Factor Analysis with J-Quants Data: Inputs and Evaluation
Factor analysis turns security characteristics into values and evaluates their relationship with later returns.
Define the feature first
Factor analysis turns security characteristics into values and evaluates their relationship with later returns. Freeze formula, direction, lookback and universe before examining outcomes.
A fictional twenty-observation momentum feature needs an explicit price basis and cutoff. Leave warm-up values missing unless a separate, documented definition justifies another treatment.
Keep labels out of strategy inputs
A future-return label is an evaluation answer. Its start price, end price and horizon need a fixed definition separate from feature construction.
A fictional five-observation horizon leaves missing tail labels. Report original eligibility and missing reasons, and never feed quantile outcomes or future returns back into the strategy.
Read correlation with grouped results
IC summarizes association with later labels, while quantiles show outcomes for ranked groups. Both require inspection of dates, surviving samples, missingness and cost conventions.
A fictional three-security ranking supports limited conclusions. One strong daily correlation is insufficient evidence of general predictive ability; examine sample counts and stability across periods.
Record exploration and dependence
Trying many windows and directions creates selection risk. Preserve attempted conditions and distinguish selection periods from later diagnostics.
Overlapping forward labels are not automatically independent observations. Describe trial counts, sample size and date coverage instead of treating many daily statistics as calibrated inference.
Use the documented diagnostics
KabuForge Demo 2 separates price-derived features and evaluation labels, then inspects IC, Rank IC, quantiles and coverage. It does not establish arbitrary financial factor import or automatic factor selection.
Check input identity and invalid-date reasons. Keep evaluation artifacts out of feature inputs and retain the course's research-only interpretation.
Further reading and official references
Continue with Listed Issues and Market Segments, What Is J-Quants? Research Uses and Limits, Designing Inputs for Japanese Equity Backtests to connect input definitions, availability and evaluation. All numerical and company examples in this article are fictional teaching examples. For real input, prioritize the documentation for your version and your own acquisition and execution records.
Check the official KabuForge GUI research course for documented workflow scope, and J-Quants: official V2 equities implementation, J-Quants: official V2 financial implementation for provider documentation or current official implementation. Confirm exact units and access limits against the applicable official specification and contract. General research design guidance does not establish that every proposed check is implemented in the GUI.
This article is educational and does not recommend securities or trades, or guarantee returns. Historical research in the public course does not fully establish that only historically available information was used. Read the disclaimer.
Continue this topic
Applies to: main source candidate (unpublished) · Public source docs/en_US/GUI_RESEARCH_COURSES.md
Reviewed · KabuForge