Using AI to create a robust investment plan
I’ve been dabbling in AI. Began using precursor “Expert Systems” about 20 years ago, but the new apps are more generalized and interesting. I’m aware of the limitations and anyone who wants to use something like Gemini or ChatGPT should also be aware. They can (and do) generate false information with apparent confidence. This can deceive users. Such disinformation has been given the name "hallucination" or "confabulation" by AI experts. Interesting names for inaccuracy.
However, using precise prompts seems to improve the response.
I’ve been running tests on Gemini using a range of prompts to build an in-retirement portfolio. I call these "tests" because I have my answer to compare this AI expert to. My early explorations indicate this might be a useful tool to add to my other modelling methods such as Monte-Carlo simulations. I’m also exploring approaches to a more robust withdrawal strategy. All of this is to aid my younger spouse.
If anyone is interested in Gemini's complete response, simply copy and paste my prompts into a query. However, AI being what it is, I expect every response will be slightly different.
Adding a few terms to the prompts does generate differing results. For example, I asked Gemini to:
“Provide a model of an in-retirement portfolio. The portfolio is $1,500,000. Annual withdrawal is 5%. The portfolio is to have a duration of 25 years. Provide an optimal mix of stocks and cash or bonds. Stocks may include ETFs, individual stocks in the US and global. Minimize risk.”
I used 5% withdrawal as a means to stress the response. I received a 1,000 word response. A statement of goals and assumptions (inflation, returns and withdrawal rate) were made. Then Gemini went on to provide an “Optimal Mix of Stocks, Bonds, and Cash (Risk-Minimized Focus)”, a “Detailed Asset Allocation”, “Withdrawal Strategy” and “Management & Rebalancing”. The portfolio was 50/50 bonds-cash/equities. It used specific ETFs in its examples.
Gemini summed its response this way: “This model provides a robust framework for a risk-minimized, in-retirement portfolio designed for a 25-year duration with a 5% annual withdrawal. The emphasis on high-quality bonds and diversified equities aims to generate income, preserve capital, and offer some inflation protection, all while prioritizing risk management.”
Adding Dividends. I then modified the query: ““Provide a model of an in-retirement portfolio. The initial value is $1,500,000. The initial withdrawal rate will be 5%. The duration is 25 years. Consider inflation and the possible returns from US and global stocks, as well as cash and bonds. Should dividend paying stocks be included and if so, what percentage of equities?”
I received a more detailed 1,500 word response. This portfolio was 60% bonds-cash and 40% equities. It included a list of example ETFs for stocks and bonds. The allocation percentages of each were provided. A Bucket withdrawal approach was outlined.
Gemini summed it this way: “This comprehensive portfolio model balances income generation, inflation protection, and growth, aiming to provide a sustainable retirement income stream for 25 years while effectively managing risk and leveraging the benefits of dividend-paying stocks.”
However, using precise prompts seems to improve the response.
I’ve been running tests on Gemini using a range of prompts to build an in-retirement portfolio. I call these "tests" because I have my answer to compare this AI expert to. My early explorations indicate this might be a useful tool to add to my other modelling methods such as Monte-Carlo simulations. I’m also exploring approaches to a more robust withdrawal strategy. All of this is to aid my younger spouse.
If anyone is interested in Gemini's complete response, simply copy and paste my prompts into a query. However, AI being what it is, I expect every response will be slightly different.
Adding a few terms to the prompts does generate differing results. For example, I asked Gemini to:
“Provide a model of an in-retirement portfolio. The portfolio is $1,500,000. Annual withdrawal is 5%. The portfolio is to have a duration of 25 years. Provide an optimal mix of stocks and cash or bonds. Stocks may include ETFs, individual stocks in the US and global. Minimize risk.”
I used 5% withdrawal as a means to stress the response. I received a 1,000 word response. A statement of goals and assumptions (inflation, returns and withdrawal rate) were made. Then Gemini went on to provide an “Optimal Mix of Stocks, Bonds, and Cash (Risk-Minimized Focus)”, a “Detailed Asset Allocation”, “Withdrawal Strategy” and “Management & Rebalancing”. The portfolio was 50/50 bonds-cash/equities. It used specific ETFs in its examples.
Gemini summed its response this way: “This model provides a robust framework for a risk-minimized, in-retirement portfolio designed for a 25-year duration with a 5% annual withdrawal. The emphasis on high-quality bonds and diversified equities aims to generate income, preserve capital, and offer some inflation protection, all while prioritizing risk management.”
Adding Dividends. I then modified the query: ““Provide a model of an in-retirement portfolio. The initial value is $1,500,000. The initial withdrawal rate will be 5%. The duration is 25 years. Consider inflation and the possible returns from US and global stocks, as well as cash and bonds. Should dividend paying stocks be included and if so, what percentage of equities?”
I received a more detailed 1,500 word response. This portfolio was 60% bonds-cash and 40% equities. It included a list of example ETFs for stocks and bonds. The allocation percentages of each were provided. A Bucket withdrawal approach was outlined.
Gemini summed it this way: “This comprehensive portfolio model balances income generation, inflation protection, and growth, aiming to provide a sustainable retirement income stream for 25 years while effectively managing risk and leveraging the benefits of dividend-paying stocks.”
The post Using AI to create a robust investment plan appeared first on HumbleDollar.
No comments have been added yet.
