AI BACKTEST / PORTFOLIO VARIANT ENGINE
Portfolio Backtesting: Backtest Portfolios, Compare Results, One Variable at a Time
Describe your assets, weights, or goal. AI builds the rules, creates one-variable alternatives, and explains the results.
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…Gold +10%Equity −10%Quarterly rebalance
One controlled change in each portfolioMore gold improved the worst drawdown, but the result is sensitive to the selected start year.
Next suggested test: shift the start datePORTFOLIO BACKTESTING BASICS
What Is Portfolio Backtesting?
Portfolio backtesting is the process of applying a portfolio’s assets, weights, contribution rules, rebalancing schedule, and date range to past market data. The goal is not to prove that an investment will work in the future. The goal is to see how a clear portfolio decision behaved under real market conditions before you put money behind it.
PortfolioBacktest focuses on controlled comparisons: keep the base portfolio fixed, change one variable, and measure what changed in return, drawdown, volatility, and consistency. AI helps structure plain-English ideas into testable assumptions, proposes useful variants, and explains the measured trade-off. For the technical approach behind that workflow, read the AI backtesting guide.
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.
60 / 30 / 10 portfolio100% VTI benchmark
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
PORTFOLIO BACKTESTING GUIDE
How to use a portfolio backtest without fooling yourself.
How to Backtest a Portfolio
Start with a complete base portfolio, not a single ticker. Enter the assets, target weights, initial value, rebalancing rule, and date range. A useful backtest should make those assumptions visible before showing performance. If a portfolio contains SPY, BND, GLD, QQQ, individual stocks, or another supported asset, each symbol should be validated against market data before the result is run.
After the base portfolio is defined, run the calculation and review the result as a decision record: what was tested, which data window was used, and which rules produced the outcome. That makes the backtest easier to repeat, audit, and share.
Why One Variable at a Time
Most portfolio tools let users compare two or three portfolios, but the comparison often mixes several changes at once. That makes the result hard to interpret. If one portfolio changes the equity weight, adds gold, changes the start year, and uses a different rebalance schedule, the result may look better, but you cannot tell which decision created the improvement.
Controlled portfolio comparison solves that problem. Change one allocation, asset, contribution rule, start date, or rebalance rule at a time. The result becomes easier to read: what improved, what worsened, when the difference appeared, and whether the same conclusion survived a different market period.
Key Metrics to Compare
A portfolio backtest should not be judged by final value alone. Return matters, but so do maximum drawdown, annualized volatility, recovery time, and consistency across different start dates. A portfolio that finishes slightly higher but suffers a much deeper drawdown may not fit the same investor objective as a portfolio with steadier behavior.
PortfolioBacktest presents results as trade-offs. For example, a bond-heavy variant may reduce drawdown while lowering long-term return. A technology-heavy variant may increase ending value while concentrating risk. The useful question is not simply “which line is higher?” It is “which change created the result, and is that trade-off acceptable?”
Free vs Paid Portfolio Backtesting
Free portfolio backtesting is enough for occasional comparisons: define a portfolio, generate controlled variants, use AI to explain the difference, and share a branded result. This is designed for users who need a clear answer to a single allocation question or want to test a portfolio idea before doing deeper work.
Paid workspace features are for repeated research and long-term monitoring. They focus on saved comparisons, version history, unlimited active tests, cleaner exports, and automated monitoring when new market data changes a saved conclusion. The same engine powers the interface, API-style workflows, and future MCP or Skill access, so the paid layer is about persistence, automation, and professional reuse rather than a different calculation method.
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.
PortfolioBacktest 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.
- Structures natural-language ideas
- Identifies hidden assumptions
- Designs one-variable comparisons
- Explains differences and limitations
- Suggests the next useful experiment
- Loads versioned market data
- Applies weights and portfolio rules
- Calculates performance and risk
- Runs identical rules across variants
- Produces reproducible result fingerprints
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