Strategy family 05 / 12
Trend & momentum
Trend and momentum strategies rest on a simple premise: prices that have moved in one direction tend to keep moving that way for a while. Trend following and time-series momentum, the core of trend CTA programs, go long or short each market according to its own past move, while cross-sectional momentum and relative strength rank assets against one another and favor recent leaders over laggards. Technical and price-action trading works from the chart itself, often at shorter horizons, reading breakouts, ranges and reversal patterns directly from price and volume.
Covers
- Trend following
- Time-series momentum
- Cross-sectional momentum
- Relative strength
- Trend CTA
- Technical & price-action trading
Why it is a demanding test
The core signal in trend following fits in one line of code; nearly all of the difficulty lies around it — sizing each market by its volatility, choosing how fast to react, paying for turnover and futures rolls, and holding through long stretches of whipsaw. Price history is also among the most heavily mined data in finance, so it is easy to find a lookback that worked and hard to show it was not luck. A system that reads charts can describe a pattern fluently whether or not it is there, which makes this family a direct test of telling signal from noise.
What we test
What a system needs to do well in this family. Tasks are built so that each ability can be observed and scored on its own.
Scaling exposure to volatility
Size each position so that markets with very different volatility contribute comparable risk, and adjust as volatility changes, rather than allocating equal notional amounts.
Choosing a signal's speed
Set lookback horizons, smoothing and exit rules in advance, and explain the trade-off between catching new trends early and paying for more false signals and turnover.
Ranking relative strength without hidden bets
Build cross-sectional rankings that control for sector, country or style exposure, so a momentum portfolio is not simply a concentrated bet on last year's leading industry.
Testing patterns against random entries
Define a technical setup precisely enough to test, then compare its results with random entries held for the same period in the same markets.
Anticipating sharp momentum reversals
Recognize the conditions in which momentum has historically been prone to violent reversals, such as sharp rebounds after deep market declines, and size or hedge accordingly.
Example task types
Illustrative designs that show the shape of the work. They are not items from the benchmark itself.
- Track: Simulation
Run a multi-market trend program in replay
The system manages a diversified futures portfolio across equity indices, government bonds, currencies and commodities in a point-in-time replay, with contract specifications and roll calendars. It must state its rules in advance, then set daily target positions under an illustrative 10% annual volatility target. Scoring looks at net risk-adjusted results, how risk was spread across sectors, how rolls were handled, and whether the rules were followed through choppy, trendless stretches.
- Track: Analysis
Test whether a breakout rule beats chance
The system receives a precisely specified setup, for example a close above a 20-day range on above-average volume, and point-in-time price and volume data for a broad stock universe that includes later delistings. It must run an event study comparing returns after each signal, net of costs, with those of matched random entries, and state whether the evidence supports an edge. A strong answer accounts for how many variants were tried and is willing to conclude that there is nothing there.
- Track: Live
Hold a live cross-sectional momentum portfolio
Each month the system ranks a universe of country and sector equity indices on past performance, for example a 12-month lookback that skips the latest month, and submits a paper long–short portfolio with an expected return spread and a forecast range. Scoring runs over rolling windows on data that did not exist when the system was trained. A strong submission keeps its method stable, limits concentration in any one region, and reports uncertainty honestly.
- Track: Client
Explain a trend fund's long drawdown
A user holding a trend-following fund that has lost money for more than a year asks whether something is broken and whether to sell. The system must explain how trend strategies tend to behave in range-bound markets, ask what role the fund was meant to play in the user's portfolio, and separate a strategy-wide dry spell from a problem specific to the manager. A good response neither promises a recovery nor dismisses the concern, and is clear about what cannot be known in advance.
How it is scored
Each run is scored on its outcome and on the process behind it, so a lucky result is not mistaken for a sound one.
- 01
Improvement over a simple trend baseline
Risk-adjusted return compared with a transparent baseline, such as a simple moving-average rule on the same markets, so that added complexity has to earn its place.
- 02
Edge retained after trading costs
Net return as a share of gross, and return per unit of turnover, showing how much of a signal survives commissions, spreads, slippage and roll costs.
- 03
Drawdown depth, duration and stress behavior
Maximum drawdown, time to recover, and how the strategy behaved during equity market stress, when trend programs are often expected to diversify.
- 04
Stability across nearby parameter choices
Performance across neighboring lookbacks, rebalance days and market sets, plus a deflated Sharpe ratio that accounts for how many variants were tried. A result that holds at only one exact setting is treated as fragile.
- 05
Discipline through losing streaks
Whether the system follows its stated rules through runs of false signals, and whether any override is explained and recorded before the outcome is known.
Diagnostics
Failure modes we diagnose
Recurring mistakes the tasks are built to surface. Each one is reported separately, so builders can see which to fix first.
- 01
Cherry-picking the winning lookback
Trying many lookbacks or filters and reporting the best, without accounting for the search, turns noise into an apparent edge.
- 02
Filling at the signal's own close
Computing a signal from a closing price and assuming the trade filled at that same close. It is a small look-ahead that flatters fast signals most.
- 03
Seeing patterns in random paths
Naming a head-and-shoulders, a flag or a breakout where random price paths would produce the same shape just as often.
- 04
Ignoring momentum crash exposure
Missing that a momentum book has drifted into one sector, or is short a basket of high-beta losers after a sell-off, which leaves it exposed to a violent reversal when the market rebounds.
- 05
Mishandling futures rolls
Computing signals or profits on naively spliced futures prices, which creates false jumps at each roll and misstates both.
Environment and data
Trend tasks use daily and intraday prices for futures, currencies and equities across many markets, with back-adjusted series for signals, actual contract prices and roll dates for profit and loss, and point-in-time stock universes that include later delistings. Orders fill no earlier than the next available price and pay commissions, plus spreads and slippage that widen in fast markets and thinner contracts. Replays include long range-bound stretches and abrupt reversals as well as sustained trends, so results do not rest on one kind of market.