Is AIvestor a ready-to-use investment product?
No. AIvestor is a research experiment.
I am testing the decision-making process in investing using AI, agent-based architecture, and market data.
It is not investment advice, not a product to buy, and it does not promise any profits.
It is a laboratory designed to explore whether it is possible to invest more calmly, consistently, and transparently.
Why test on the Warsaw Stock Exchange (GPW), often called a “banana exchange”?
I chose GPW deliberately. It is a market I know, I have data access, and I understand its specifics.
It is volatile, sometimes undervalued, and prone to speculation – which makes it a perfect experimental proving ground.
If the system can work here, it has even more potential on larger markets.
Why focus on low-liquidity and small-volume stocks?
This is not arbitrary – it is part of the system’s configuration.
The user can set filters for:
- which market to trade,
- minimum turnover thresholds,
- whether to exclude illiquid “plankton” stocks.
Currently, filters are set low to generate as many diverse test cases as possible.
Additionally, the user can:
- disable automation and select stocks manually,
- request on-demand analysis of specific companies,
- add instruments to the portfolio manually.
In such cases, the system only handles monitoring and managing positions.
Why not test directly on Wall Street?
Because the initial goal was to quickly close the experimental loop.
I know GPW, data is accessible, and entry barriers are lower.
Building such a system from scratch is challenging enough – at this stage the priority was to test whether the architecture itself makes sense.
Why paper trading, if only real money proves anything?
Paper trading is standard in research.
The system runs on real-time market data via broker API, the only difference is simulated capital.
This allows safe testing of dozens of scenarios, including extreme ones, and even deliberately “breaking” a portfolio without financial risk.
At this stage, the focus is not profit but decision mechanics.
Do micro-positions prove anything?
Yes – because every position, regardless of its size, triggers the same cycle:
analysis → TP/SL → monitoring → decision to reduce, close, or extend.
The system can add to a position if:
- the user’s strategy allows it,
- there are free funds available,
- current exposure is below the defined limit.
This demonstrates that the portfolio is not static.
The purpose is not the amount of money, but testing whether the system behaves consistently across scenarios.
Can you really trust AI in investing?
It is not about “blind trust.”
AIvestor is not a black box – every decision has an explanation.
The aim is not to outperform funds or HFT algorithms, but to check whether AI can support investors in a way that is transparent and aligned with their intent.
Does this work with large capital and scale?
Not in the current version. This is a laboratory.
If the approach proves valid, scaling is primarily an engineering challenge, not a conceptual one.
That is why I start small – to validate the fundamentals first.
Isn’t this just a “toy for geeks”?
Maybe. But every innovation starts this way.
Personal computers were once “toys for geeks.”
For me, this is a research experiment addressing real questions about the future of decision intelligence.
Why experiment at all, if the results don’t guarantee profit?
Because no investment ever guarantees profit.
Here, the question is not ROI, but:
can a system make decisions that are consistent, transparent, and aligned with user intent?
Why not just ask ChatGPT for an investment strategy?
One prompt = one answer.
AIvestor is different:
- it runs continuously,
- reacts to events (prices, news, sentiment),
- remembers context (decision history, limits, prior assumptions),
- manages the portfolio over time (updates TP/SL, reduces, closes, extends positions),
- and logs + explains every decision.
Most importantly: AIvestor curates its own context.
It integrates data from ~30 scrapers and trusted sources (industry portals, financial reports, ESPI announcements, company metrics).
That way it is always clear what information was used and why a decision was made.
It is not “just a prompt” – it is a decision architecture with controlled data sources.
If it’s so good, why not invest real money?
Because this is an experiment. Capital is simulated on purpose – to safely test risky and extreme scenarios without consequences.
The goal at this stage is decision mechanics, not financial return.
How can one person compete with hedge funds and their million-dollar budgets?
I am not competing with funds.
AIvestor is a research lab, not a hedge fund.
The purpose is to test an agent-based, transparent decision approach – not to beat HFT benchmarks.
Why write posts instead of just “making money”?
Because I am not selling this system and not promising profits.
Communication is part of the experiment – showing process and lessons learned, not a “magic algorithm.”
This is only paper trading – real markets would crush it.
Paper trading is a conscious choice – a standard in research.
The system works with real market data, the only difference is simulated capital.
This makes it possible to test hundreds of scenarios – before any move to a real account would make sense.
What about crises? How would AIvestor handle 2008 or 2020?
The system is designed to support Black Swan scenarios.
Through user-defined risk parameters and continuously monitored market sentiment (affecting the percentage of portfolio exposed to investments), the portfolio can be reduced during extreme uncertainty.
This is intended to mitigate the impact of shocks such as 2008 (subprime), 2020 (COVID), or 2022 (war).
There are hundreds of auto-investment systems. Why is yours unique?
Most systems are black-box SaaS tools or simple TA algorithms.
AIvestor:
- runs in a multi-agent architecture,
- builds context from ~30 data sources,
- operates in event-driven mode,
- provides full explainability – every decision has reasoning and a log.
And what if your system just doesn’t work?
That’s possible – and that’s exactly why it’s an experiment.
I am not testing a guarantee of success, but whether agent-based, transparent decision-making makes sense.
If it doesn’t – that itself is a valuable conclusion.
How do you cope with new models / versions of LLMs?
This is part of the system’s configuration. Whenever a new model is released, it usually represents my intent more accurately. Since AIvestor is an experiment, new models simply make it “smarter.” If the concept proves itself, I may later add backtesting and simulations to measure the impact of switching models.
How do you avoid historical data overfitting?
AIvestor is not a high-frequency trading (HFT) system. I’m not building a model to squeeze micro-patterns from the past. Instead, each agent follows a user-defined strategy (in natural language) and reacts to events in real time. It’s closer to having a “digital twin” of the investor than an optimized statistical model.
What about period performance differences (e.g., pre- and post-COVID)?
Same answer as above: this is not an HFT backtesting engine. AIvestor is designed to operate now, in the current market, following strategy and sentiment rather than trying to perfectly predict shifts like COVID.
How does the system handle black swan events?
Through sentiment monitoring. The MarketSentimentAgent evaluates news flow and global context, allowing the system to reduce exposure or stop opening new positions during crises.
Do you take into account execution costs?
For development purposes, I operate entirely on the broker’s demo environment. Execution costs are not a focus at this stage.
What about trade latency and slippage?
This is not HFT and not production-grade execution. I’m aware of latency effects, but they are not modeled at this stage of the experiment.
Financial data quality is notoriously hard for retail investors. How do you address that?
Again, AIvestor is an experiment, not a product. I don’t sell anything, so I rely on public, curated, and reliable data sources (e.g., ESPI, official WSE feeds). The point is consistency and explainability, not perfect data arbitrage.
Do exchange rates affect your results?
Currently I operate only on the local market (WSE: WIG + NewConnect). No FX spread applies. My broker also does not charge commissions below a certain annual limit, and I’m far below that threshold.
Do you analyze financial statements as well?
Yes, company reports are included in the data pipelines. They enrich the context for LLM-based reasoning.
Why not focus on a single market, given how different they are?
That’s exactly what I do. AIvestor is limited to the Polish stock exchange for now (WIG + NewConnect), for reasons also explained in the this FAQ.
What about fast alpha decay?
I don’t optimize for “alpha discovery” in the traditional quant sense. The goal is not maximizing statistical edge but ensuring consistent, explainable, and strategy-aligned behavior.
Where does trading strategy and domain knowledge come from?
Strategy = the user’s intent. AIvestor allows users to define strategies hierarchically:
- global rules,
- instrument-specific strategies,
- position-level overrides.
The LLM is instructed to follow these intents, with indicators and metrics already pre-calculated. It’s less about “finding alpha,” more about enforcing discipline.
Did you check that you’re legally ok to scrape the data you are scraping?
I use public sources. And importantly: this is not a commercial product, it’s a research experiment. I don’t sell access, I don’t charge fees, I don’t give investment advice.
Do LLMs capture stock-specific features like seasonality in retail?
LLMs don’t “magically know.” I explicitly instruct them which factors to consider, and I can embed structured signals (like seasonality) into the context. This way, the reasoning process stays under control.
Could the system unintentionally engage in market manipulation (e.g., spoofing/layering)?
That’s a very valid point. For now, since AIvestor runs in demo mode and within a limited scope, this is not an issue.
Even in a scenario of broader usage, users’ “strategies” and “intents” would be diverse and executed at different times, so systemic manipulation risk remains low.
In future iterations, a dedicated ComplianceAgent could be added to monitor system behavior and proactively detect and prevent manipulative patterns.
Is AIvestor just another quant trading bot or HFT experiment?
No – AIvestor is not a quant trading product, and it does not compete with hedge funds or HFT systems. They already do that, and they do it much better.
AIvestor is a completely different concept: an autonomous decision-support system that protects the investor from themselves, automating consistency instead of chasing fleeting market edges.
💡 In one sentence: AIvestor does not try to beat the market in milliseconds – it protects investors from their own emotions by turning strategy into consistent action.