swaabi-lab/fx-swap-points-japan

Daily swap point dataset from Japanese FX brokers

20,882 rows · 63 currency pairs · 10 companies · 2026-05-16 to 2026-08-14

Swap points (rollover interest) published by Japanese retail FX companies, collected every business day and normalised so that rows can be compared: amounts in JPY per 10,000 units of the base currency, each paired with the number of days it covers.

Licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) · Version 1.0 · Generated 2026-08-14

Why this exists

Each company publishes its own swap points on its own page, in its own units, and overwrites them daily — the previous day is simply gone. This table keeps that history in a form where two companies can be placed side by side, which is what makes a question as basic as how has the gap between these two moved this month answerable at all. Published under a licence that lets you use it.

Two things to know before comparing rows

1. Divide by the accrual days first. buy_jpy_per_10k is the published figure, and how many days it covers varies (three days at once around a weekend is normal). Comparing raw figures across dates or across companies will make a swap that rose look like it fell by two thirds. Pre-divided columns are provided.

2. 3,267 rows are days with no accrual. grant_days = 0, amounts 0, per-day columns empty. Companies differ on which days those are, and that disagreement is itself data, so the rows are kept. Averages change depending on whether you include them — filter to grant_days > 0 to reproduce the figures published on the site.

Columns

Column	Meaning
date	Trading day the swap applies to (YYYY-MM-DD).
pair	Currency pair, base/quote. The amounts below are per 10,000 units of the base (left-hand) currency.
broker	Name of the FX company, as published by that company.
product	Account course. 通常 (standard) or LIGHT — some companies run two courses with different terms.
buy_jpy_per_10k	Swap on a long position, in JPY per 10,000 units, exactly as published. Positive means received, negative means paid. NOT divided by the number of accrual days. Pairs quoted in a currency other than JPY are converted to JPY.
sell_jpy_per_10k	Swap on a short position. Same units and sign convention.
grant_days	How many days the amount covers. Values spanning a weekend or a holiday pay several days at once. 0 means the company observed the day but granted nothing for it.
buy_jpy_per_day	buy_jpy_per_10k divided by grant_days, rounded to 4 decimals. Use this to compare across days or across companies. Empty where grant_days is 0.
sell_jpy_per_day	sell_jpy_per_10k divided by grant_days, rounded to 4 decimals. Empty where grant_days is 0.

What is deliberately not in here

Removals are stated rather than made silently — a dataset that quietly drops a company is worse than one that was never published, because the omission becomes the reader's problem to discover.

• 1 individual observation judged unreliable by a daily automated check (a series jumping against its own neighbours, or an order of magnitude away from other companies on the same pair). They are named with company and figure on the site's per-pair pages, not hidden.
• 0 rows where the amount and the accrual count contradict each other (an amount recorded against zero days), which cannot be reduced to a per-day figure.
• Days with no observation at all. Absent rather than zero — companies publish on different schedules, so not every company exists on every day.

How it is collected

Directly from each company's own published page or file, every business day, by program. Not scraped from an aggregator and not copied from another site. Each day's collection passes an automated validation gate (date continuity, sign consistency, cross-company outliers) before it is published; anything that fails the gate is not shipped.

How far the comparison is underwritten

Companies publish in different shapes: whether the date column means the trade date or the value date, how many units the amount refers to, whether it is denominated in yen or in the quote currency, how many days it covers, and which column is the buy side. Putting them on one axis therefore rests on 50 assumptions (10 companies x 5). 48 of them (96%) are checked against the company's own published material or a basis outside this site; 2 are corroborated only by agreement with other companies — which is not a guarantee, because a mistake shared by every company would pass that test. 0 remain unchecked. The machine-readable breakdown by kind is in assumption_verification in the JSON distribution.

Method in detail: https://swaabi-lab.com/methodology/

Updates

The files in this repository are a snapshot. The live distribution at https://swaabi-lab.com/data/ is rebuilt four times a day (04:00, 10:00, 16:00, 22:00 Japan time); a company's change appears in the first rebuild after it publishes. Per-company last-observed dates are in as_of_by_broker inside the JSON distribution.

These figures are a record, not a forecast. Swap points are revised by each company every business day, a pay side can turn into a receive side, and nothing here implies the levels will hold. A copy of this file taken today describes today.

Licence and attribution

CC BY 4.0 — commercial use, modification and redistribution are all permitted; attribution is the only condition.

Source: Swaabi Lab (https://swaabi-lab.com/data/) / CC BY 4.0

What the licence covers is this dataset — the collection, the unit normalisation and the arrangement. It is not a claim over the figures each company publishes; their own sites' terms are their own. When quoting figures, state the observation date: swap points change daily. If you modify the data before redistributing it, say so.

Files

• swap-daily.csv — UTF-8; the first line is a # source comment, skip it as a comment when reading
• swap-daily.json — same rows plus units, licence and the exclusion counts above
• Always-current versions: https://swaabi-lab.com/data/

import pandas as pd
df = pd.read_csv("swap-daily.csv", comment="#")
# or the always-current copy:
df = pd.read_csv("https://swaabi-lab.com/data/swap-daily.csv", comment="#")

Notes

Provided without warranty; collection or conversion errors may remain. Always confirm current terms with the company itself before acting on anything here. This dataset and the site publishing it are not investment advice and make no recommendation about any company or trade.

Corrections and questions: https://swaabi-lab.com/contact/ (a public correction log is kept at https://swaabi-lab.com/corrections/)