The formula answers a precise question: given that an analyst likes a stock, how much alpha should the portfolio expect from it?
$$z_i = \frac{Rating_i - \overline{Rating}}{\sigma_{Rating}}$$
Raw ratings are meaningless on their own — a "7 out of 10" tells you nothing unless you know the analyst's average and spread. The Z-Score — definition">z-score converts the raw rating into "how many standard deviations above the analyst's own average is this call?" [1]
$$\alpha_i = IC \cdot z_i \cdot \sigma_{\epsilon_i}$$
Three inputs multiply together: [1]
| Term | What it is | What it does |
|---|---|---|
| $IC$ | Information Coefficient — Correlation — definition">correlation between the analyst's past scores and realised returns | Scales the whole signal. Zero skill → zero alpha, regardless of the rating |
| $z_i$ | Standardised rating from Step 1 | Captures the strength of the current call |
| $\sigma_{\epsilon_i}$ | Residual (idiosyncratic) Volatility — definition">volatility of stock $i$ | Captures the room for that call to manifest in returns |
Say an analyst rates three stocks: A = 8, B = 5, C = 3.
Mean rating = 5.33, Standard Deviation (Volatility) — definition">standard deviation = 2.05.
| Stock | Rating | z-score | $\sigma_\epsilon$ | $IC$ | $\alpha_i$ |
|---|---|---|---|---|---|
| A | 8 | +1.30 | 28% (mid-cap) | 0.05 | +1.82% |
| B | 5 | −0.16 | 12% (large-cap) | 0.05 | −0.10% |
| C | 3 | −1.14 | 28% (mid-cap) | 0.05 | −1.60% |
Stock A is a mid-cap with a strong positive signal → meaningful positive alpha forecast.
Stock B is a large-cap with a near-neutral signal → barely any alpha, and the 12% residual vol further compresses it. [1]
Skill is the multiplier. If $IC = 0$, then $\alpha_i = 0$ for every stock, no matter how confident the analyst sounds. Before trusting any active strategy, ask: what is this manager's historical IC? [1]
Small-cap and mid-cap offer more alpha opportunity. Residual volatility for large-caps is ~10–15%; for small-caps it is ~25–40%. The same IC and the same conviction produce structurally larger alpha forecasts in smaller stocks — which is one reason active management has a harder time adding value in Nifty 50 stocks. [3]
Standardisation removes bias. Without z-scoring, a permabull analyst who rates everything 8/10 would artificially inflate every alpha forecast. The z-score anchors on relative opinion, not absolute level. [1]
The Grinold Formula — definition">Grinold formula produces an alpha Vector — definition">vector — one $\alpha_i$ per stock. That vector feeds into the Fundamental Law of Active Management ($IR \approx IC \cdot \sqrt{B}$), where $IC$ from this formula becomes the manager's edge-per-bet, and $B$ is how many such bets are made. [2]
Apply this → Go to the Finding Alpha module to see how the alpha vector from Grinold feeds into portfolio construction and the Fundamental Law.