using AI for pension projections

SouthCoastBoy

Has anyone used an AI tool to validate their projection numbers?

I used chatgpt and found it very interesting and useful, especially how resilient my pot would be in a crash and adds another available tool for pension planning with the obvious caveats.

dunstonh

Has anyone used an AI tool to validate their projection numbers?

Given the number of errors and inconsistencies AI is prone to, even with very simple calculations, I wouldn't trust it on a single chat prompt or even a handful of chat prompts. You would need to run it under a project with a large amount of instructions and checks. You would need to feed it the external data and not rely on its own lookup. You would want it on a high level of thinking and best model. You would then want to test the results against other non-AI modelling tools for validation before being happy to rely on it.

SouthCoastBoy

Yes agree, don't take the responses at face value, however I found it useful to get another perspective. I'm sure over time it will improve, its definitely not going to be perfect from the start, however in 5 to 10 years time it could be a different story.

Aminatidi

I wouldn't use them to validate but I do use it as a set of numbers and augmentation of anything else I've done.

NormalNorman

I’ve used AI to construct graphs from my data. Don’t have the skills or interest in creating complex graphs myself. Also useful when comparing funds. I agree with the above, you really need a good set of prompts for your circumstances.

Dead_keen

Yes. I used Claude to take my balance sheet, spending, state pension expectations, tax rules, expected life, failure rate and so on to create a number of scenarios with Monte Carlo. So, for example, I can ask it how much can I spend, post-tax, so that I have a 10% chance of failure by the time there is only a 10% chance of me or my partner surviving. It's very good and I get it to do other things (get bank, credit card data automatically and categorise it, show my gilt ladder, populate my 'what happens when I die' document, reminds me of my financial todos, view spending by categories, and sub-categories and so on by month, tell me my tax liability for the year, the potential IHT due, etc).

It is quite sophisticated, unlikely to halucinate and so on. But whether the answer is any better that the 4% safe withdrawal guideline is a different question.

GenX0212

If you think AI models are good with complex numbers then try giving one a half completed Sudoko…..

dunstonh

It is quite sophisticated, unlikely to halucinate and so on. But whether the answer is any better that the 4% safe withdrawal guideline is a different question.

There is no 4% safe withdrawal guideline in the UK. That's a US-based study. The equivalent for the UK is lower at around 3.5% in your mid 60s. It falls to under 3% in your mid-50s. It only gets to over 4% in your 70s

SouthCoastBoy

I not sure about these safe withdrawal rates, as the past does not necessarily match the future, also there is the fx element to consider and also the fact that many uk investors invest in world stocks not just uk. I've gone for 2.5% withdrawal rate for first 9 years, then down to about 1% when my spouses state and db pensions kick in

Dead_keen

I've gone for 2.5% withdrawal rate for first 9 years, then down to about 1% when my spouses state and db pensions kick in

One thing that is interesting with Monte Carlo modelling is that you can see what the upside and downside might be. So you can guess at what you might be able to spend on so that you have a 10% run out of money by the time you have a 10% chance of being alive. But it will also illustrate how you might end up with £28 million (in today's money) instead. So spending too little means that you might end up being able to afford the whole graveyard, and still pay eight figures of IHT.

dunstonh

I not sure about these safe withdrawal rates, as the past does not necessarily match the future

That's why they're not generally referred to by professionals as safe withdrawal rates. You could be asking for a missale if you start using phrases like that.

However, reliable UK data goes back to about 1915. So it's a pretty long period, and it would cover most scenarios. That said, the worst 10 year period for equities was March 1999 to Feb 2009 and we are currently in the worst period for bonds