Exposure Effects Are Not Automatically Regime Alpha

Name the performance claim precisely

A strategy can lose less during a decline because it held less of the declining asset. That observation does not establish skill in identifying the decline. Separate absolute return, active return against a benchmark, and alpha under a stated risk model. A regime label may organize an exposure policy without adding predictive information. Begin with the narrow question: did the rule achieve something beyond a simpler allocation with comparable constraints? This is an evaluation question, not a recommendation to increase, reduce, or time market exposure.

Reconcile holdings before fitting explanations

For a single period, reconcile portfolio return to beginning-of-period holdings and the corresponding asset returns, then account for costs and any intra-period trades. Track invested weight, cash, borrowing, gross exposure, and net exposure separately. A low net position can still contain substantial offsetting exposures. When allocations vary, preserve the time series of actual simulated holdings rather than replacing it with an average weight. The return of an average-weight portfolio need not equal the return of the original changing allocation.

Use alpha as a model-dependent residual

A factor regression relates portfolio excess returns to selected factor returns; its intercept is an estimated alpha conditional on that specification. Fama and French's original factor research illustrates the distinction between common exposures and residual performance. The factor set, currency, cash reference, frequency, estimation window, and treatment of dependence all matter. Omitting a relevant exposure can put its contribution into the intercept. A fitted positive intercept is not direct proof of forecasting skill, and a statistically uncertain intercept should not be described as established alpha.

Regime-specific regressions divide the available sample while adding estimation choices. Report how many observations and distinct episodes support each fit. Do not treat an extreme coefficient from a sparsely occupied state as a stable exposure estimate. A full-sample attribution may explain a completed history while remaining unavailable for decisions made within that history.

Build useful counterfactual allocations

Compare the regime policy with a fixed allocation, a cash-inclusive policy benchmark, and a simple rule that does not use the disputed regime information. Choose these comparisons before inspecting their relative outcomes. A retrospectively exposure-matched benchmark can help explain results, but it is a diagnostic rather than automatically an implementable alternative. Keep its hindsight adjustments explicit. The published definitions of market, size, and value factors also remind the evaluator that an exposure measure has a construction rule, not merely a familiar name.

Read a fictional mechanical advantage

Consider an invented portfolio holding 40% in one market and 60% in zero-return cash, with no trades or costs during each scenario. When the market gains 8%, the portfolio gains 3.2% and trails the fully invested comparison by 4.8 percentage points. When the market loses 8%, it loses 3.2% and leads by 4.8 percentage points. The two rows are alternative one-period scenarios, not a sequential backtest. No classifier is involved, so the downside advantage cannot be evidence of regime-detection skill.

Separate timing from the exposure level

To study timing, ask whether the changes in exposure contributed beyond a predeclared constant-exposure comparator, after costs and under the same risk constraints. Align every allocation with information available before its outcome. Rebuilding labels with revised data or smoothed states changes the question to retrospective association. Inspect distinct episodes rather than only the average of selected state returns. If a timing result disappears when a single episode is removed, report that concentration instead of treating the number of daily observations as a substitute for repeated evidence.

Report attribution without converting it into proof

Present the holdings reconciliation, benchmark-relative result, factor specification, residual uncertainty, turnover, and cash treatment together. Gross and net comparisons must use compatible cost scopes. Explain where attribution does not uniquely identify a cause: leverage changes, volatility scaling, and changing asset mix can overlap in their effects. A useful conclusion might identify lower participation as the main explanation while leaving timing skill unresolved. These are methods for examining hypothetical research claims, not investment advice, a forecast, or evidence that a particular regime allocation has earned alpha.

Inspect the comparison

Primary references

References support the stated methods and data conventions. Original fictional examples are not observations from the cited sources.

• Fama and French: Common Risk Factors in the Returns on Stocks and Bonds
• Kenneth R. French: Description of Fama/French Factors

Continue the investigation