I am not clear, are you saying the +2% healthcare incorporates Medicare, not a separate add-on, meaning it's not +2% healthcare PLUS an additional 2% for Medicare?
@pizzaman It looks like Build > Get Started > Scenario Assumptions > Inflation Rates includes:
| General Inflation Rate |  Healthcare Costs | Long Term Care Costs | College Expenses | Medicare Premiums | Medicare IRMAA Income Limits |
|---|
You must be talking about the online version, that is not an option in Gold 2025, which is what I use. That explains it 😉.
Another valuable resource for predicting returns is tpaw's regression analysis of the sp500. https://tpawplanner.com/plan/expected-returns-and-volatility
...These are purely valuation based estimates, so returns tend to be quite small lately.
Mathews runs the regression for 5, 10, 20, 30y return averages daily, and in each of those spans for "all data" and "data since 1950". With this info of 8 data points you can establish expected returns for short term (averaging his four 5y and 10y figures) and long term (the 20 and 30 year figures). For instance, currently, the 7y estimate would be roughly 4.4% real (6.8% nominal/ 3.3% cagr real), and the 25y estimate would be roughly 5.8% (8.4% nominal, 4.2% cagr real). There is a lot of variation around the regression, so consider these the center of a wide cloud of returns.
@jkandell, Thank you for sharing this interesting and valuable resource!
I noticed the suggestion to average the 'All Data' and 'Since 1950' samples. Could you walk me through the rationale?
My instinct is that this approach lacks a clear 'point of view' and creates a double-counting bias for the more recent data. Generally, I find it preferable to identify the single 'best' model through backtesting or bootstrap analysis rather than blending them. I’d love to hear why you feel an average is more reliable than choosing the specific framework you find most fundamentally sound.
I noticed the suggestion to average the 'All Data' and 'Since 1950' samples. Could you walk me through the rationale?
My instinct is that this approach lacks a clear 'point of view' and creates a double-counting bias for the more recent data. Generally, I find it preferable to identify the single 'best' model through backtesting or bootstrap analysis rather than blending them. I’d love to hear why you feel an average is more reliable than choosing the specific framework you find most fundamentally sound.
I don't think Mathews has given a rationale. (His purpose is to find a discount to use in his own actuarial bootstrap analysis, the heart of his site.) But--just speaking for myself--I'm okay with the "double counting" because it gives more weight to more recent results, an informal exponential average if you will.
One of the criticisms of going back to 1871 (shiller's earliest) is that the data itself is less reliable and that the regime was different enough to make analogizing questionable. One answer to that problem is to start newer e.g. 1928. But that produces problems of statistical significance. So another response is to include all data but with less weight to the older stuff. (Pralana chooses a third answer: add in some user expected returns to fluff out the historical data with more data points.)
So with TPAW we have 1848 monthly data points, 900 of which are counted twice. No one doubts that the post WW2 (1950 to present) data is the current regime, with US companies the most powerful, the most innovative, situated in a tax-favorable and stable environment, with US military and cultural dominance in the background. (As an aside, I do worry that that economic regime may be ending this year, RIP, but it's too early to tell.) There are also good arguments that interest rates, and post Keynesian monetary policy, the change in CAPE from buybacks, elimination of gold standard, and the existence of consumer mutual funds all serve to make the more recent data distinct from older data.
This is especially true of the 5y regression: you'll note the 30 year "all data" and "1950 only" regressions are identical, but the difference increases as we get closer to short term predictions. We know short term predictions are less reliable, so averaging the two helps minimize the error.
So, in short, Mathews strikes a middle ground between the benefits of including all the data at our disposal while favoring the more recent, valid, and reliable data. It's rough-and-ready to be sure. There is no reason you couldn't just use his "all the data" or "just 1950 on" if you thought one were more valid. I'll also note that Mathews includes the simple 1/CAPE--the purest form of the valuation model (though very low: 2.3%!)--in his overall stock average. (I ignore it in my own numbers.)
FWIW, I don'tuse the 15y and 30y figures from my post as my only two data points. I also look at Research Affiliates own regression (which includes international), and I look at the Growth & Yield (Gordon) model. And, thanks to you, I am going to start including Damodaran's i-erp model as one of my "short term 10 year" data points. Since I have no reason to favor one model over the others, I will likely just average all the 10y ones and then also average all the 30y ones. (Why? When in doubt: minimize your error. :)) The problem with monte carlo and bootstrap is they generalize, so miss the current context (more than random chance of return to mean). The valuation-returns regression captures that the best; but it misses things in the other methods.
The most sophisticated analysis after Damodaran is Karsten Jeske's. His is purely historical based like Pralana's (so doesn't generalize with bootstraps or monte carlo) but does take into account the current CAPE level (revised by him, like Damodaran, to account for buybacks) and also the relation of the current sp500 to its recent "high". Unfortunately, he only filters these with the consumption-smoothed-withdrawal amount, not with return rates. So it's only an indirect data-point, not something that we can add into the mix for deriving expected returns. But I do compare his own withdrawal amount to my own withdrawal amount as a check and balance.
Yes it is a separate entry - so healthcare is +2% above the 3% general inflation and so is Medicare.
I can't find the source, but I started using 3% for Medicare inflation (above cpi) after I read that the official government actuaries who review the books are assuming that rate of increase.
Yes it is a separate entry - so healthcare is +2% above the 3% general inflation and so is Medicare.
I can't find the source, but I started using 3% for Medicare inflation (above cpi) after I read that the official government actuaries who review the books are assuming that rate of increase.
I’m skeptical that the current "+3% healthcare spread" will last through my 50+ year retirement plan.
Right now, the US spends ~18% of GDP on healthcare (vs. ~12% in the rest of the developed world). If we assume healthcare grows 3% faster than everything else for the next five decades, it would mathematically account for ~1/2 of our total GDP in 50 years. That's clearly impossible. While I can’t predict exactly how the sector will evolve, I’m highly confident that this growth gap will have to close well before my retirement plan concludes.
Predictions are worthless: Stocks blow past expectations in 2025
@Pizzaman, I agree that short-term predictions on the range of ~1 year are of very little use, because a 1-year prediction of ~4% +/- 17% Standard Deviation is basically admitting that you have no idea. However, I believe that long-term predictions about average returns over a 30+ year can be moderately accurate because the annual expected returns accumulate and the annual "noise" tends to cancel out.
I also believe that making some long-term predictions is an essential and unavoidable part of the the retirement planning process.
If we make any decisions at all, then we can't help but have made some (relative) predictions.
I predict that over the long-run stocks will outperform bonds, which will outperform cash. So I buy mostly stocks, with a few bonds, and I keep very little cash.
If I predicted otherwise, I would invest differently and my investment choices would reveal my (different) predictions... even if pretend to deny that I made any predictions at all.
Show me what your investments are and I'll tell you what your (relative) predictions are.
Predictions are worthless:
Stocks blow past expectations in 2025
Forecasters unsure
if trend will continue
Daniel de Visé
USA TODAY
A week after President Donald Trump’s so-called 'Liberation Day' on April 2, he paused most of his 'reciprocal' tariffs, dialing them back to 10%. The stock market soared. Annabelle Chih/Getty Images file
As recently as June 2025, many stock analysts predicted that the year would end without without any real gains in the stock market.
At that time, several prominent forecasters projected the S&P 500 index would close out 2025 in a range between 5,600 and 6,100. The S&P had started the year around 5,900.
When a turbulent 2025 finally ended, the S&P closed at 6,845.5, up more than 16%.
That was great news for stock owners. But it also raised some questions in hindsight: How could so many forecasters have been that far off on where the market was headed? What changed between June and December?
To answer them, let’s go back to the start of 2025.
The S&P finished 2024 at 5,881.6, wrapping a spectacular year that saw the index rise by 23%.
In a typical year, stock forecasters 'tend to cluster around 5% to 10% gains in their year-end forecasts,' said Jeffrey Buchbinder, chief equity strategist at LPL Financial. Stocks gain about 10% in an average year. So, all else being equal, a year-end forecast in the 5%-10% range isn’t likely to be far off.
LPL Financial forecast that the S&P would end 2025 somewhere between 6,275 and 6,375, gaining roughly 7%-8%, according to a Jan.1 roundup of stock market forecasts by Bloomberg. Bank of America predicted, rather ominously, that the S&P would end the year at 6,666. JPMorgan Chase put the figure at 6,500.
But many of those forecasts shifted dramatically after President Donald Trump’s so-called 'Liberation Day.' On April2, Trump announced a universal 10% tariff on all imports, with additional import taxes on many countries, displaying the figures on an oversized board.
By April8, the S&P had plummeted below 5,000, down almost 20% from its then-record high of two months earlier.
'Investors sold first and asked questions later,' said David Meier, a senior investment analyst at The Motley Fool.
Traders feared Trump’s tariffs would seed runaway inflation, and that consumers would stop spending. They also feared the unknown: U.S. tariffs hadn’t ranged so high in more than half a century.
'The tariff rates that he had on the board were essentially ridiculous,' Meier said. 'Meaning, they were so high but did not have any real justification under them. So, the market reacted, in my opinion, perfectly rationally.'
A week after Liberation Day, Trump paused most of his 'reciprocal' tariffs, dialing them back to 10%. The stock market soared.
'The worst-case scenario following Liberation Day did not come to pass,' said Eric Teal, chief investment officer at Comerica Wealth Management.
But uncertainty remained, and the S&P would not reach a new record high until late June.
It was during those spring months that stock market forecasters dialed back their projections and recast 2025 as a year of meager gains.
Analysts still widely believed Trump’s tariffs would trigger inflation and hamper spending. They feared a recession.
The worst fears did not bear out. America’s annual inflation rate didn’t climb past 3%.
Dire predictions about tariffs and inflation assumed U.S. consumers would bear the brunt of those taxes. That also didn’t happen. Only about 20% of Trump’s tariffs 'passed through' to consumers, according to a study by the National Bureau of Economic Research.
As imported products traveled from their country of origin to American retailers to consumers, the tariff impact softened at every stop.
'That inflation never really showed up,' Buchbinder said. 'Companies did a really good job managing it. Our trade partners ate some of it.'
Another factor that dampened stock predictions for 2025 was the prospect of an AI bubble. Throughout that year and continuing into this one, Wall Street observers have debated whether the stock market has entered 'bubble' territory: in this case, a run-up in the prices of tech stocks fueled by outsized expectations about artificial intelligence.
Perhaps the best evidence of a bubble lies in ratios of stock prices to company earnings, which sit at a historic high. Price-to-earnings ratios tell you if a stock is overvalued.
That metric was higher only once before, at the peak of the dot-com bubble in 1999-2000.
Investor surveys suggest that stock owners know all about the AI bubble. They continue to buy AI stocks anyway.
In a recent survey by The Motley Fool, 93% of investors with AI stocks said they plan to hold or expand those investments over the next year. Only 7% plan to decrease their AI holdings.
AI investment 'just blew past everybody’s expectations' in 2025, and corporate earnings came in higher than expected, Buchbinder said. Those trends drove stock prices higher.
What, then, do forecasters expect from the stock market in 2026?
LPL predicts the S&P 500 will end the year at 7,400, an 8% gain. Comerica Wealth sets the same target. Wells Fargo Investment Institute has a year-end target of 7,500, a nearly 10% gain.
The biggest drag on those projections might be the looming midterm elections. Midterms tend to work out poorly for the party in power, seeding potential volatility.
'We’ve really emphasized playing defense this year,' Teal said.
On the plus side, investors may feel growing confidence in the president’s economic pragmatism.
One lesson of Liberation Day, to many economic observers, was Trump’s sensitivity to stocks. His April2 tariffs lasted, as noted, one week. At other pivotal moments in his second administration, Trump has backed off from policy decisions that sent the stock market reeling.
For stock traders, that’s a welcome trend.
'The president, given enough time, really does care about the markets,' said Sameer Samana, head of global equities and real assets at Wells Fargo Investment Institute. 'That seems to be his report card for himself.'
Yes it is a separate entry - so healthcare is +2% above the 3% general inflation and so is Medicare.
I can't find the source, but I started using 3% for Medicare inflation (above cpi) after I read that the official government actuaries who review the books are assuming that rate of increase.
I’m skeptical that the current "+3% healthcare spread" will last through my 50+ year retirement plan.
Right now, the US spends ~18% of GDP on healthcare (vs. ~12% in the rest of the developed world). If we assume healthcare grows 3% faster than everything else for the next five decades, it would mathematically account for ~1/2 of our total GDP in 50 years. That's clearly impossible. While I can’t predict exactly how the sector will evolve, I’m highly confident that this growth gap will have to close well before my retirement plan concludes.
Solid point. And come to think of it, I think the social security actuaries were speaking of the short-medium term with their 3% above inflation assumption. But given that expenses early on matter the most for overall retirement, I don't think this can be discounted. Since PRC only allows one time frame for medical and social security inflation, the issue is kind of moot isn't it? I suppose the rational response is to slightly raise one's medical expenses but not the full 3%.
My assumptions below, as of Jan 6 2026.
My updated Capital Market Assumptions are documented below as of Feb 11, 2026.
I'm providing this update because I've changed my methodology from a 1-period forecast model to a 3-period forecast model.
Multi-period Model
2026-2030
- S&P 500 - Expected Return Real CAGR% 5.97%
- Inflation - Real CAGR% 2.32%
2031-2045
- S&P 500 - Expected Return Real CAGR% 5.59%
- Inflation - Real CAGR% 2.36%
2046+
- S&P 500 - Expected Return Real CAGR% 5.26%
- Inflation - Real CAGR% 2.40%
~Equivalent Single-Period Model
2026+
- S&P 500 Expected Return
- Nominal
- Arithmetic 9.54%
- Std Dev 17.50%
- Geometric 8.00% (CAGR%)
- Real
- Arithmetic 6.94%
- Std Dev 17.13%
- Geometric 5.47% (CAGR%)
- Nominal
- Inflation
- Arithmetic Avg 2.41%
- Std Dev 1.57%
- Geo Avg 2.40%
Key Sources:
* Equity Risk Premia based on monthly estimates by Aswath Damodaran [ https://pages.stern.nyu.edu/~adamodar ]
* Treasury Bond Yields based on daily updates by US Treasury [ https://home.treasury.gov/resource-center/data-chart-center/interest-rates/TextView?type=daily_treasury_yield_curve&field_tdr_date_value=202602 ]
* Inflation based on monthly estimates by Cleveland Federal Reserve [ https://www.clevelandfed.org/indicators-and-data/inflation-expectations ]
* Standard Deviations based on 2001-2025 averages [MoneyChimp - http://www.moneychimp.com/features/market_cagr.htm ]
Future Revisions:
As a computational purist, I'm not totally happy yet with some of the simplified sources, approximations, and computational methods that I've employed above, so I'll probably make minor tweaks to my spreadsheets in the upcoming months. However, I anticipate t the resulting changes should be small compared to the normal monthly revisions and should be small enough that the differences won't be visible to outside observers.
Examples:
- Estimates of long-term real US GDP growth were sourced and averaged across several AI agents. In the future I plan to source/validate those directly myself.
- I applied the FED's 10-year Inflation CAGR% estimate to my 5-year forecast period. In the future I plan to compute the 5-year CAGR% from the FED's year-by-year inflation estimates.
