What is Information Ratio?

In Plain English

The Sharpe ratio measures return per unit of total risk. The information ratio measures something more specific: active return per unit of active risk. The distinction matters for anyone running a strategy against a benchmark. A manager who tracks the S&P 500 closely but consistently outperforms it by a small margin might have a lower Sharpe ratio than a manager who takes huge bets — but the first manager is delivering superior alpha more consistently, which is exactly what institutional allocators care about.

Active return is simply the portfolio's return minus the benchmark's return. If your portfolio returned 12% when the benchmark returned 10%, your active return is 2%. Tracking error is the volatility of that active return series — how consistent is the outperformance? A manager who delivers exactly +2% every month has zero tracking error and an infinite information ratio. One who delivers +2% on average but swings between -10% and +14% month to month has high tracking error and a much lower information ratio.

The information ratio is the active manager's Sharpe ratio. It rewards consistency and penalizes noise. A manager with IR = 0.5 is considered skilled. An IR above 1.0 over a multi-year period suggests genuine, sustainable alpha. Many quantitative funds target specific information ratios when constructing portfolios — combining signals specifically because their combination improves IR even when individual signals have modest standalone IR.

Grinold and Kahn's Fundamental Law of Active Management formalizes this: IR ≈ IC × √N, where IC is the information coefficient (correlation between signal and realized return) and N is the number of independent bets per year. This law explains why diversification of signals and increased bet frequency are both valuable — even with a fixed IC, you can raise IR by taking more independent bets.

Technical Definition

The information ratio over period T:

IR = (r̄_p − r̄_b) / TE

where r̄_p is the mean active return, r̄_b is the mean benchmark return, and TE (tracking error) is the annualized standard deviation of excess returns r_p,t − r_b,t.

Fundamental Law of Active Management (Grinold):

IR = IC × √N (approximate, assumes uncorrelated, equal-IC bets)

where IC = Corr(signal, forward return) and N = number of independent active bets per year.

Transfer coefficient (TC) corrects for portfolio construction constraints that prevent full expression of every signal. IR = IC × TC × √N in the constrained case, where TC ≤ 1.

How VectorFin Uses This

Alternative data signals add value in a portfolio context through IR improvement, not necessarily through high individual IC. VectorFin's sentiment_drift signal has an IC of approximately 0.04-0.07 on a 60-day forward return horizon across the 5,000+ ticker universe. Applied across 1,000 liquid names (N=1,000 annual bets), the theoretical contribution to IR is meaningful even at modest IC.

The whystock_score composite signal combines 8+ sub-signals including Piotroski F-Score, momentum, and sentiment drift. The composite's IC exceeds any individual component because the sub-signals are partially uncorrelated — their errors cancel when combined, which is the quantitative case for data aggregation.

GET https://api.vectorfinancials.com/v1/signals/whystock-score/{ticker}?date=2024-10-01

When evaluating VectorFin signals for your strategy, request the IC data table via the Pro API to see signal-level IC by sector, market cap, and time horizon.

Code Example

import requests
import pandas as pd
import numpy as np

API_BASE = "https://api.vectorfinancials.com"
API_KEY = "vf_your_api_key_here"

def compute_information_ratio(active_returns: pd.Series) -> float:
    """Compute annualized information ratio from a series of active returns."""
    mean_active = active_returns.mean() * 252
    tracking_error = active_returns.std() * np.sqrt(252)
    return mean_active / tracking_error if tracking_error > 0 else 0.0

# Simulate: long top-quintile whystock_score, short bottom-quintile
# vs the equal-weighted benchmark
tickers = ["AAPL", "MSFT", "NVDA", "GOOGL", "META", "AMZN", "TSLA", "JPM"]

scores = {}
for ticker in tickers:
    resp = requests.get(
        f"{API_BASE}/v1/signals/whystock-score/{ticker}",
        params={"date": "2024-10-01"},
        headers={"X-API-Key": API_KEY},
    )
    if resp.ok:
        scores[ticker] = resp.json()["score"]

df = pd.DataFrame.from_dict(scores, orient="index", columns=["score"])
df = df.sort_values("score", ascending=False)

print("Signal ranking:")
print(df.to_string())
print(f"\nTop quintile: {df.head(2).index.tolist()}")
print(f"Bottom quintile: {df.tail(2).index.tolist()}")
print("\nUse forward return data to compute realized IC and information ratio")