Deterministic baseline · no RAG · no agents

From an economic release
to a portfolio weight.

A point-in-time quantitative baseline that converts macro surprises into Growth and Inflation states, forecasts next-month asset returns, and produces constrained allocations.

Selected decision September 30, 2026 8 / 8 releases observed

Growth surprise −0.13σ Modestly below consensus

Inflation surprise +0.17σ Modestly above consensus

Equity47%

Treasury31%

IG Credit22%

System architecture

One traceable path from data to decision

Every box produces a structured output that can be audited before it moves downstream.

Input

Macro releases

First-release actuals and pre-release consensus for eight indicators.

744 observations

Transform

Surprise engine

Actual − consensus, scaled only by prior forecast-error history.

Look-ahead safe

State

Macro factors

Growth, Inflation, Policy, ANFCI, GPR, and prespecified interactions.

Monthly panel

Forecast

Ridge models

Separate expanding-window next-month forecasts for each asset.

Past-only tuning

Allocate

Black–Litterman

Blends model views with a market-implied prior and forecast uncertainty.

Long only

Output

Portfolio weights

Bounded, turnover-aware allocations with modeled transaction costs.

Month-specific weights

Interactive demo

See exactly how a surprise becomes a signal

Select a release from the active decision month or edit the inputs. The formula recalculates immediately.

Release

Actual Consensus

Prior surprise volatility (σ)

The scale uses only earlier releases. This prevents the current observation from changing its own z-score.

Raw surprise162 − 55 = 107

Standardized surprise107 ÷ 1075.7 = +0.10σ

A small positive Growth surprise.

September 2026

What entered the model

Component z-scores roll into separate Growth and Inflation composites.

Composite states

Growth−0.13σ

Inflation+0.17σ

Reading the state

Industrial Production and ISM outweighed positive Payrolls and Retail Sales. Core CPI created the only material positive Inflation surprise.

Macro surprise states over time

Monthly equal-weight composites, standardized using prior data only

GrowthInflation

Model output

Forecast, allocation, and evidence

The baseline is useful, but the current test does not yet establish superior risk-adjusted performance.

1 · Ridge forecastsSep 30 decision

Expanding-window Ridge uses only data available through this decision month.

2 · Portfolio bridge

Black–Litterman moderates the Ridge views

It combines the benchmark-implied starting point with Ridge forecasts, while discounting views that have been unreliable.

01Neutral anchor

Benchmark prior

The 45 / 35 / 20 benchmark and trailing covariance imply a risk-aware starting return.

45% Equity35% Treasury20% Credit

02Model evidence

Ridge views

The selected month’s next-month forecasts become Black–Litterman’s absolute views.

03Risk context

Risk covariance (Σ) and view uncertainty (Ω)

Σ (capital sigma) is the trailing 36-month covariance matrix of asset returns. Ω (capital omega) is a diagonal matrix of each Ridge view’s prior forecast-error variance.

Σ = asset co-movement · Ω = forecast uncertainty

04Blended result

Posterior returns

The moderated return estimates feed the long-only, turnover-aware optimizer.

RMSE means root mean squared error, the typical size of earlier forecast misses. Larger errors create a larger Ω and therefore less trust in the Ridge view.

Selected decision

Compare raw and moderated return views

Raw Ridge views are the model’s standalone forecasts. Select the Black–Litterman result to see how the prior and forecast uncertainty moderate them.

3 · Recommended allocationLong only

100%allocated

Full backtest evidenceContext—not an input

4 · Subsequent outcomeNext month

Outcome data is evaluation evidence—not an input available at the selected decision date.

Out-of-sample portfolio growth

After modeled transaction costs, against fixed 45 / 35 / 20 benchmark

MVPBenchmark

Annual return10.03%Benchmark 9.46%

Sharpe ratio1.33Benchmark 1.34

Max drawdown−3.65%Benchmark −3.56%

Average turnover5.35%Monthly one-way

Latest-month walkthrough

What happened in September 2026?

Releases arrived

Eight scheduled macro releases were observed by the September month-end decision date.

Surprises offset

Soft Industrial Production and ISM pushed Growth slightly negative. Core CPI pushed Inflation slightly positive.

Forecasts stayed restrained

Ridge forecast positive Equity returns and approximately flat-to-negative fixed-income returns, with strong shrinkage.

Allocation tilted modestly

Black–Litterman moved the benchmark to 47% Equity, 31% Treasury, and 22% IG Credit.

Guardrails

What this MVP proves—and what it does not

Point-in-time controls

Macro z-scores use prior observations only. GPR is delayed to the following decision month. ANFCI uses an assumed five-day publication lag.

Current evidence

The MVP beat the benchmark by 0.57 percentage points annualized, but delivered slightly worse Sharpe and drawdown over only 33 realized months.

Remaining limitation

ANFCI and GPR use latest revised histories. A formal locked test should reconstruct real-time vintages before making a performance claim.

From an economic releaseto a portfolio weight.