Trend following is a family of rules
A trend-following strategy changes exposure in response to a defined direction measure. That measure might use an instrument's trailing return, a moving average or a breakout. The label alone does not specify the product, holding period, portfolio or risk model.
Time-series momentum compares an instrument with its own past returns. Cross-sectional momentum ranks instruments relative to one another. A single-chart CFD rule is not interchangeable with a diversified futures portfolio, even if both are described as trend following.
Start by naming the implementation. Will it trade one instrument or a universe? Long only or long and short? At completed daily bars or intraday? What happens when the signal changes? Put those answers into the strategy template before comparing results.
What published research supports—and what it does not
Moskowitz, Ooi and Pedersen's Time Series Momentum studies an own-return effect across futures and forwards. Hurst, Ooi and Pedersen's long-horizon historical study examines a specific constructed implementation over a much longer record.
Those papers provide evidence about their instruments, signals, assumptions and samples. They do not establish a win rate for a new three-bar rule, prove that every trend strategy profits in a crisis or determine a suitable allocation for the reader. Historical reconstruction and a live executable strategy are different objects.
When reading any result, identify the return series, volatility sizing, rebalancing, transaction costs and market universe. Then ask which of those conditions your own proposed implementation actually shares. A familiar label is not enough.
A small rule that can be inspected by hand
For this teaching exercise only, use one fictional cash instrument, one unit and a long-or-flat position. At each completed session close, compare that close with the arithmetic mean of the previous three closes, excluding the current close. No action is available during the three-observation warm-up.
| Decision | Exact exercise rule |
|---|---|
| Entry signal | While flat, close above the previous-three-close mean creates a buy for the next session open. |
| Exit signal | While holding, close below that mean creates a sell for the next session open. |
| Equal values | Keep the existing position; no new signal. |
| Execution | Use the next open shown in the invented series; no same-close fill. |
| Exposure | One unit, no shorting, no pyramiding and no leverage in the arithmetic exercise. |
| Costs | USD 1 assumed total round-trip charge; no other costs in this simplified example. |
The example loses money without breaking its rules
| Session | Open | Close | Previous 3 mean | Opening fill | Decision after close |
|---|---|---|---|---|---|
| D1 | 100.00 | 100.00 | — | None | Warm-up |
| D2 | 101.00 | 102.00 | — | None | Warm-up |
| D3 | 103.00 | 105.00 | — | None | Warm-up |
| D4 | 105.00 | 104.00 | 102.33 | None | Buy next open |
| D5 | 105.00 | 107.00 | 103.67 | Buy at 105.00 | Hold |
| D6 | 106.00 | 103.00 | 105.33 | None | Exit next open |
| D7 | 101.00 | 99.00 | 104.67 | Sell at 101.00 | Stay flat |
At D4, the previous closes are 100, 102 and 105: their mean is 102.33. The close at 104 creates a buy instruction, filled at the invented D5 open of 105.00. D5 remains above its reference mean. At D6, the mean of 105, 104 and 107 is 105.33; the close at 103 creates an exit instruction.
The sale occurs at the D7 open of 101.00, not at D6's close. The one-unit gross result is USD -4.00. After the assumed USD 1.00 charge, the net result is USD -5.00. The gap between signal and fill is part of the example, not an execution error to remove.
This demonstrates signal timing, a reversal and a losing trade. It is not a historical backtest or an estimate of expectancy. One winning example would have proved no more. Download the invented observations to reproduce the calculation:
session,open,close,previous_three_close_mean,opening_fill,closing_decision D1,100.00,100.00,,None,Warm-up D2,101.00,102.00,,None,Warm-up D3,103.00,105.00,,None,Warm-up D4,105.00,104.00,102.3333,None,Buy next open D5,105.00,107.00,103.6667,Buy at 105.00,Hold D6,106.00,103.00,105.3333,None,Exit next open D7,101.00,99.00,104.6667,Sell at 101.00,Stay flat
Where a real implementation can depart from the worksheet
Persistent direction can allow a rule to stay with a move; repeated reversals can trigger late entries and costly exits. Neither outcome is guaranteed. Lookback length changes responsiveness, turnover and the opportunities represented in the sample.
| Issue | What to define before evaluation |
|---|---|
| Data adjustments | Corporate actions for securities; contract rolls and consistent construction for futures. |
| Execution | Attainable prices, spread, slippage, rejected orders and latency. |
| Short exposure | Borrow or contract access, financing, margin and liquidation rules, if shorts are part of the system. |
| Portfolio risk | Instrument sizing, correlated exposure, leverage and concentration. |
| Failure state | Action on a stale quote, missing bar, disconnect or breached account limit. |
Daily trend research does not automatically transfer to a five-minute chart. A CFD mid-price feed is also not a futures execution record. Use the risk-system guide to specify the control around the rule, and the calculator to check contract-dependent sizing separately.
Test the rule you wrote, not the chart you prefer
Freeze the universe, signal, entry, exit and cost model. Record every variant tried. Reserve data that did not influence selection and preserve trades crossing a split under a declared rule. Inspect turnover, drawdown, duration, exposure and net outcomes—not only the most favorable year.
Compare against a relevant baseline with comparable exposure and costs. A leveraged, short-capable futures strategy cannot be evaluated fairly by placing its headline return beside an unleveraged cash holding without explaining the difference.
Then run incoming simulation using the unchanged specification and log intended versus actual orders. The validation guide explains the evidence stages. Read the drawdown guide before interpreting a past maximum decline as a future loss ceiling.
Questions worth answering
What is the difference between time-series and cross-sectional momentum?
Time-series momentum evaluates an instrument against its own history. Cross-sectional momentum compares instruments with one another. The portfolio, signal and risk rules still determine the strategy actually being tested.
Does a moving average crossover prove that a trend will continue?
No. It is a transformation of historical prices. It can define a repeatable decision rule but cannot guarantee the next move. Delayed signals, repeated reversals and costs must be represented in the test.
Does trend following protect a portfolio in every crash?
No. Historical findings for particular diversified implementations do not guarantee protection in a new event. Signal delay, sudden reversals, instrument access, leverage and execution can change the outcome.
What percentage of my portfolio should use trend following?
There is no allocation percentage established by this guide. A portfolio decision requires the actual strategy, risks, correlations, costs and personal constraints. Do not transplant a percentage from a different study or portfolio.
Sources and scope
- Moskowitz, Ooi and Pedersen: Time Series Momentum
Primary research on own-return momentum in the studied futures and forwards; not evidence for the invented three-close example.
- Hurst, Ooi and Pedersen: A Century of Evidence on Trend-Following Investing
Historical reconstruction of a particular implementation. Its scope and assumptions matter when interpreting the result.
- Bailey et al.: The Probability of Backtest Overfitting
Selection risk when many configurations are tried on finite data.
The seven observations and fills are invented for this article. No live performance, market-size estimate, universal allocation or guaranteed crisis protection is claimed.