What Is a Factor? Comparing Security Characteristics with Numbers
Learn what a factor measures, how scores differ from rankings and trading orders, and why factor evaluation does not establish investment profitability. Public KabuForge examples keep features separate from future-return labels.
What a factor means
In equity research, a factor is a numerical representation of a security characteristic used for comparison. For example, you can calculate past price changes for each security using the same method. Instead of saying a stock ‘looks strong,’ you specify what was compared.
Factor investing uses such characteristics to inform security selection or allocation. Measuring a characteristic and establishing whether it helps explain future returns are separate questions. Understand the measure first, then evaluate it.
A score describes what was measured and how
Price momentum is one example: it measures past price changes over a specified period and compares securities. Changing that period changes the values, even for the same security.
A score needs context: the calculation window, price definition and reference date. A score transformed into a rank is also different from the underlying price return. A rank-based score of 0.8 does not mean an 80% return.
Scores, rankings, portfolios and orders are different
The following teaching example uses fictional codes and invented figures. These are neither real securities nor KabuForge run results. The ranking rule places larger past price changes first.
| Fictional code | Past price change | Rank |
|---|---|---|
| EX-A | +8% | 1 |
| EX-B | +3% | 2 |
| EX-C | −2% | 3 |
The table compares past price changes measured consistently. It does not establish that EX-A will rise the most in the next period. Future declines also remain unknown until those outcomes are observed.
After ranking securities, you still need to specify which to hold, how much to allocate and when to review the holdings. Those are portfolio rules. Orders require further decisions about quantities and timing based on existing holdings and cash.
A factor score describes a characteristic; a rank compares securities; a portfolio specifies intended holdings; an order instructs a trade. A high score alone does not determine an order quantity or price.
The public course stores features separately from future returns
The second KabuForge public research course uses 20-session and 60-session price momentum to demonstrate feature calculation and separate evaluation. Follow its definitions of observation periods and prices, rather than changing them after seeing results.
Features use closing prices through the session before the signal date, or D-1. The evaluation label is the simple return from the execution open to the open of the fifth subsequent observed session. It provides the later outcome used to examine the relationship between the characteristic and subsequent price changes.
Future-return labels are stored in a separate file and used only for evaluation. Rows near the end without sufficient future prices remain missing. Do not feed labels or evaluation results into security selection on the same date: that introduces information unavailable at the decision time. See data availability and look-ahead bias.
What to examine when evaluating a factor
One question is whether securities with higher scores subsequently had higher returns. IC measures correlation between factor values and future returns; Rank IC measures correlation between their ranks. See factor and IC basics.
Quantile analysis groups securities by score and compares subsequent returns across groups. This can reveal whether outcomes change progressively across the ranking, rather than looking only at the top group.
IC and quantile returns are different from portfolio profits after costs. Holding rules, turnover, execution assumptions and fees all affect portfolio results. Correlation alone cannot establish profitability without examining transaction costs and turnover.
A three-security example cannot establish statistical stability
The public course uses a small example to teach calculations and interface interpretation. Three securities allow at most three quantile groups, and one security can strongly affect the result. Good results in this example do not establish a stable relationship across a broad universe.
Days with constant values or many missing observations contain limited information for comparison. Breaking ties by security code does not create new predictive information. Overlapping future-return windows also mean daily evaluations are not all independent observations.
Repeatedly changing a factor’s window or direction after viewing results makes it easier to fit the historical period. Fixing the definition first and checking periods not used in development relates to overfitting and out-of-sample evaluation.
Read features and evaluation separately in KabuForge
In the factor diagnostics tutorial, first check the inputs and calculation method, and confirm that feature files and evaluation labels are separate. Then read IC, quantile returns, coverage—the proportion of securities for which the calculation is available—and changes in group membership.
Before drawing conclusions from a chart, explain what the score measures, which information it uses and which subsequent return period it is compared with. These checks help turn a comparison of security characteristics into research that can be inspected.
Source: the second course in KabuForge’s GUI research courses. The fictional-code table was created solely for explanation.
This article is educational material about research procedures. It does not recommend particular securities or trades or guarantee future returns. See the disclaimer.
Continue this topic
Applies to: main source candidate (unpublished) · Public source docs/en_US/GUI_RESEARCH_COURSES.md
Reviewed · KabuForge