Time-series momentum strategy
On ES futures from January 2016 to July 2026, the time-series momentum rule made $158,958 after costs and made money in 8 of 11 years.
TrendSwingDaily barsBacktested Nov 6, 2026
Time-series momentum backtest, the result
Contract
E-mini S&P 500
-
+$159K
Net after costs
After $767 in fees and slippage
-
8 of 11
Years that made money
January 2016 to July 2026
-
−$75.9K
Largest drawdown
Peak to trough, 1 contract
-
$3.8K
Average trade
43 trades
Net after costs by month, January 2016 to July 2026.
Time-series momentum rules
At each month's end, go long if the price is up over the last 12 months, and short if it is down.
A drawing of the rules on a made-up day, not a result.
- At the close of each month's last trading day, the strategy reads the change in price over the last 252 trading days.
- If it is up, it holds one contract long for the month ahead. If it is down, one contract short.
- A roll closes the position, and the first daily close on the new contract takes the same side back.
- Decides on
- Daily bars
- Holds
- Overnight and across days
- Parameters
- lookback 252, session minute 1,020
Results by year
It made money in 8 of 11 years on ES. Its best year was 2021, +$51,911, and its worst 2022, −$16,693.
| Year | Net | Trades | Win rate | Drawdown | Buy and hold |
|---|---|---|---|---|---|
| 2016 | −$1,227 | 0 | −$1,962 | +$11,497 | |
| 2017 | +$21,924 | 4 | 100% | −$3,991 | +$21,936 |
| 2018 | +$874 | 4 | 75% | −$18,478 | −$8,589 |
| 2019 | +$517 | 6 | 50% | −$25,001 | +$35,436 |
| 2020 | −$11,755 | 5 | 60% | −$71,282 | +$27,761 |
| 2021 | +$51,911 | 4 | 100% | −$13,941 | +$51,361 |
| 2022 | −$16,693 | 5 | 20% | −$44,661 | −$45,939 |
| 2023 | +$9,344 | 5 | 80% | −$28,029 | +$38,435 |
| 2024 | +$42,772 | 4 | 100% | −$30,062 | +$42,947 |
| 2025 | +$36,860 | 4 | 75% | −$69,341 | +$37,010 |
| 2026, Jan to Jul | +$24,430 | 2 | 50% | −$36,954 | +$24,355 |
Time-series momentum backtest results
On ES it made $159,725 before costs. Fees and slippage took $767 of that.
- Net = before costs − fees − slippage
- ES: $159,725 − $155 − $613 = $158,958
The interval holds zero: this test cannot tell the average trade from no edge.
Time-series momentum code and Pine Script
The strategy as the backtest ran it, and the Pine Script our compiler wrote from it for TradingView.
1from atlantive import BUY, NEW_YORK, SELL, Bars, Rolling, Session, Strategy2 3 4class TimeSeriesMomentum(Strategy):5 """Time-series momentum: long or short for the month ahead on the sign of the last twelve months.6 7 A day is the CME trading day, 18:00 to 17:00 New York (the run's daily bars). At the close of each8 month's last trading day the strategy reads the change in price over the last 252 trading days:9 the mean of the daily changes over that window has the same sign. Up, it holds one contract long10 for the month ahead; down, one contract short. A roll closes the position, and the first daily11 close on the new contract takes the same side back, from the sign the 252 days show there if no12 month has ended yet on that contract's replay. Each decision is taken at a daily close and fills13 at the next session's first one-second bar, at the 18:00 New York reopen.14 """15 16 data = "bars"17 18 class Params:19 lookback: int = 25220 session_minute: int = 102021 clip: int = 122 23 class State:24 clock: Session(zone=NEW_YORK)25 bars: Bars(interval_ns=86_400_000_000_000)26 changes: Rolling(window=252)27 last_close: float28 seen: int29 target: int30 31 def on_event(self, view, ev, orders):32 self.clock.update(ev)33 if not self.bars.update(ev):34 return35 close = float(self.bars.prev_close)36 if self.seen > 0:37 self.changes.update(close - self.last_close)38 self.last_close = close39 self.seen += 140 closed_day = self.clock.prev_day41 if self.clock.prev_minute >= self.p.session_minute:42 closed_day = closed_day + 143 next_day = self.clock.day44 if self.clock.minute >= self.p.session_minute:45 next_day = next_day + 146 # The civil month of each trade date (Hinnant's days-to-civil, every divisor a literal).47 z = closed_day + 71946848 era = z // 14609749 doe = z - era * 14609750 yoe = (doe - doe // 1460 + doe // 36524 - doe // 146096) // 36551 year = yoe + era * 40052 doy = doe - (365 * yoe + yoe // 4 - yoe // 100)53 mp = (5 * doy + 2) // 15354 month = mp + 355 if mp >= 10:56 month = mp - 957 if month <= 2:58 year = year + 159 closed_month = year * 12 + month60 z = next_day + 71946861 era = z // 14609762 doe = z - era * 14609763 yoe = (doe - doe // 1460 + doe // 36524 - doe // 146096) // 36564 year = yoe + era * 40065 doy = doe - (365 * yoe + yoe // 4 - yoe // 100)66 mp = (5 * doy + 2) // 15367 month = mp + 368 if mp >= 10:69 month = mp - 970 if month <= 2:71 year = year + 172 next_month = year * 12 + month73 if self.changes.count < self.p.lookback:74 return75 if next_month != closed_month or self.target == 0:76 self.target = 177 if self.changes.mean() < 0.0:78 self.target = -179 if self.target > 0:80 if view.position < 0:81 orders.ioc(side=BUY, qty=abs(view.position) + self.p.clip)82 elif view.position == 0:83 orders.ioc(side=BUY, qty=self.p.clip)84 else:85 if view.position > 0:86 orders.ioc(side=SELL, qty=abs(view.position) + self.p.clip)87 elif view.position == 0:88 orders.ioc(side=SELL, qty=self.p.clip) 1//@version=62// TimeSeriesMomentum, exported from Atlantive.3// Chart simulation: decisions use completed bars; market orders fill at the next bar open.4// Native fills, fees, data gaps and session boundaries can differ. Re-test on the target symbol.5// Use standard time-based candles and the strategy's timeframe, not synthetic chart prices.6// Pine v6 trims old trade records at its order limit; the strategy keeps calculating.7// Margin is explicitly disabled to retain the native signal simulation's sizing.8strategy("TimeSeriesMomentum", overlay=true, initial_capital=100000.0, commission_type=strategy.commission.cash_per_contract, commission_value=1.68, default_qty_type=strategy.fixed, default_qty_value=1, calc_on_every_tick=false, process_orders_on_close=false, slippage=0, margin_long=0, margin_short=0)9 10if timeframe.in_seconds() != 8640011 runtime.error("TimeSeriesMomentum was exported for the 1 day timeframe; set the chart to 1D.")12 13lookback = input.int(252, "lookback", minval=-9007199254740991, maxval=9007199254740991)14session_minute = input.int(1020, "session_minute", minval=-9007199254740991, maxval=9007199254740991)15clip = input.int(1, "clip", minval=-9007199254740991, maxval=9007199254740991)16 17_q_safe_int(x) =>18 if math.abs(x) > 900719925474099119 runtime.error("an integer left the exact chart range; this run cannot be compared faithfully")20 int(x)21_q_int(x) => _q_safe_int(x < 0 ? math.ceil(x) : math.floor(x))22_q_iadd(a, b) => _q_safe_int(a + b)23_q_isub(a, b) => _q_safe_int(a - b)24_q_imul(a, b) =>25 if a != 0 and math.abs(b) > 9007199254740991 / math.abs(a)26 runtime.error("an integer multiplication left the exact chart range; this run cannot be compared faithfully")27 _q_safe_int(a * b)28_q_ifloordiv(a, b) =>29 _q_r = a % b30 _q_q = (a - _q_r) / b31 _q_safe_int(_q_r != 0 and ((_q_r < 0) != (b < 0)) ? _q_q - 1 : _q_q)32 33_q_close = _q_int(math.round(close * 1000000000.0))34 35var float last_close = 0.036var int seen = 037var int target = 038// clock: completed chart bar's opening timestamp (America/New_York).39_q_clock_prev_minute = hour(time, "America/New_York") * 60 + minute(time, "America/New_York")40_q_clock_prev_second = _q_clock_prev_minute * 60 + second(time, "America/New_York")41_q_clock_prev_weekday = (dayofweek(time, "America/New_York") + 5) % 742_q_clock_prev_day = int(timestamp("UTC", year(time, "America/New_York"), month(time, "America/New_York"), dayofmonth(time, "America/New_York"), 0, 0) / 86400000)43// Current clock: this chart bar's scheduled close, known without a future bar.44// A native next event after a data gap can carry a later time.45_q_clock_minute = hour(time_close, "America/New_York") * 60 + minute(time_close, "America/New_York")46_q_clock_second = _q_clock_minute * 60 + second(time_close, "America/New_York")47_q_clock_weekday = (dayofweek(time_close, "America/New_York") + 5) % 748_q_clock_day = int(timestamp("UTC", year(time_close, "America/New_York"), month(time_close, "America/New_York"), dayofmonth(time_close, "America/New_York"), 0, 0) / 86400000)49_q_clock_seen = 150var _q_changes_buf = array.new_float(252, 0.0)51var int _q_changes_idx = 052var int _q_changes_count = 053var float _q_changes_off = 0.054var float _q_changes_sum = 0.055var float _q_changes_sumsq = 0.056var int _q_changes_skipped = 057float close_l = na58int closed_day = na59int next_day = na60int z = na61int era = na62int doe = na63int yoe = na64int year_l = na65int doy = na66int mp = na67int month_l = na68int closed_month = na69int next_month = na70 71bool _q_running = true72close_l := _q_close * 1.073if seen > 074 _q_sample1 = close_l - last_close75 if na(_q_sample1) or math.abs(_q_sample1) > 1.7976931348623157e30876 _q_changes_skipped := _q_changes_skipped + 177 else78 if _q_changes_count == 079 _q_changes_off := _q_sample180 if _q_changes_count >= 25281 _q_old2 = array.get(_q_changes_buf, _q_changes_idx) - _q_changes_off82 _q_changes_sum := _q_changes_sum - _q_old283 _q_changes_sumsq := _q_changes_sumsq - _q_old2 * _q_old284 else85 _q_changes_count := _q_changes_count + 186 _q_dev2 = _q_sample1 - _q_changes_off87 array.set(_q_changes_buf, _q_changes_idx, _q_sample1)88 _q_changes_sum := _q_changes_sum + _q_dev289 _q_changes_sumsq := _q_changes_sumsq + _q_dev2 * _q_dev290 _q_changes_idx := (_q_changes_idx + 1) % 25291 if _q_changes_idx == 0 and _q_changes_count >= 25292 _q_origin2 = _q_changes_off + _q_changes_sum / _q_changes_count93 _q_total2 = 0.094 _q_totalsq2 = 0.095 for _q_j2 = 0 to 25196 _q_e2 = array.get(_q_changes_buf, _q_j2) - _q_origin297 _q_total2 := _q_total2 + _q_e298 _q_totalsq2 := _q_totalsq2 + _q_e2 * _q_e299 _q_changes_off := _q_origin2100 _q_changes_sum := _q_total2101 _q_changes_sumsq := _q_totalsq2102last_close := close_l103seen := _q_int(_q_iadd(seen, 1))104closed_day := _q_int(_q_clock_prev_day)105if _q_clock_prev_minute >= session_minute106 closed_day := _q_int(_q_iadd(closed_day, 1))107next_day := _q_int(_q_clock_day)108if _q_clock_minute >= session_minute109 next_day := _q_int(_q_iadd(next_day, 1))110z := _q_int(_q_iadd(closed_day, 719468))111era := _q_int(_q_ifloordiv(z, 146097))112doe := _q_int(_q_isub(z, _q_imul(era, 146097)))113yoe := _q_int(_q_ifloordiv(_q_isub(_q_iadd(_q_isub(doe, _q_ifloordiv(doe, 1460)), _q_ifloordiv(doe, 36524)), _q_ifloordiv(doe, 146096)), 365))114year_l := _q_int(_q_iadd(yoe, _q_imul(era, 400)))115doy := _q_int(_q_isub(doe, _q_isub(_q_iadd(_q_imul(365, yoe), _q_ifloordiv(yoe, 4)), _q_ifloordiv(yoe, 100))))116mp := _q_int(_q_ifloordiv(_q_iadd(_q_imul(5, doy), 2), 153))117month_l := _q_int(_q_iadd(mp, 3))118if mp >= 10119 month_l := _q_int(_q_isub(mp, 9))120if month_l <= 2121 year_l := _q_int(_q_iadd(year_l, 1))122closed_month := _q_int(_q_iadd(_q_imul(year_l, 12), month_l))123z := _q_int(_q_iadd(next_day, 719468))124era := _q_int(_q_ifloordiv(z, 146097))125doe := _q_int(_q_isub(z, _q_imul(era, 146097)))126yoe := _q_int(_q_ifloordiv(_q_isub(_q_iadd(_q_isub(doe, _q_ifloordiv(doe, 1460)), _q_ifloordiv(doe, 36524)), _q_ifloordiv(doe, 146096)), 365))127year_l := _q_int(_q_iadd(yoe, _q_imul(era, 400)))128doy := _q_int(_q_isub(doe, _q_isub(_q_iadd(_q_imul(365, yoe), _q_ifloordiv(yoe, 4)), _q_ifloordiv(yoe, 100))))129mp := _q_int(_q_ifloordiv(_q_iadd(_q_imul(5, doy), 2), 153))130month_l := _q_int(_q_iadd(mp, 3))131if mp >= 10132 month_l := _q_int(_q_isub(mp, 9))133if month_l <= 2134 year_l := _q_int(_q_iadd(year_l, 1))135next_month := _q_int(_q_iadd(_q_imul(year_l, 12), month_l))136if _q_changes_count < lookback137 _q_running := false138if _q_running139 if next_month != closed_month or target == 0140 target := _q_int(1)141 if (_q_changes_count == 0 ? 0.0 : _q_changes_off + _q_changes_sum / _q_changes_count) < 0.0142 target := _q_int(-1)143if _q_running144 if target > 0145 if _q_int(strategy.position_size) < 0146 _q_qty3 = _q_int(_q_iadd(math.abs(_q_int(strategy.position_size)), clip))147 if _q_qty3 <= 0148 runtime.error("market order quantity must be a positive whole unit")149 _q_pos3 = _q_int(strategy.position_size)150 if _q_pos3 > 0 or _q_pos3 < 0 and _q_qty3 < -_q_pos3151 runtime.error("this order would add to or partially reduce a position, so its trade ledger would not be faithful")152 strategy.order("q1", strategy.long, _q_qty3)153 else if _q_int(strategy.position_size) == 0154 _q_qty4 = _q_int(clip)155 if _q_qty4 <= 0156 runtime.error("market order quantity must be a positive whole unit")157 _q_pos4 = _q_int(strategy.position_size)158 if _q_pos4 > 0 or _q_pos4 < 0 and _q_qty4 < -_q_pos4159 runtime.error("this order would add to or partially reduce a position, so its trade ledger would not be faithful")160 strategy.order("q2", strategy.long, _q_qty4)161 else162 if _q_int(strategy.position_size) > 0163 _q_qty5 = _q_int(_q_iadd(math.abs(_q_int(strategy.position_size)), clip))164 if _q_qty5 <= 0165 runtime.error("market order quantity must be a positive whole unit")166 _q_pos5 = _q_int(strategy.position_size)167 if _q_pos5 < 0 or _q_pos5 > 0 and _q_qty5 < _q_pos5168 runtime.error("this order would add to or partially reduce a position, so its trade ledger would not be faithful")169 strategy.order("q3", strategy.short, _q_qty5)170 else if _q_int(strategy.position_size) == 0171 _q_qty6 = _q_int(clip)172 if _q_qty6 <= 0173 runtime.error("market order quantity must be a positive whole unit")174 _q_pos6 = _q_int(strategy.position_size)175 if _q_pos6 < 0 or _q_pos6 > 0 and _q_qty6 < _q_pos6176 runtime.error("this order would add to or partially reduce a position, so its trade ledger would not be faithful")177 strategy.order("q4", strategy.short, _q_qty6) How it was tested
The published rules, never tuned on the data, with real costs on every fill. Every strategy here is tested the same way.
- Market data
- CME Globex daily bars, Jan 2, 2016 to Jul 31, 2026 ZN from March 2020
- Rules
- As published, with the published parameters, never tuned on this data
- Size
- 1 contract on every trade
- Contract
- The most traded one, rolled as the market moved
- Fees
- The broker's $0.50 a side, plus CME's and the NFA's fees when the window opens
- Slippage
- 1 tick a side on ES, 2 in NQ's regular session and 3 outside it, 1 on other markets, where it was not measured
- Stops and entries
- Tested on 1-second closes and filled at the next bar
- Contracts traded
- ES: 43 contracts, ESH6 to ESU6ZN: 26 contracts, ZNM0 to ZNU6GC: 54 contracts, GCG6 to GCQ66E: 43 contracts, 6EH6 to 6EU6
- Costs paid
- ES: $155 in fees, $613 in slippageZN: $77 in fees, $578 in slippageGC: $254 in fees, $750 in slippage6E: $206 in fees, $369 in slippage
- Warm-up
- Each new contract first reads the last 253 trading days, moved onto its prices
- Flagged days
- ES: NoneZN: NoneGC: None6E: None Days the data holds too few bars for. Flagged, not removed.
Could you pass
a prop firm challenge
with the time-series momentum rule?
A backtest says whether a strategy makes money. Our AI runs its trades through the rules of every prop firm account, then tells you which one it would pass and at what size.
No card needed. Free credits every month.
- Most money: Blue Guardian Futures Direct 50K, $5,084
- Best payout odds: Funded Futures Family Straight to Funded 50K, 98%
- Fastest pass: Phidias Express to Live 50K, 11 days
$5,084
Median net across 2,000 simulated runs
- Evaluation
- None
- First payout
- 97%
- Fees paid
- $1,521
Median net across 2,000 simulated runs
Questions about the time-series momentum rule
What is the time-series momentum rule?
At each month's end, go long if the price is up over the last 12 months, and short if it is down. The rules follow Tobias Moskowitz, Yao Hua Ooi and Lasse Heje Pedersen, Time Series Momentum, Journal of Financial Economics (2012). Atlantive backtested it on CME futures from January 2016 to July 2026 after costs.
Is the time-series momentum rule profitable on ES futures?
In this backtest it made money: on ES futures from January 2016 to July 2026 it made $158,958 after costs, over 43 trades, positive in 8 of 11 years; on ZN it made $8,923, on GC it made $95,566 and on 6E it lost $17,668. Buying and holding one ES contract over the same window made $236,212. The average trade was $3,791, its 95% interval −$465 to $7,875. These are hypothetical results of a backtest, not of trading.
What are the time-series momentum rule's rules?
At the close of each month's last trading day, the strategy reads the change in price over the last 252 trading days. If it is up, it holds one contract long for the month ahead. If it is down, one contract short. A roll closes the position, and the first daily close on the new contract takes the same side back. The rules follow Tobias Moskowitz, Yao Hua Ooi and Lasse Heje Pedersen, Time Series Momentum, Journal of Financial Economics (2012). Its parameters are the published ones, never tuned on the data it was tested on.
How was the time-series momentum rule backtested?
On CME Globex data from January 2016 to July 2026, one contract, on daily bars, rolled to the most traded contract as the market moved, with the broker's and CME's fees and the slippage measured on each market. Stops and entries were tested on 1-second closes and filled at the next bar. Nothing was fitted to the data.
Can I run the time-series momentum rule in TradingView?
Yes. The page carries its Pine Script, written by Atlantive's compiler from the same code the backtest ran, for TradingView's strategy tester. A chart's fills and fees can differ from the backtest's, so test it again on the symbol you trade.
Can I change the time-series momentum rule and test it again?
Yes. Rerun it in Atlantive: a new chat opens with the strategy's rules in the box, for you to read, change and send. The agent writes the code, backtests it on the market and window you name, and reports the result with its costs.