Config as it was
A vendor may be able to tell you what its data said in March 2024. Nobody outside your firm can tell you what your USD curve config was. Masseef records every version.
As of any date
Run any day again, exactly as your desk saw it.
Every value and every config version is kept with two dates: when it applied, and when you learned it. Corrections never overwrite. So the whole object graph, from a swap down to its underlyings, can be read as it was on any date, and tonight's break and last year's backtest both run on the data and config of the day.
BOOK · USD RATES RV
Valued as of 14 Jun 2024
USD 10Y swap, 14 Jun close
3.904%3.888%
Vendor A
USD curve config
v3
approved 31 May
18 Jun, 09:15. Vendor A restates the 14 Jun close to 3.888%.
Original run14 Jun, 17:30
12,480,210
Re-run as known on 14 Junsame data, same config
12,480,210
Matches the original run
Re-run as known todayincludes the correction
12,493,875
+13,665 · cause: input restated 18 Jun by Vendor A
The question behind every break
What did this look like on that date, and why?
Every fund and bank answers it many times a year. Few have a system that does. The cost is already being paid, spread across lines nobody adds up.
- 01
Reconstruction by hand
An analyst restores an archive or a backup to answer one question.
- 02
Snapshot estates
Dated extracts, the jobs that produce them, and the rules for keeping them.
- 03
Break explanation
People working out every night whether an input was wrong then or has been corrected since.
- 04
Lineage diagrams
Tools that document how data flows. Useful, but a diagram can't reproduce a number.
- 05
Audit preparation
Days assembling evidence, then the auditor's hours testing it.
- 06
Research preparation
Weeks per study getting the history right before the research can start.
As-of reconstruction
Ask for any date, as it was known on any date.
Drag the slider. The chart shows the 10-year swap rate as it was known on that day, and the config versions known at the time. The valuation on the right follows.
USD 10Y swap, as known
Value before the correction
Curve config versions
BOOK VALUED AS OF 14 JUN 2024
- 14 Jun close used
- 3.904%
- Source
- Vendor A, received 14 Jun 17:02
- Curve config in force
- USD v3
- Book value
- 12,480,210
What the desk saw on the day.
- Tonight's break. Compare what was known then with what is known now, and the difference names its cause.
- The auditor's question. Show the figure, and show what it was built from.
Technically: bitemporal, for reference data, time series, configuration, trades and positions.
Research
Backtests that don't flatter you.
Most backtests apply today's curve setup, today's corrected data and today's identifiers to the past. The bias runs one way: they look better than reality did. Run on what was known at the time, your backtests will look worse, and that is the point.
Data as it was known
Values before later corrections, with each correction kept as its own fact and dated when it arrived.
Identifiers as they were
Ticker and identifier changes resolve by date. A 2020 backtest finds FB, a 2023 one finds META, and both are the same company.
Running research with agents? Point-in-time correctness matters more when nobody inspects each test. Why
Bring one question you can't answer today.
For example: what was your USD curve config on 15 June 2024? We'll show you how Masseef answers questions like that, on a running system.