AI chatbots and investing

More and more people ask general-purpose AI chatbots which stocks to buy, or have them review a portfolio, including portfolios created with aivest. Chatbots are good at explaining terms and at helping you prepare questions for your bank. But their answers sound just as competent when they are not based on any data. A fluent answer is not the same as a calculated one.

This page lists what to check before you act on a chatbot's figures. It is general information, not investment advice.

Current prices

A chatbot that does not look up current market data answers from what it learned during training, and prices from that memory can be months or years old. Share counts calculated from old prices are wrong: a proposal that is supposed to fit your budget can cost considerably more or less once you place the orders.

Check: compare every price with a current quote from your bank or the exchange.

Does it add up?

For each position, multiply the number of shares by the current price and, for foreign stocks, by the current exchange rate. The sum should match your amount, and each position's share of the sum should match the weight the chatbot states. If it does not, the proposal was not calculated from current data.

Risk figures

"This lowers your risk" or "a volatility of about 12 %" are claims that need a calculation: the price history of every holding, converted into your currency, over a stated period, including how the holdings move together. Ask which data and which period were used, and whether the figure was actually calculated. If the answer cannot show this, treat the figure as an estimate.

Replacing stocks with well-known, solid-looking companies does not automatically reduce risk. Large companies, especially from the same sector, often rise and fall together, so a portfolio of individually "safe" names can fluctuate more than one whose stocks move more independently of each other.

Consistent answers?

Ask the same question again in a new conversation, or object once. If the answer changes completely without any new data, neither version was based on data. Some chatbots agree with almost any objection; that tells you something about the conversation, not about the portfolio.

Well-known names

Asked to "improve" a portfolio, chatbots tend to suggest the same handful of famous companies. That is a preference learned from text, not the result of a risk calculation. And if many people follow the same suggestions, they all end up holding the same few stocks.

Personal advice

Investment advice that takes your personal situation into account (income, assets, tax situation, goals) is a regulated activity in Germany and the EU: it requires authorisation, and the adviser is liable for the advice. A general-purpose chatbot has neither. If a chatbot asks about your personal circumstances to tailor a portfolio, keep in mind that nobody is responsible for the result.

The same applies to taxes: withholding tax on foreign dividends, for example, depends on where you live and on tax treaties. General rates quoted from memory are often not what you will actually pay; ask your tax adviser or your bank.

How aivest does it

aivest is not investment advice either: its portfolios do not take your personal circumstances into account. What it does differently is to calculate instead of recall:

• Current data. Stock prices are updated every trading day after the US market close, exchange rates every working day from the reference rates of the European Central Bank.
• Calculated risk. The risk of every portfolio is calculated from about ten years of price history of all candidate stocks, converted into your portfolio currency, including how the stocks moved together.
• Checked before you see it. Every new version must pass a quality check of its risk profile. If none passes, you receive an error message instead of a weaker portfolio.
• Share counts that add up. They are calculated from current prices and exchange rates for the amount you enter.

Where the AI comes in: the portfolio is put together by a genetic algorithm, an AI method that evolves many candidate portfolios over generations and keeps the best. An AI language model also helps to check the data of the roughly 2,000 companies once a month. The AI does not predict prices. More under AI Portfolio Generator.

If you ask a chatbot about an aivest portfolio, the checks above apply to its answer as well.