A holding period is the length of time a position, or a specified lot within it, is held. For a completed trade, it runs from entry to exit under a declared timing convention. It determines the interval over which the trade earns its return and carries market exposure, financing or borrow costs. A strategy's intended holding time can differ from the duration actually achieved.
Nasdaq's glossary gives the basic time-held definition. In a backtest, making that definition operational requires entry and exit timestamps, a clock and a rule for matching quantities.
Five days between six observations
Consider a synthetic daily ledger. One unchanged lot of one unit is bought on day 2 at 16:00 UTC and sold in full on day 7 at 16:00 UTC. The day labels denote consecutive calendar days, with each observation exactly 24 hours after the previous one. Entry and exit mean executed fills; there are no additions or partial exits.
For entry time and exit time , elapsed holding time is
Here , or 120 hours. Counting day labels 2, 3, 4, 5, 6 and 7 gives six observation points, with only five intervals between them. Including both endpoint observations does not add another day of elapsed exposure.
Original synthetic daily timeline for one unchanged lot of one unit. Every observation is at 16:00 UTC on consecutive calendar days. The alternative open-position case uses the same observation cutoff, without an exit.
A count in trading sessions answers a different question. State the venue calendar and whether the entry and exit sessions are counted. Likewise, a bar count needs the bar frequency and endpoint rule. Missing observations or a market closure do not stop calendar time, so a row count alone is insufficient to recover elapsed duration.
Decide what is being held
A fill ledger can produce several legitimate duration measures, provided the convention is visible:
| Unit of measurement | Entry and exit convention |
|---|---|
| Matched lot or quantity slice | Each quantity has a matched acquisition and disposal; partial exits close only the matched quantity |
| Position episode | First entry from flat through the final return to flat; a sign reversal closes one episode and starts another |
| Open position at a reporting cutoff | Entry through the cutoff measures current age; the eventual exit remains unknown |
Scaling into a position creates lots with different entry times. A first-in-first-out matching rule and a specific-lot rule can therefore give different lot durations even when they describe the same fills. The age of the original lot is not automatically the age of newly added quantity. A flat-to-flat episode may continue through several partial reductions and additions.
Choose the instrument, account and strategy scope too. Offsetting positions in separate strategies can leave aggregate net exposure flat while both underlying trades remain open. A net-position series alone cannot identify their separate holding periods.
Open age is an incomplete duration
In the example, suppose the lot is still open at the day-7 reporting cutoff. Its observed age is five days. Its final holding period is unknown: no exit has occurred. This is an application of right censoring, where observation ends before the event of interest. NIST's explanation of censoring distinguishes an observed run-time from a completed event time; here the event is a trade's exit.
An average holding period computed only from closed trades describes those closed trades. It may leave out long-lived open positions. Report that population and the open-position ages separately. If a backtest forces liquidation at its final timestamp, that is a specified exit with execution and cost assumptions, rather than evidence that the strategy would naturally have closed then.
Use the duration without mixing clocks
The entry rule and exit rule define the intended trading process. A maximum holding-time rule is a limit in that process; a profit target, stop or other exit can end the trade earlier. A triggered order can also wait for a fill. Measure realised duration from the declared execution timestamps rather than substituting signal times.
An investment horizon describes when capital is intended to serve an investor's goal. It can span many successive trades. A holding-period return measures performance over a specified interval; the holding period itself measures time, not profit.
Duration also matters when research outcomes overlap. Trades entered on different dates can share some of the same return increments. Labels and Overlapping Outcomes develops feasible entry and exit references, shared outcome information and unfinished labels. The linked page provides a public preview; the full lesson requires membership.
Finally, a trading-duration convention is not a universal tax rule. For example, the IRS's 2025 Publication 550, for U.S. federal tax treatment of securities traded on an established market, starts the holding-period count after the purchase trade date and ends it on the sale trade date. It explicitly distinguishes trade dates from settlement dates. Other assets and circumstances have their own rules. Use the applicable legal convention for that purpose, rather than treating a timestamp subtraction as a complete tax classification. IRS Publication 550
Sources
- Nasdaq — Holding period: basic definition as time held.
- Marcos López de Prado, Advances in Financial Machine Learning, first edition, Wiley, 2018, Chapter 14, §14.3, “General Characteristics”: holding periods, position episodes and the distinction from bet frequency.
- NIST/SEMATECH e-Handbook — Censoring: observed run-times when an event has not occurred before observation ends.
- Internal Revenue Service — Publication 550 (2025), Holding Period: the stated U.S. federal tax counting convention and its distinction from settlement dates; not a jurisdiction-independent definition.