Exchange Calendars and Trading Holidays

Every exchange publishes a calendar of which days it's open, which days it's fully closed (holidays), and which days it closes early (the day after Thanksgiving in the U.S., or Christmas Eve), and these vary by country, exchange, and even asset class within the same country — bond markets and equity markets don't always share the same holiday list. A backtest or live system that simply assumes "every Monday through Friday is a trading day" will silently misalign with reality on dozens of days a year.

The consequences show up in specific, avoidable ways: a strategy that computes a daily return using the previous "day's" close actually spans a three-day gap over a holiday weekend without adjusting for it; a model trained on daily bars mistakenly counts a holiday as a zero-volume, zero-return day rather than excluding it entirely; or a live system placed a resting order the night before a market holiday and it never got a chance to fill or cancel as expected. Early closes cause a subtler version of the same problem — a strategy that always trades in "the last 30 minutes" trades at the wrong time of day, and volume/liquidity assumptions calibrated to a normal session don't hold on a half-day.

The fix is to maintain (or use a maintained library providing) an explicit trading calendar per exchange, keyed by date, marking full sessions, early closes, and full closures, and have every date-handling piece of the pipeline — data ingestion, backtest day-stepping, live order scheduling — consult it rather than reasoning about weekdays directly.