Specific (Idiosyncratic) Return

Specific return — also called idiosyncratic, residual, or alpha — is the portion of a strategy's return that a named factor model cannot explain. In FM103's Factor Attribution sub-pill it is the residual after subtracting the per-factor weighted contributions. This article explains what specific return means, when to celebrate it, and when to worry about it.

Definition

For each period, the strategy's realised return r_p is decomposed by the Brinson framework (see Brinson Attribution):

where w_i is the strategy's exposure to factor i and r_i is the realised return of factor i (top quintile minus bottom quintile) over the period.

What it actually contains

Specific return absorbs everything the named factors don't capture. That includes:

• Genuine stock selection. If your strategy picks superior names within each factor bucket, the picking shows up as specific return.
• Unnamed factor exposure. Sector tilt, country bias, currency mismatch, or any latent risk factor the model omits.
• Implementation luck. Random period-specific outcomes that don't represent a repeatable pattern.
• Measurement error. Factor returns are themselves estimated — small misestimation flows into the residual.

How much specific return is "good"?

When large specific return is a red flag, not skill

The default assumption when specific return exceeds 50% should be "there is a factor I am not naming." The most common culprits:

• Sector concentration. If your value screen happens to load on financials and the period was a financials-rebound period, "specific return" is really "financials sector return." A sector factor would absorb it.
• Country bias. US-only strategies hide US-specific macro tailwinds in specific return.
• Size bias. The named factors are typically size-neutralised; if your strategy is implicitly small-cap, the small-cap premium goes to specific return.
• Quality omission. Strategies using only value + momentum often pick up unintended quality exposure; add quality to the factor mix to absorb it.

The fix is to add candidate factors to the model and re-run attribution. If specific return drops materially, you've found the missing factor.

When large specific return is genuine skill

If after adding all reasonable candidate factors specific return is still large and the per-period sign is consistent (rather than driven by 1–2 outsized periods), the strategy may have genuine stock-selection skill. Selection skill is rare and valuable. To validate it:

1. Hit rate of specific return. Fraction of periods where it's positive. 65%+ is suggestive of real skill; 50% with one big period is luck.
2. t-statistic. Mean specific return divided by per-period standard error, scaled by sqrt(N). A t-stat > 2.5 over 20+ periods is meaningful.
3. Out-of-sample. Specific return is the most overfit-prone part of a backtest. Treat any large in-sample specific return with suspicion until out-of-sample data confirms it.

Linking to the Risk Decomposition sub-pill

FM103's Risk Decomposition sub-pill reports R² from a ridge regression of portfolio returns onto factor returns. The relationship:

• R² > 0.6 + small specific return: Consistent. Factors explain both the level and the variance.
• R² < 0.3 + large specific return: Consistent. Factors explain little.
• R² > 0.6 but large specific return level: Factors track the wiggles but miss the mean. Probably a sector or country level effect.
• R² < 0.3 but small specific return: Factors track the mean but miss the wiggles. Unusual; investigate noise.