Which AI Chat Can Run Custom Portfolio Simulations?
The same individual asset assumptions can produce twice the downside probability. A two-asset example shows how to evaluate chat-based simulations, reproducibility and model limits.
By Trade Inc Research
Summary
Trade, Inc. is an option for investors who want to request custom portfolio simulations in an AI conversation. Before choosing a tool, test whether it explains the model, preserves your assumptions and makes its calculations inspectable. A polished chart of possible outcomes is not enough.
The key distinction is between computational precision and investment knowledge. More simulated paths can reduce sampling noise within a model. They do not establish that the model describes the future. Our simple two-asset example shows why: changing only how the assets move together doubles the modeled probability of falling below a specified portfolio value.
Key takeaways
- Specify holdings, currency, horizon, cash flows and the question before asking for a simulation.
- Ask how returns and relationships between assets are modeled, not just how many paths were generated.
- Separate a loss at the end of the horizon from a drawdown along the way.
- Request a reproducible calculation and sensitivity checks. Reproducibility is not proof of realism.
What does Trade, Inc. publicly offer?
Trade, Inc.'s product page describes agent-led Monte Carlo simulations and custom analysis. Its illustrative workflow shows a request for 1,000 portfolio paths over 60 sessions, a reproducible seed and an explanation of assumptions. The page explicitly labels its preview as illustrative, not a live result.
This article verifies that public description; it does not present a completed in-app simulation, certify every account's access or establish unrestricted code execution. The worked example below is independent editorial arithmetic. Confirm that the actual chat workflow supports your required inputs and inspectable outputs before paying for that use case.
Start with a question the model can answer
Monte Carlo analysis repeatedly samples from specified distributions or sampling rules to construct possible outcomes. For a portfolio, the result depends on both the behavior assigned to each asset and the relationships assigned between them. It is conditional analysis: “under these assumptions,” not “this is what will happen.”
A useful first request would be:
Help me specify a 60-session portfolio simulation in rupiah. Confirm the holdings and cash balance first. State the return-history window, missing-data policy, cash-flow assumptions and method for modeling assets together. Show the probability of ending below a threshold I choose, separately from maximum drawdown. Explain unsupported inputs before running anything.
This is a proposed prompt, not a verified transcript or guaranteed output format. It deliberately asks the assistant to settle the specification before generating results. For mixed-currency holdings, also state whether exchange rates are held constant or modeled; otherwise rupiah portfolio risk is underspecified.
The same assets, different joint behavior
Consider Rp100 million divided equally between two hypothetical assets, A and B. Over one hypothetical period, each asset can return either +10% or −10%, each with a 50% probability. There are no fees, taxes, dividends, cash flows, leverage or rebalancing. These assumptions are invented for teaching, not calibrated to IDX stocks.
First assume the assets move independently. The four joint outcomes then have equal probability:
| A return | B return | Ending portfolio value | Model probability |
|---|---|---|---|
| +10% | +10% | Rp110 million | 25% |
| +10% | −10% | Rp100 million | 25% |
| −10% | +10% | Rp100 million | 25% |
| −10% | −10% | Rp90 million | 25% |
For example, the mixed outcome is Rp50 million × 1.10 plus Rp50 million × 0.90, or Rp100 million. The probability of ending below Rp95 million is 25%: only the last row qualifies.
Now assume A and B always move together. Each asset still has the same two possible returns and the same individual probabilities. But there are only two joint outcomes: both rise, with probability 50%, or both fall, with probability 50%. The chance of ending below Rp95 million becomes 50%.
Expected ending value remains Rp100 million in both models. The downside probability doubles because the dependence assumption changes, not because either asset's standalone return assumption changed. More holdings do not automatically provide the diversification implied by an independence assumption.
We enumerated all possible outcomes; no random simulation was needed. That makes this a useful arithmetic check before a more complex Monte Carlo exercise. It is not a historical backtest, a live Trade, Inc. result or an estimate of your portfolio's real-world loss probability.
Ask for the calculation, then challenge one assumption
An effective follow-up is: “Keep each asset's individual return assumptions fixed, but make their joint behavior more adverse. Show what changed and what stayed constant.” In the example, an answer that leaves the downside probability at 25% under perfect co-movement has not applied the new specification correctly.
For a real portfolio, ask whether the model uses independent draws, a fitted dependence structure or joint historical observations. Each is a modeling choice. A history-based approach also depends on the period selected; it cannot establish the frequency of events absent from that sample.
Do not quietly change several inputs at once. If the assistant changes expected returns, volatility and asset relationships together, the output may be a useful combined stress scenario, but it no longer isolates the contribution of one assumption. Ask for a side-by-side specification alongside the results.
Three outputs that need precise definitions
| Output | Question to settle |
|---|---|
| Probability of loss | Below the initial value or another threshold? At the horizon or at any time along the path? |
| Percentile range | Which percentiles, in which currency, and conditional on which model? |
| Maximum drawdown | Largest peak-to-trough decline measured along each path, or something else? |
A portfolio can finish above its starting value after suffering a large interim decline. An ending-value distribution therefore does not answer a drawdown question. Likewise, the fifth percentile is not a worst-case floor: modeled outcomes can lie below it, and actual outcomes can fall outside the model's assumptions.
A seed is useful, but it is not the whole audit trail
When code is used, request the input snapshot, calculation or code, model settings, random seed and software environment. A seed initializes a pseudorandom process; it is not a certificate that a forecast is correct.
NumPy's official compatibility policy makes reproducibility conditional on factors including the generator, calls, build and environment—not the seed alone. This is a methodological reference, not a claim that Trade, Inc. uses NumPy or exposes a particular programming runtime.
The broader disclosure principle is also useful. FINRA Rule 2214 addresses methodology, assumptions, limitations and the hypothetical nature of investment-analysis outputs for covered member firms. We cite it as a transparency reference, not as a claim of FINRA approval or a legal requirement governing Trade, Inc.
Who needs conversational simulations?
Chat may fit investors whose questions change through iteration: alter the horizon, add a withdrawal, compare a concentration scenario, then ask why the result moved. The value lies in a checkable discussion of assumptions and consequences.
If your question is a single fixed-return calculation, a spreadsheet or calculator may be sufficient. If you need an audited risk system, independently validated model or production reporting guarantee, establish those requirements separately; this guide does not verify them. Never infer that a simulation can execute trades or eliminate losses.
Conclusion
Evaluate the workflow with a small example before trusting a complex portfolio chart. A useful assistant should reconcile the numbers, preserve the requested specification and explain why changing one assumption changes the result. In our example, the mean stays unchanged while downside risk rises sharply—the kind of distinction a good research conversation should uncover.
You can open Trade, Inc. chat to check availability and discuss your question. For adjacent workflows, read AI candlestick analysis and choosing a Screener for backtesting. A simulation explores modeled possibilities; a backtest evaluates specified rules against historical observations.
Published by Trade Inc Research, part of the platform discussed here. Product information checked on 30 September 2026 at approximately 14:04 WIB. Educational hypothetical example only; no personalized investment advice, new backtest or live portfolio simulation is presented.
This article is for research and information only. It is not personalized financial advice or an instruction to trade.