How long should you wait before trading a newly listed perpetual?

Crypto exchanges list new perpetual futures every week, and Binance alone has listed 832 USDT-margined perpetuals, so a research universe built on that exchange mixes contracts listed last week with contracts that have traded for years. Both kinds usually share one cost assumption and one strategy library, even though a new contract has no settled basis, no settled funding schedule and far more volatility than an old one.

That raises a descriptive question about how a contract's behaviour changes as it ages and how long the change lasts, plus a practical one: does a fixed set of trading rules earn anything different on a young contract than on a mature one after real costs? If it does, a listing calendar is an opportunity set; if it does not, age is still something to control for, because young contracts pollute any backtest that pools them with old ones.

Answering the second question cleanly runs into two problems, since listings arrive in waves that follow the market cycle, so pooling young contracts across years mostly measures the years, and young contracts are far more volatile, so any rule that scales with volatility looks different on them without age mattering at all.

Everything below is built around those two problems, with the exact methodology behind every number on a separate reproducibility page.

Contracts: The Binance universe is every USDT-margined perpetual that has ever traded, rebuilt from the public data archive rather than from the live exchange listing, so contracts the exchange later removed are included. Age counts from the first day the perpetual traded with real volume, not from the token's launch or its spot listing. Contracts listed up to the end of 2021 form a calibration set used only to choose the trading rules, while those listed from 2022 onward form the Binance test set.

Rules: 20 mechanical entry-and-exit rules, each with a target, a stop and a maximum holding time, were chosen once from a fixed grid on the calibration set using mature contract-days only. They then run unchanged on every contract at every age, with nothing tuned per age or per contract.

Comparison: Absolute returns on young contracts mix the age effect with whatever the market did that week, so the measured quantity is a difference: the same rules on the same calendar day, on a young contract (3 to 30 days old) against mature contracts (over a year old) in the same volatility bucket. Matching on the day removes the market regime, and matching on volatility removes the mechanical effect of young contracts simply moving more.

Costs: Every fill pays the taker fee, the quoted half-spread measured for a contract of that age and the price impact of a $10,000 order measured from the order book. Positions also pay or receive the funding actually charged on the contract, and no cost is taken from a price-range estimator, since the one tested below fails on this data.

Intrabar order: Rules trade on hourly bars, which leaves the order of events open whenever one bar touches both the target and the stop, and on Binance that order is reconstructed from one-minute data. For the out-of-sample test every such bar resolves as a stop, which can only understate what the rules earn, and Binance is reported both ways so the size of that handicap is visible.

Out-of-sample test: Every design choice was fixed on Binance and frozen before the test data was touched. That test uses a second exchange, Bybit, restricted to price histories the Binance data never contained: 79 contracts whose token never had a Binance perpetual and 77 that Bybit listed at least 30 days before Binance did, counting only the days before the Binance listing. Tokens are matched across venues by identity rather than symbol, since the same token often trades under a different multiplier prefix on each, giving 156 contracts and 4,284 young contract-days with a volatility estimate, compared on the same calendar day against mature Bybit contracts listed from 2022 onward, in a test run once.

This table covers the 691 Binance contracts listed from 2022 onward with enough history to measure, taking the median across contracts in each age bucket. Funding is converted to an eight-hour equivalent because new contracts often settle every four hours or every hour.

Volatility falls with every bucket, from 2.44 times the mature level on the first day to 1.33 times it by the end of the first month, while correlation to Bitcoin rises with every bucket from 0.20 to 0.50. On its first day a contract trades 48 times the volume it will trade once it is a year old.

Funding stays at 3.2 to 4.2 times the mature level for the whole first year rather than just the first weeks, with 0.75 bp per eight hours at 8 to 30 days against 0.18 past a year, a ratio of 4.15, and its dispersion peaks on the first day.

When a contract was listed also decides its settlement schedule. Every contract listed from 2020 to 2022 started on eight-hour funding, against 84.0% of 2024 listings and 89.3% of 2025 listings that started on four hours. Across all 832 contracts, 47.6% never settle at eight hours and 32.7% change interval at least once, and although none started on hourly settlement, 6.7% of payments in the first week and 8.1% in days 8 to 30 are hourly.

Quoted spreads are widest on the first day and settle within about two weeks. Binance published best bid and offer data from May 2023 to March 2024, and 86 contracts were listed inside that window, which allows each of them to be compared with itself later on.

In absolute terms these spreads are small: the median quoted half-spread is 2.57 bp on the first day, 1.56 bp at 2 to 7 days and 1.22 bp at 8 to 30 days, against 1.18 bp for contracts past a year. Between days 8 to 30 and maturity the ratio is 1.04, although that comparison sets different contracts from different listing years against each other and so is not a within-contract measurement.

Order-book depth moves against the spread pattern, measured as the cost of sweeping a $10,000 buy through the book, excluding the half-spread, on one fixed sample of days for every contract: ages 3, 7, 14, 28, 45, 75, 120, 160, 240 and 300 plus eight days spread across its life past one year.

A contract's book is deepest in its first month, where the same contract costs less than a third as much to trade in size during days 3 to 30 as it does once it is a year old, and that holds for close to nine contracts in ten. Its mature measurements necessarily fall later in calendar time than the young ones, so part of this gap could be a market-wide change in depth, which this design does not separate from age.

Contracts in the Bybit out-of-sample set are much thinner, with a $10,000 market order costing a median of 29.0 bp against mid at 2 to 7 days and 29.1 bp at 8 to 30 days, of which 3.8 and 4.0 bp is half-spread, while Bybit contracts listed from 2022 and past a year old cost 29.8 bp, all measured from the exchange's order-book archive on 706 contract-days.

Where quotes are not available, a standard shortcut estimates the spread from each bar's high, low and close. The Abdi-Ranaldo estimator was tested against real quoted spreads on 987 contract-days across the 86 contracts listed inside the quote window.

Since the estimator measures the full spread, about half of the gap in level is definitional. Its rank correlation with the real spread is +0.035 over all 987 pairs, so the estimator carries no information about which contract-day is expensive to trade, and 362 of its 987 estimates are exactly zero. On hourly crypto perpetual data it should not be used to set per-contract costs.

Each share counts only contracts old enough to have reached that age before the archive ends in July 2026, and of the 692 contracts listed from 2022 onward, 108 no longer trade.

Neither venue shows a difference, measured as the library's average net P&L on a young contract-day minus the same on matched mature contract-days, in basis points of position per day, with 95% intervals from a bootstrap clustered by week.

On Binance the rules do slightly worse on young contracts, with an interval that includes zero. For Bybit, where the test was frozen in advance and run once, the estimate is -1.46 bp with an interval of about 17.6 bp either side, ruling out any age effect on the long-short book larger than that in either direction. Reading the hourly bars one way or the other barely matters, since resolving every double touch as a stop moves the Binance estimate by 0.08 bp.

No bucket inside the young window differs from zero either. In the first week, where the sample is smallest, the point estimates are the most negative (-16.70 bp on Binance, -23.63 bp on Bybit), yet both intervals include zero.

Splitting the book by side gives a consistent direction on both venues, though neither one-sided book was the registered test:

Long positions do worse on young contracts and short positions do better, by similar amounts on both venues, which fits young contracts drifting down against mature ones over their first month while the long-short book nets the drift out. It remains a lead, since only the Binance long arm excludes zero and that sample was the development set.

The library is weak by design, being the best 20 rules from a fixed grid of 13,770 configurations on contracts listed up to 2021, and none of the 12,555 configurations with enough activity was profitable after costs on those contracts, so the comparison asks whether age changes what these rules earn and a stronger library could respond to age differently.

Its out-of-sample set is also smaller than the Binance one, with 3,295 matched young contract-days against 15,244. That leaves an interval of about 17.6 bp per day either side of the estimate, and a smaller effect would go undetected.

Bybit's order-book archive starts in 2023, so contract-days before it are charged the median measured cost of their age bucket rather than their own. Execution cost on Bybit is measured at the registered $10,000 size, where these thin contracts are expensive, and a smaller book would pay less on both legs. Age on Bybit counts from the Bybit listing, so 23 contracts in its mature leg are tokens that already had a Binance perpetual before 2022.

Hourly bars that touch both the target and the stop resolve as a stop on Bybit, which can only understate what the rules earn, and the two Binance readings show how little that matters here.

Age and calendar time move together as well. Matching young and mature contracts on the same day removes the market regime from the trading comparison, but the descriptive depth and spread results compare a contract with itself at different dates and cannot separate age from a market-wide drift.

For the exact universe, rule grid, cost measurements and inference, see the reproducibility page.