Japanese Equity Backtests with J-Quants: Input Design

Define what information drives a decision, when that decision occurs and when execution follows. A close-based feature does not establish that execution at the same close was possible.

In a fictional next-open study, separate feature construction, quantity scheduling and execution price. Write that sequence before selecting or transforming input columns.

A present-day universe may omit securities that disappeared during historical testing. State whether the experiment uses date-specific membership or a deliberately fixed teaching sample.

Requested dates and usable result dates can differ because of warm-up and unavailable future labels. Record effective boundaries, exclusions and retained sample counts.

Features must use information allowed at the decision time. Future returns belong to evaluation, with their horizon fixed before analysis.

A fictional five-observation label is unavailable at the tail when five later observations do not exist. Count valid features separately from evaluable feature-label pairs instead of repairing the tail with another horizon.

Fees, dividends, rounding, cash constraints and price basis can change the account path even with identical signals. Specify them alongside the input rather than after viewing returns.

Skipping an unaffordable fictional opening order and resizing it are different strategies. Align benchmark dates and return definitions, and distinguish fills from completed trades.

The public KabuForge course validates and freezes selected local daily bars, then performs preflight and research. It separates prior-observation decisions from next-observation opening fills and records skipped orders.

The course is not certification for every Japanese security or arbitrary API response. Check the receipt for your own input and retain its stated historical availability limitation.

Continue with Adjusted Prices and Stock Splits, Factor Analysis: Inputs and Evaluation, Price Data Fields and Observation Dates 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, JPX: J-Quants datasets 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.

Explore the j-quants learning path

• J-Quants Adjusted Prices and Stock Splits
• J-Quants and Point-in-Time Research
• Factor Analysis with J-Quants Data: Inputs and Evaluation