MacroForge Macro
Macro relationship
intelligence.
Which macro signals actually move an asset — and when? MacroForge discovers those relationships and pressure-tests them out-of-sample, reads the regime they live in, and carries them through to the exposures of a real equity book. Honest evidence, first.
Discover
Find the relationship, then try to break it.
Relationship scan
Screen the macro universe for the signals that genuinely lead an asset — tested on data they never saw, and filtered for the false positives that come from searching thousands of series.
Relationship shapes
See the real, non-linear shape of a relationship — where it strengthens, where it saturates, where it flips — not just a single average slope.
Event studies
How an asset typically behaves after a signal fires — the turning points, and the path that tends to follow.
Read the state
Know which world you are in.
Regime inference
States inferred from the data rather than assumed — with the probability of each one, the dates the model thinks the world changed, how long each episode ran, and the historical episodes today most resembles.
Nowcasts
An estimate of a series for a period that has not been published yet — with a standard error and an interval, so the uncertainty arrives with the number instead of after it.
Transmission & causality
Which market moves first and which follows: conditional causal structure with lags and orientation, and shock-transmission networks showing who exports volatility to whom.
Act on it
From a relationship to a position.
Macro equity model
Top-down and bottom-up macro models on a single-stock universe: every security’s exposure to each macro factor, with confidence around it — so you can see what a book is really long of.
Correlated macro stress
Move a macro factor and the others move with it, the way they actually co-move. Correlated ±1σ shocks, not one-at-a-time bumps that never happen in isolation.
Tail risk
Where the distribution stops behaving — the macro states in which the left tail thickens, measured rather than assumed.
Uncertainty, stated
A number is not a forecast without its error.
A nowcast arrives with the range the model considers plausible around it. Here the point estimate sits below 50 — but the interval crosses it, and that is the finding. Reporting 49.1 alone would claim a contraction the data does not support.
- 95% interval
- 45.6 … 52.6
- Standard error
- 1.79
- Convergence
- Converged
Estimated from a dynamic factor model on point-in-time inputs — each one the value that was known at the time, not today’s revised number.
Notebooks · coming
Build on it, in the browser.
The hosted notebook platform that ships with Fabrisk is coming to MacroForge. Write Python against the same point-in-time data and the same models the screens use — without deploying anything locally.
- Ask your own question
- The screens answer the questions we anticipated. When yours is not one of them, write it yourself — against the same models and the same history the product runs on, not an export of them.
- Open it and start
- Nothing to install, nothing to download, no environment to build. It is already running, with the data already there.
- Yours, and your desk’s
- A private space for work in progress and a shared one for the team — so research a colleague can open and re-run does not end up as a file on somebody’s laptop.
The data
Every study runs on what was known at the time.
A curated, revision-aware macro warehouse. Each input is the value as first published, not today’s restated figure — so a backtest cannot quietly use a number that did not exist yet, and a revision is a fact you can study rather than a contamination you never notice.
The standard
Predictive is not the same as tradeable.
Every relationship is scored on data it never saw — and penalised for the thousands of combinations we tried — before it earns a rating. The in-sample view sits beside it, clearly labelled, never in its place.