Charta is on Google Play: practise on the go. Get the app
Charta

Weigh up why alphas are refined during implementation and the methods used to do so.

FRM® Part II · Risk Management and Investment Management · Portfolio Construction · FRM2.IM.4.b

The idea

Raw alphas are rarely fit to feed straight into an optimizer. They carry noise, overconfidence and unintended biases, and an optimizer will turn a large or erroneous alpha into a large position. Refining them has three common steps. Scaling sets the size of alphas to what the information can support. Here IC is the information coefficient, and the score is a standardized signal with a cross-sectional standard deviation of one. The rule of thumb is below. Alpha = volatility × IC × score. Trimming caps extreme values, for example at three standard deviations, so no single forecast dominates. Neutralizing removes unintended bets. Remember that with no costs or constraints, doubling every alpha doubles the active positions and the active risk.

Formulas

Scaled alpha

α = ω × IC × score

ω
the asset's residual volatility
IC
the information coefficient
score
the standardised signal, mean zero and cross-sectional standard deviation one

Volatility multiplies, not divides, and IC enters once, not squared. Trim the result if it exceeds the cap, often three standard deviations of the alphas.

A practice question

From the question bank, with the reason each wrong option is wrong.

Why are raw alphas usually refined before being passed to an optimizer?

  1. Refining guarantees that the realized alpha of the portfolio will be positive in the next measurement period.Why not: Refinement improves consistency with the information; it cannot guarantee positive realized alpha.
  2. Optimizers cannot accept alphas with negative values, so every alpha must be shifted upward before it is used.Why not: Optimizers handle negative alphas without trouble; negative alphas simply lead to short or underweight positions.
  3. Refining makes the sum of all alphas equal the portfolio's target active risk.Why not: Alphas are returns and the target active risk is in different units; the two are not set equal.
  4. They are noisy and often overstated, and an optimizer would turn the flaws into extreme or unintended positions. - correct

The optimizer treats each alpha as true. Scaling, trimming and neutralizing bring raw forecasts in line with what the information can support and remove unintended bets.

A harder one

Alphas have been scaled so that their cross-sectional standard deviation is 1.5%. The firm trims alphas at three standard deviations. A stock's scaled alpha is 7.2%. What alpha goes into the optimizer?

  1. 4.5% - correct
  2. 7.2%Why not: Leaves the alpha untrimmed; the cap is 3 x 1.5% = 4.5%.
  3. 0.0%Why not: Treats trimming as removing the outlier; trimming caps an extreme alpha but keeps its sign and size up to the cap.
  4. 1.5%Why not: Caps at one standard deviation rather than three standard deviations.

The cap is 3 x 1.5% = 4.5%. An alpha above the cap is set to the cap, so 7.2% becomes 4.5%.

The hardest tier

Two stocks have the same score of +2.0. The IC is 0.05 and residual volatilities are 15% for stock X and 35% for stock Y. The refined alphas for X and Y are closest to:

  1. 10.0% and 10.0%Why not: Multiplies IC by score and omits residual volatility, so both stocks get the same alpha.
  2. 3.5% and 1.5%Why not: Assigns the larger alpha to the less volatile stock; alpha rises with residual volatility.
  3. 1.5% and 3.5% - correct
  4. 0.2% and 1.2%Why not: Uses variance instead of volatility (0.15² and 0.35²).

Alpha = volatility x IC x score: X = 0.15 x 0.05 x 2 = 1.5%; Y = 0.35 x 0.05 x 2 = 3.5%. The same standardized signal implies a bigger move in a more volatile stock.

In the app: 6 practice questions on this outcome, spaced review that brings it back before you forget it.

Start free →

More in Portfolio Construction