45%35%20%
Quant baseline · no RAG · no agents
One signal.
Three portfolios.
Your original three-asset model starts at 45% SPY, 35% IEF, and 20% LQD, then adjusts the mix each month. Compare it with fixed 60/40 and S&P 500 portfolios.
Test design
What is learned, shown, and sealed
Development results are visible. January 2023 onward remains reserved and its realized performance is hidden.
Interactive signal
How a release becomes a surprise
The denominator is calculated from earlier forecast misses only, so today’s release cannot change its own scale.
Release missActual − forecast = 0
↓
Standardized surprise0.00σIn line with expectations
Below
Above
Development evidence
Three scenarios on the same months
All returns use adjusted SPY, IEF, and LQD prices and the same 5 bp per dollar traded cost rule.
Original three-asset model
Start at 45 / 35 / 20
The model adjusts the original starting allocation each month.
Starting allocation
Annual return—
Volatility—
Sharpe—
Max drawdown—
Growth of $100
Common scored overlap · Oct 2020–Dec 2022
Original model60/40S&P 500
| Scenario | Annual return | Volatility | Sharpe | Max drawdown | Cumulative |
|---|
What the comparison tells us
The original model reduced fluctuations, but earned less
Across these 27 development decisions, the S&P 500 portfolio delivered the highest return. The original three-asset model had the lowest volatility, but its lower return outweighed that reduction in risk.
100% SPY earned 11.9% annualized and 28.7% cumulatively. It also had the highest volatility (19.5%) and the deepest peak-to-trough loss (−23.9%).
The fixed mix earned 4.3% annualized and 9.9% cumulatively, with lower volatility than SPY alone. Its −20.5% drawdown was the smallest of the three.
The model earned 1.1% annualized and 2.4% cumulatively. Volatility was lowest at 12.5%, but its −21.1% drawdown was slightly worse than 60/40. Its Sharpe of 0.08 was also the lowest.
What this means for the MVP: The model did not outperform either fixed portfolio in this development sample. It reduced volatility, but the results do not yet show that its forecasts and allocation changes added value.
How to read the metrics
Annual return is the compounded growth rate expressed per year. Cumulative return is the total gain across the sample. Volatility measures how much monthly returns fluctuate, expressed annually. Max drawdown is the largest observed fall from a previous portfolio peak. Sharpe compares return with volatility; higher is better. Here it uses annualized return divided by annualized volatility and assumes a zero cash rate.
This development sample is short. Results describe this period; they do not establish future outperformance.
Decision replay
Open any model decision
Historical test is sealed.Signals and model weights are shown for demonstration. Realized returns remain hidden.
Macro state
Growth—
Inflation—
Ridge → Black–Litterman
Σ = how assets move together. Ω = how uncertain each forecast is. More uncertainty means less weight on the Ridge view.
Model allocation this month
Starting allocation: 45% SPY / 35% IEF / 20% LQD. These bars show the model’s adjusted weights for the selected month.
Subsequent outcome
The original model, explained
How the starting mix becomes a decision
The original three assets stay the same. Their weights change as the model combines economic news, forecasts, and risk.
45% SPY holds U.S. equities. 35% IEF holds 7–10 year U.S. Treasuries. 20% LQD holds investment-grade corporate bonds. This neutral mix anchors Black–Litterman.
Payrolls, retail sales, industrial production, and ISM form the Growth surprise. CPI, Core CPI, PPI, and hourly earnings form the Inflation surprise. Each group averages its available standardized surprises.
Ridge estimates next-month returns for SPY, IEF, and LQD using earlier data. Shrinkage limits extreme forecasts. Policy expectations, financial conditions (ANFCI), and geopolitical risk (GPR) add context.
Black–Litterman combines the starting mix’s implied returns with Ridge forecasts. The optimizer then balances expected return, risk, and turnover within allocation limits.
Black–Litterman after Ridge
Starting mix + model views + uncertainty → moderated returns
The neutral return estimate implied by the 45 / 35 / 20 starting mix and asset covariance.
The three Ridge return forecasts. The replay above shows each raw view alongside its BL result.
Capital sigma: how asset returns move together, estimated from up to 36 earlier months.
Capital omega: each view’s earlier forecast-error variance. Higher uncertainty reduces the view’s influence.
Allocation limits: SPY 20–70%, IEF 15–60%, LQD 10–40%. Weights sum to 100%. The 60/40 and SPY-only reference portfolios keep their fixed allocations.
The previous detailed page is saved.
You can still open the original presentation, including its longer Black–Litterman walkthrough and signal charts. It uses the earlier dataset and results.
Open previous MVP page →Data and limits
What is in the updated MVP
Eight Bloomberg release and consensus series from the Oct 26 workbook. Surprise scaling uses prior observations only.
Chicago Fed ANFCI and Caldara–Iacoviello GPR. Both use latest revised history, not a locked real-time vintage.
SPY, IEF, and LQD adjusted-close returns. ETF history in the current source begins in October 2016.
The workbook’s 3-month and 6-month implied policy paths begin in September 2019. Earlier missing policy fields use past-only median imputation inside Ridge.