatten_before_close_seconds=120,
    ),
    ...,
)

Status

	L0	L1+
Protocol, NYSE, AlwaysOpen, Resolution	Shipped
`pandas_market_calendars` integration	Shipped (opt-in)
Half-day (13:00 close) detection	Shipped
Gate order submission on `is_open`		Shipped via watchdog
Scheduled flatten before close		Shipped via watchdog
Option / future expiry scheduling via calendar		Planned
LULD / circuit-breaker halt detection		From feed
Non-US venues (LSE, TSE)		Add per-venue impls