Evaluating Equity Factors with Information Coefficient and Sharpe

Summary

The document raises a practical factor-validation question using a stock-ranking signal assigned scores from one to ten and evaluated against five-day returns over a three-year sample. It reports observations, trading days, average returns, and return standard deviations for the lowest and highest score groups, then asks how to assess predictive effectiveness.

The author distinguishes a sample-size-based t-statistic from an annualized Sharpe calculation, proposing to scale the latter by the square root of the number of five-day periods in a trading year. The material is framed as a question and does not include an answer, an information coefficient calculation, or a backtest of the signal. Its figures alone do not settle statistical significance or investment value: overlapping holding-period returns and dependence among observations can affect uncertainty estimates, while a portfolio Sharpe requires a defined return series and treatment of costs and exposures.

Key ideas

• The example evaluates a ranked equity factor against five-day forward returns across score groups.
• The question separates inference about a mean return from annualized risk-adjusted strategy performance.
• The proposed Sharpe scaling uses the square root of the number of five-day periods per trading year.
• The document does not provide an information coefficient result or a resolution to the validation question.
• Overlapping horizons and correlated observations can make simple sample-size-based significance calculations unreliable.

Tags

# Measuring Information Coefficient and Sharpe Ratio


# Measuring Information Coefficient and Sharpe Ratio












I have been looking through / using Quantopians' Alphalens library to measure/create new factors, and I had some questions in evaluating the credibility of the factor.

This is what I have:

- I have created a factor that ranks stocks from ranking 1 to 10.

- Each factor is supposed to have a predictive power for 5-day total return for groups of stocks in each score.

These are the statistics that I have gathered using above score's return over 3-year period.

- factor with score = 1

- number of data points = 12,659

- number of trading days: 850

- average return over 5 days: 0.0026

- standard deviation of return over 5 days: 0.06

- factor with score = 10

- number of data points = 11,397

- number of trading days: 850

- average return over 5 days: -0.01

- standard deviation of return over 5 days: 0.058

From the above, I am not sure what is the correct way of measuring the effectiveness of the factor. I know that t-statistics can be calculated using:

```
sqrt[number_of_samples] * (average return over horizon) / (sample standard deviation over horizon)
```

However, from the strategy's sharpe-ration perspective, below are used:

```
sqrt[252 trading days / 5 - because we are talking about 5 day return?] * (average return over 5 days) / (sample standard deviation over 5 days).
```

Is this correct way of evaluating the signal in above example?

To summarize, my question is the confusion coming in the process of identifying / validating the effectiveness of the factor that I have constructed.

I appreciate your time and help in advance.

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.