Diversification Without Guarantees

Decide what is being diversified

There are several distinct diversification questions. Market diversification concerns underlying exposures. Method diversification concerns how decisions respond to the same information. Horizon diversification concerns the speed of those responses. Operational diversification concerns dependencies such as data, intermediaries and settlement processes. A portfolio can improve one dimension while remaining concentrated in another. Counting products alone cannot establish that all these dimensions have been addressed.

For example, several differently named rules might all use the same trailing price history and produce the same directional state. Their labels add variety to a report, but their current behaviour may be nearly identical. Conversely, two genuinely different decision processes can still depend on one data feed. A useful review therefore asks both what changes the exposure and what could prevent the process from functioning.

Use a transparent portfolio calculation

Consider two hypothetical sleeves with annual volatilities of 15% and 10%, weighted 60% and 40%. At zero correlation, variance is 0.6 squared times 0.15 squared plus 0.4 squared times 0.10 squared, or 0.0097. The square root is approximately 9.8489%. This is a deliberately specified covariance example, not an estimate for any real investment or a suggested allocation.

At an assumed correlation of 0.5, the cross term is 0.0036, bringing variance to 0.0133 and volatility to 11.5326%. At correlation one, volatility is 13%. MIT's portfolio-theory teaching provides the underlying mean-variance framework. The changing answer illustrates a conditional benefit: diversification depends on the joint behaviour of the sleeves, and that dependence must remain visible when presenting a single summary number.

Do not confuse an asset class with a decision rule

Managed-futures discussions often refer to strategies that can change direction across multiple markets. That description differs from simply holding a passive collection of commodity prices. A portfolio's behaviour depends on how exposures are selected and resized, not just on the names of the underlying contracts. Even where the same markets appear in two portfolios, their directional and timing rules can create different paths.

Moskowitz, Ooi and Pedersen's Time Series Momentum examines a particular historical construction across several asset classes. Its multi-market design is useful primary context for this distinction. It does not make every publication labelled trend following equivalent to that study. A standalone educational comparison of deterministic directions is narrower still: without contract execution and investment accounting, it does not establish the diversification of investable returns.

Measure rule overlap directly

A simple diagnostic asks how often two fully defined rules occupy the same eligible directional state. Suppose two hypothetical sequences agree on 72 of 90 comparable observations. Their agreement rate is 80%. If nine additional dates are unavailable for one rule, those dates should not silently become disagreements or neutral states. Report the common eligible count and the unavailable count beside the percentage.

Agreement is not return correlation. Two rules can disagree mainly during small movements or differ greatly in size despite sharing direction. Still, state overlap is useful when the research question concerns method behaviour rather than investment performance. Examine whether differences occur around reversals, during warmup or because of tie conventions. This tells a more precise story than claiming that separate parameter names create independent strategies.

Consider horizons and common stresses

A faster rule and a slower rule can react differently to a change in price direction. That difference may reduce some forms of timing concentration, but both can still be affected by repeated reversals. Similarly, multiple market exposures can depend on shared financing or currency conditions. These are possible dependencies to test with specified scenarios, not universal claims about how markets must behave.

A hypothetical scenario matrix might include a persistent movement, an abrupt jump, a rapid reversal and missing observations. Record each rule's eligible states and adjustment requirements under the same assumptions. A method that looks different during a smooth path may converge with others during a discontinuity. Diversification analysis becomes more informative when it includes conditions under which the apparent differences disappear.

Include the cost of maintaining differences

Combining methods can create offsetting instructions. In a fictional example, one method requests plus three units and another requests minus two units of the same defined exposure. Their net request is plus one, while the gross requested magnitude is five. Whether execution can be netted depends on the account and implementation design. A calculation should not assume both full gross costs and perfect netting simultaneously.

Rebalancing also maintains the intended mixture only by changing positions. A method that contributes little independent behaviour may still add turnover or operational complexity. The relevant comparison is therefore broader than a correlation table. Ask what distinct behaviour is obtained, what resources it requires and which assumptions permit offsets. Complexity alone is neither evidence of diversification nor a reason to dismiss a genuinely useful difference.

Write a conditional conclusion

A defensible diversification statement identifies the exact objects compared, the common observation period and the measures used. It distinguishes historical covariance from hypothetical stress behaviour and operational dependencies. It also identifies limitations: short samples, missing states or unavailable contract detail. Readers can then see where the conclusion is supported and where it remains an assumption rather than an established property.

Avoid turning a conditional reduction in estimated variance into a promise of protection. A diversified portfolio can still decline, experience a long recovery or require simultaneous cash payments. The useful outcome is an improved map of dependencies and a clearer account of the trade-offs. That map supports better questions about a portfolio without pretending to guarantee what any future market episode will deliver.

Source record

1. MIT OpenCourseWare: Portfolio Theory
2. Moskowitz, Ooi and Pedersen: Time Series Momentum, NYU-hosted paper