Who this is for — Readers who want to navigate trend following, momentum, mean reversion, carry, relative value, market making and event-driven approaches without mistaking a family name for a ready-to-use strategy.
Systematic strategy families group systems that share a mechanism or a form of portfolio construction. They are useful taxonomies, not natural boundaries: one strategy may combine several families, and the same label may describe very different implementations.
The literature documents historical returns associated with some families, but does not guarantee that they will persist, be investable or remain available at every scale. The universe, period, data, signal definition, exposures, costs and selection among tests can change the conclusion. In Cyclepedia, a family explains the economic question; the specification defines the strategy that was actually tested.
An operational taxonomy
| Family | General idea | Typical construction | Risks to make visible |
|---|---|---|---|
| Trend following / time-series momentum | an instrument's past direction contains information about its own future direction | long or short according to its own historical signal | rapid reversals, whipsaw, volatility, leverage, roll and crisis correlation |
| Cross-sectional momentum | relatively strong instruments are compared with weak ones | ranking within a universe, often long-short or long-only | momentum crashes, turnover, factor exposures and universe composition |
| Mean reversion / contrarian | a deviation tends to narrow towards a conditional reference | position against the move or spread | unstable reference, “catching a falling knife”, tails and frequent costs |
| Carry | exposures associated with differences in yield, curve or convenience are held | ranking or portfolio across assets or contracts | crash losses, funding, roll, currency and crowding |
| Relative value / statistical arbitrage | a relationship among instruments makes a deviation relatively unusual | spread, hedge or neutralised portfolio | relationship break, model risk, shorting and joint liquidity |
| Market making / liquidity provision | prices are quoted to capture spread while managing inventory and adverse selection | passive orders, inventory control and cancellations | adverse selection, latency, queues, jumps and inventory risk |
| Event-driven | a defined event changes the distribution or convergence of prices | rules around announcements, mergers, rebalances or corporate actions | event probability, gaps, specialist data and legal or operational constraints |
| Execution / scheduling | a decision is implemented while seeking to reduce execution cost and risk | slicing, routing and a choice between urgency and impact | improper benchmark, non-fill, impact and information leakage |
“Arbitrage” requires particular precision. A riskless theoretical arbitrage and an empirical relative-value trade are not the same thing. Pairs trades, convergence trades and statistical arbitrage can suffer divergence, funding constraints, stock-loan recalls, costs and jumps. Calling them arbitrage does not make convergence certain.
Trend and momentum: two distinct axes
Time-series momentum compares an instrument with its own past and can take a positive or negative sign. Cross-sectional momentum ranks instruments at the same point in time and takes relative exposure to winners and losers. The two signals may agree or disagree: in a broadly falling market, relative “winners” may still have a negative absolute trend.
Trend following is a broader family: moving averages, breakouts and state models can produce trend exposure without being equivalent definitions. Signal speed determines turnover, responsiveness and whipsaw risk; volatility normalisation changes concentration and leverage. There is no universally correct window.
The studies by Jegadeesh and Titman and by Moskowitz, Ooi and Pedersen document two historical forms of momentum under specific samples and designs. They are foundations for studying the phenomenon, not parameters to transfer automatically to different markets, costs and periods.
Mean reversion: towards which reference?
A mean-reverting strategy assumes that a deviation is temporary relative to a relevant mean or relationship. The reference may be the history of the same price, a spread between instruments, a fundamental value, a risk model or a microstructure relationship. If the reference changes, the position may look “increasingly cheap” while accumulating risk.
The following must be distinguished:
- statistical mean reversion, an estimated property of a series or spread;
- cross-sectional contrarian, buying relative losers and selling winners;
- liquidity reversal, recovery after temporary pressure;
- value, price relative to an economic measure over what is often a long horizon.
These concepts can overlap, but they are not synonyms. Stationarity or cointegration tests depend on the model and sample and do not guarantee convergence within the holding period.
Carry: expected return or compensation for risk
“Carry” has different meanings across asset classes. It may refer to interest rate differentials in currencies, curve slope, futures roll yield, bond yield, or the cost or benefit of holding an exposure. Any comparison must state the definition, base currency, hedges, financing, collateral and roll.
Observed carry is not necessarily an almost-certain gain: it may compensate for exposures that lose during liquidity contractions or abrupt changes. The cross-asset study by Koijen and co-authors constructs a comparable definition within its design; a practical implementation must still demonstrate investability, costs and tail risk.
Relative value and market making
Relative value seeks to isolate a relationship rather than the market's broad direction. The hedge is estimated, however; it can change and leave basis, factor, currency or duration risk. The pairs-trading research by Gatev, Goetzmann and Rouwenhorst is a specific historical design: it does not show that every correlated pair converges.
Market making differs from a simple mean-reversion signal. Its result depends on captured spread, queue priority, fill probability, adverse selection, inventory and the ability to cancel or hedge. A backtest on OHLC bars rarely observes all these states; automatically assigning a fill at the bid or ask can turn nonexistent priority into simulated profit.
Execution algorithms, finally, are not necessarily an alpha family. They may seek to minimise implementation shortfall or execution risk relative to an execution benchmark for a position that has already been decided.
How to compare two strategies
| Dimension | Questions to ask |
|---|---|
| Mechanism | what behaviour, risk or friction should support the result? |
| Horizon | how long do the signal, position and label information last? |
| Exposures | do beta, duration, volatility, currency, credit or liquidity explain the P&L? |
| Turnover | how much gross performance is consumed by spread, impact and financing? |
| Capacity | how do fills and costs change as quantity grows? |
| Dependence | do trades overlap or concentrate in the same episodes? |
| Tail | where do gaps, short squeezes, deleveraging or relationship breaks appear? |
| Diversification | do systems have different names but the same latent exposure? |
Average return correlation is only a starting point. Conditional dependence and simultaneous losses in relevant scenarios may reveal hidden concentration across systems.
Illustrative classification example
A portfolio buys futures with a positive signal on their own past returns, sells those with a negative signal and scales positions using a volatility estimate. It may be classified as time-series momentum or trend following. The specification must still define contracts, roll, frequency, window, volatility estimate, leverage limit, gap handling and costs. Changing one of those elements may alter risk more than the family label does.
This example illustrates the taxonomy; it is neither an operational recommendation nor a claim of returns.
Sources
- Narasimhan Jegadeesh and Sheridan Titman, Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency — foundational study of cross-sectional momentum in the sample examined.
- Tobias J. Moskowitz, Yao Hua Ooi and Lasse Heje Pedersen, Time Series Momentum — evidence and construction of time-series momentum in the cross-asset design studied.
- Ralph S. J. Koijen et al., Carry — definition and comparison of carry across asset classes within the study sample.
- Evan Gatev, William N. Goetzmann and K. Geert Rouwenhorst, Pairs Trading: Performance of a Relative-Value Arbitrage Rule — methodology and historical evidence for one pairs-trading rule.
- Werner F. M. De Bondt and Richard Thaler, Does the Stock Market Overreact? — classic study of overreaction and contrarian returns in the sample considered.