AI BACKTEST / PORTFOLIO VARIANT ENGINE

Backtest a portfolio from a plain-English idea.

Describe your assets, weights, or goal. AI builds the rules, creates one-variable alternatives, and explains the results.

See the live result →How results stay verifiable ↓
AI BACKTEST / DRAWDOWN CONTROLREADY TO STRUCTURE
INVESTMENT IDEA

“I want a long-term core portfolio with less severe drawdowns, without giving up too much return.”

AI is converting the idea into testable assumptions…
01 / REFERENCE60% VTI · 30% BND · 10% GLDAnnual rebalance · dividends reinvested
02 / AI VARIANTS
Gold +10%Equity −10%Quarterly rebalance
One controlled change in each portfolio
03 / AI INTERPRETATION

More gold improved the worst drawdown, but the result is sensitive to the selected start year.

Next suggested test: shift the start date

LIVE DATA VERTICAL SLICE

One portfolio. Real closing prices. Auditable output.

A $10,000 reference portfolio using dividend-adjusted daily closes, with annual rebalancing and no fees or additional contributions.

REFERENCE PORTFOLIO60% VTI · 30% BND · 10% GLD
MARKETSTACK EOD SNAPSHOT
ENDING VALUE$17,829
ANNUALIZED RETURN9.8%
MAX DRAWDOWN-22.0%
ANNUALIZED VOL.11.3%
60 / 30 / 10 portfolio100% VTI benchmark
$25,722$21,299$16,875$12,451$8,028Jun 2, 2020Jul 31, 2026

METHODCommon trading dates · adjusted close · annual rebalance ·1537 observations

PROVENANCEMarketstack `/v1/eod` · fetched Aug 20, 2026, 2:34 PM UTC · 16 incomplete provider rows excluded

THE AI BACKTEST WORKFLOW

AI does more than translate a prompt into portfolio code.

It structures the question, designs controlled comparisons, and explains the evidence. The calculation layer remains deterministic and reproducible.

01INTERPRET

Turn an idea into explicit rules.

Describe a goal in ordinary language or enter assets, weights, and rules directly. AI converts the idea into a testable reference portfolio.

02INSPECT

Expose assumptions before testing.

Dividends, inflation, contributions, costs, rebalancing, and data substitutions are surfaced before they affect the result.

03DESIGN

Propose informative variables.

AI reasons about which allocation, asset, timing, or rule changes could answer the user’s question most directly.

04GENERATE

Create controlled variants.

Each proposed portfolio changes one variable, preserving a clean comparison with the reference portfolio.

05EXPLAIN

Interpret the measured difference.

AI connects changes in return, drawdown, volatility, and result consistency to the periods and assets that contributed to them.

06CONTINUE

Recommend the next useful test.

Instead of searching thousands of parameters, AI proposes the next comparison with the highest information value.

A CLEAR DIVISION OF LABOR

AI designs and explains the comparison. The engine computes it.

PortfolioCompare never asks a language model to invent prices, calculate returns, or decide whether an investment is suitable. AI organizes and interprets the result around auditable calculations.

AI ASSISTANCE
  • Structures natural-language ideas
  • Identifies hidden assumptions
  • Designs one-variable comparisons
  • Explains differences and limitations
  • Suggests the next useful experiment
BACKTEST ENGINE
  • Loads versioned market data
  • Applies weights and portfolio rules
  • Calculates performance and risk
  • Runs identical rules across variants
  • Produces reproducible result fingerprints

AI-GUIDED EXPERIMENT DESIGN

Generate the next useful comparison—not more noise.

AI evaluates the current evidence and proposes the variable most likely to clarify the decision. Every branch includes a rationale before it is tested.

BBASE PORTFOLIO
01Gold 10% → 20%Test drawdown protection
02Start year 1999 → 2003AI recommendation · test sensitivity
03Annual → quarterlyTest implementation rule

BUILD A CONTROLLED BACKTEST

Start with an idea. End with a testable portfolio question.

Create a reference portfolio, generate controlled variants, and see which configuration is better supported by the backtest.

Start with an investment idea