With the benefit of a 40-year career in financial advice under my belt, a thought struck me when playing with Chat GPT this month.
I don’t claim any expertise in AI, but like many others, I have been lured in to explore it and see just how far it will go. Last week I asked it if it could help me execute pension income drawdown. I told it my age and pension pot size and off it went!
The results are quick, comprehensive and delivered with such confidence: “sure I can help you do that” and “great question, would you like me to build a calculation for you showing the balance in your account over time?”
The results were interesting.
Our CEO, who does know a bit about AI, pointed out that if I went through the exercise again I might get a different answer. He was right! Not a big difference, but one query prompted a question as to whether I had considered an alternative annuity, the other very similar query didn’t.
The UK regulator, the Financial Conduct Authority (FCA), have an unenviable – if not potentially impossible – task of regulating AI outputs like this, which legally in the UK should not be giving financial advice. As ever, the tech bros appear to be cracking on regardless and worrying about rules and regulations later.
The guidance (or advice, call it what you want) broadly said:
- Invest in a balanced fund, c40-60 or 60-40 equities and bonds
- Withdraw 4% if I wanted a sustainable withdrawal rate, but I could draw more
- It even went on to name check some providers of low-cost products
I won’t name check the providers as they get enough publicity already, but suffice to say they all offer un-advised solutions and completely or mainly passive solutions using index funds. Two were well-known American brands, although I’m sure that’s complete coincidence!
It was good at highlighting things like the money purchase annual allowance restricting contributions once you start drawing income. In one exchange it even suggested I might need to split out my funds into pots to avoid return sequencing risk – very helpful. It also offered to provide me with the sources of its knowledge, which all looked sensible if I wanted to check further.
To the uninitiated, it would have felt convincing, comprehensive, free, and delivered almost instantaneously. No lengthy exchange of personal circumstances and no big difficult questions around risks. What’s not to like?!
Well firstly, is it right?
Here’s what Google’s Gemini AI says about Open AI’s (mostly owned by Microsoft) Chat GPT if you ask it “is Chat GPT ever wrong?”:
ChatGPT is not always right, and studies suggest it answers programming questions incorrectly about 52% of the time. While it can be helpful, its responses may contain inaccuracies, especially on complex topics or when dealing with unique or unfamiliar queries. It’s important to verify its answers, especially for critical applications like programming.
Perhaps financial advice is easier than programming?
Then there is the issue of the differing answers and inconsistencies. In one query I’m shown a 1.5% real return (4% growth, 2.5% inflation).
In the next query, I’m shown a 4% real return. I’m pretty sure I answered the questions the same, but one query is using 1.5% real and the other 4% real – quite a difference!
I am told I could draw more than 4%, describing 5.9% (an almost 50% increase) as “slightly above” its original suggestion. Some interesting statements here on that below. I like the idea of a ‘smart drawdown strategy’ whatever that is!
To the initiated it looks like the ‘Wild West’ and very dangerous!
Consensus vs sense
But these aren’t the core reasons for my big beef with AI advice. I am sure it will get better with time and some clever firms will put the relevant guidance and crash barriers around it to keep it on the straight and narrow.
My real issue is with the fundamentals of the process it is using to offer its advice and guidance. That is ploughing over masses (and I’m sure they are big masses) of historical articles and website material and coming up with a consensus view on how to do something. The process is both historic-looking and consensus-based.
Relying on past performance and acting like the herd are in general two features that, when brought together, often spell disaster in financial matters. For example:
- The great tulip bubble of 1637
- The dot com bubble of 1999
- The Great Financial Crisis of 2008
All come instantly to mind, although I wasn’t advising in 1637!
However, I have been through property market crashes, negative equity, the collapse of Equitable Life, the demise of endowment policies, the collapse of Northern Rock, the demise of split capital investment trusts, film schemes, and more.
Some caught me out, some I steered serenely around. In 2012, my then business partner and I (and some seriously financially astute clients) worked out that, contrary to mainstream views, there was a substantial (and potentially time limited) windfall opportunity for members in DB to DC pension transfers in the post-financial-crisis quantitative easing era. As a firm we went on to complete £1bn in transfers which, as of today, have probably collectively added over £500m in value to those clients’ pension arrangements. You can read more about this here.
The standout features of the most successful advice I have given in my career so far are that it was based on current and forward forecast financial conditions, not the past, and it was against the consensus view of the time.
The core issues with ChatGPT’s advice
Coming back to Chat GPT’s advice and guidance around drawdown and putting the scary inaccuracies to one side, I see two main issues:
The 4% consensus view on pension withdrawals
With UK pensions coming into Inheritance Tax (IHT) and 6-8% p.a. forward corporate bond returns looking achievable for the foreseeable future, many 60 plus year olds like me will need to draw a lot more than 4% a year to avoid a double tax hit on any unused pension.
By my calculation, I think I should be drawing between 7% and 9% a year from my pension (depending on if I want an income linked to inflation or one that gradually falls as I get older) in order to exhaust my account by age 85 and rely on other sources of income in later life.
If you are going to only draw 4% at my age, you may as well buy an annuity and let an insurance company take the windfall money – its either them or a tax liability payable to HMRC if you leave it in your pensions.
The 60/40 blend it recommends, again a consensus view
This blend has worked well in the past and it would not be hard to justify it based on past performance, but will it work for the next 10 years?
I would be cautious about putting more than 30-40% in equities at my age in a drawdown account, and very cautious about using a passive approach, based on the focus and elevated values of US equities that dominate US indices.
I don’t want any of my pension fund invested in Tesla, based on price/earnings ratio (PE) of 200 (according to Yahoo finance) that’s rising rapidly as Tesla’s car sales tank. Even if Tesla’s humanoid robots and driverless taxis do bring the company vast riches and the share price goes to the moon, I would not want it in my pension fund. It would be very tax inefficient if I got that lucky.
If Tesla or any other big US tech firms run into trouble and hold back world indices, I could be missing out on a more predictable 6-8% p.a. compounding return from fixed income, which will more than get the job done. As the Dragons say, “I’m out”.
You can read more about the risk of US equity concentration here and withdrawal rates and IHT implications for pensions here.
What would I do instead?
I’ll look again in a few years’ time and see how (or if) it’s developed. In the meantime, I will give it a wide berth, and I suggest you do too.
Great advice is not free, but it can save or make a small fortune – if you want to invest in past returns and follow the herd, you know where to go!
