I don’t make predictions often. It’s not my style. But, every once in a while, there are a few things that I can’t stop thinking about and I want to call my shot on. Maybe I will be proven right. Only time will tell.
Here are three things I think we will see play out over the next decade.
U.S. Residential Real Estate Will Underperform Inflation
I’m going to say it—U.S. residential real estate has risen too much, too quickly. As a result, I think it will have a negative inflation-adjusted return over the next 10 years.
For the last 130 or so years, U.S. housing barely beat inflation. However, since the year 2000, prices have gone almost straight up on two occasions—the mid-2000s housing bubble and what we are experiencing today:
Of course, this doesn’t imply that we are in a massive housing bubble today like we were in the mid-2000s. Nevertheless, it’s concerning. Though the typical American’s balance sheet is much stronger than it was back in 2007, I don’t think home prices can keep going up like they have been in real terms. Unless real incomes rise much more than expected, something will have to give.
Does this mean that I believe prices will decline a lot? Not necessarily. I could see a scenario where prices don’t drop, but stay flat or increase slower than inflation over the next decade. Of course, it’s anyone’s guess what home prices will do over the next few years. Maybe they will continue to increase as interest rates decline and more homebuyers enter the market. Or they could crash as more supply comes online as sellers relocate.
Either way, I don’t think U.S. housing is going to behave in the future anything like it did since 2000. So, if you are in the market for a home, think of it as a purely consumption good and act accordingly. This is good advice in any market environment, but especially when home prices are elevated relative to history.
Another option you have is to continue renting until home prices seem more reasonable. Unfortunately, by the time home prices become “reasonable” again, you won’t be very enthusiastic to buy a house. Just look at what happened to U.S. home prices from 2009-2011. They were cheap (relative to now) because things looked so bleak at the time.
The same thing will eventually happen again in the future. An unforeseen event will cause a recession, home prices will decline, and you’ll get your chance to buy at a relative discount. But, will you have the courage to actually go through with it? That’s the hard part.
Though I may be dead wrong on U.S. housing, I’m throwing in with the historical data. U.S. housing has been odd for a few decades and I’m skeptical that such a trend will continue.
Now that we’ve discussed the future of U.S. housing, let’s examine how I think AI will develop in the coming years.
Custom Chatbots Will Be All the Rage
The next big thing I am thinking about is custom chatbots. ChatGPT, Claude, Perplexity, and the other LLMs are all great at being your portable Wikipedia. They are incredible generalists, but they don’t have deep domain expertise. To get such expertise, we would need to train them on a specific person (or set of people). This is where I think the real value will be unlocked with LLMs.
So instead of having a general AI to help you with cooking and recipes, imagine having a Martha Stewart or a Gordon Ramsay to chat with. And these chatbots would answer almost exactly how Martha or Gordon would. I know people are already trying to do this today, but I haven’t seen anything that’s blown me away yet. But, I know we will get there.
I think the real challenge of creating these custom chatbots will be getting deep personality insights from the limited information we have on someone. Yes, we may have thousands of hours of Martha and Gordon cooking, but is that enough to answer any conceivable cooking question they might get asked? I’m not sure.
You can imagine two extremes when training an LLM model. First, imagine if there was a single question that I could ask you that would perfectly predict your answer to any other question. Of course, no such magical question exists, but if it did, then we could train a model by just asking that one question.
Now think of the flip side. Imagine I sat down with you and asked you every possible question imaginable and recorded all of your answers. If I had this list of responses (all infinite number of them), then I could perfectly predict how you’d answer any question in the future since I would technically have already asked it. Of course, this is impossible since gathering and storing such information is far too costly.
Therefore, the actual solution must be somewhere in between these two extremes. Initially, it might take thousands of hours of interviews with someone to get enough information about their personality to build a custom chat bot. But, as we get better at doing it, we will find shortcuts that better predict responses. Thousands of hours of interviews will become hundreds of hours of interviews. And, one day, we will build a custom chatbot without needing much time at all.
It reminds me of a line from Neal Stephenson’s Snow Crash. In the book, one of the characters tells a story about how she got pregnant at 15 years old and no one in her family knew. Yet, one day her grandmother came to visit and figured it out within 10 minutes just by watching her face from across the dinner table. The girl was stunned that her grandmother was able to learn of her secret with so little information. As Stephenson summarized:
To condense fact from the vapor of nuance.
This is how we are going to build the chatbot that acts like you 99% of the time. We are going to figure out the vapors of nuance that define a personality. I have no idea how, but we will get there.
And once you have your own chatbot, then the sky is the limit. Your chatbot will be able to perform all the mundane tasks that you don’t want to (e.g. answer emails, schedule meetings, etc.). Influencers will be able to “talk” with millions of their fans at once.
I could even imagine financial experts licensing their chatbots to their fans to answer financial questions. For the 36% of individual investors who are do-it-yourselfers (DIYers), this might be the ideal solution that will finally serve this client segment. While DIYers would never pay an AUM fee, they might pay a $5 monthly subscription for unlimited access to their favorite financial influencer. Who knows?
Yes, there is a lot of conjecture here, but this is how I imagine things will play out as chatbots improve over the next decade. Customization will be all the rage. And, as custom chatbots help us do our work, we will continue to redefine how we think about work altogether.
More People Following Coast FIRE
As the role of work changes in our society, I think we will see more people than ever pursuing Coast FIRE. For the uninitiated, Coast FIRE is the idea that you should save up enough money for your future retirement and then coast until you get there. This allows you to transition into a lower paying, but more fulfilling, job mid-career once you’ve hit your Coast FIRE number.
With fewer individuals doing multiple decades at the same company and the rise of remote work, people are changing how they view their careers. The person writing on this topic better than anyone else is Paul Millerd. His first book The Pathless Path and his most recent, Good Work, both address the existential crisis of work gripping corporate America.
I am personally a big fan of Coast FIRE and think it is the perfect financial philosophy for a world growing tired of the traditional career path. While I don’t know what the future of work will look like, I am willing to bet that there will be far more people striving for Coast FIRE or something similar.
Life is too short to do something you don’t love. I think Coast FIRE is going to bridge the gap between the old way of working and the new one that people are coming to define for themselves.
Anyways, that’s all I have for now. I can’t know the future, but I’m still excited to see how things play out. Thank you for reading!
If you liked this post, consider signing up for my newsletter.
This is post 426. Any code I have related to this post can be found here with the same numbering: https://github.com/nmaggiulli/of-dollars-and-data
