July 27, 2026 – Welcome back to a new part of my Safe Withdrawal Rate Series. In my 10-year quest to study safe withdrawal strategies and find ways to hedge or at least alleviate Sequence of Return Risk, I’ve come across a lot of purported “solutions.” Some actually work to at least some degree. For example, a reverse glidepath can improve outcomes. Momentum strategies look promising. But most proposed solutions to Sequence Risk are ineffective (e.g., dividend yield, bucket strategies, small-cap value stocks, etc.). The flavor of the season right now seems to be Risk Parity. My blogging colleague Frank Vasquez has been making the rounds on various podcasts over the last year or so, touting the benefits of Risk Parity, a supposedly brilliant and innovative portfolio construction method that will miraculously increase your safe withdrawal rate in historical simulations. Supposedly up to 5% or more. I’m less optimistic, though. Let’s take a closer look…
Risk Parity Intro
I published a separate post that serves as a deep dive into the mechanics of Risk Parity on July 29. Also, please refer to the PDF file here. The intuitive explanation of Risk Parity is that you diversify away from an equity-heavy portfolio and try to find other uncorrelated assets. Normally, that involves a much larger allocation to nominal bonds but also commodities generally and gold in particular. Other exotic styles, like crypto, come to mind. In addition, one can also throw in tactical asset allocation strategies (e.g., managed futures/momentum) to serve as low- or even zero-correlation assets. The idea of Risk parity is that, first, you aim for a lower portfolio volatility, so hopefully you’ll experience shorter and shallower drawdowns. Second, you also spread your risk more elegantly across the so-called “four economic quadrants” (see the diagram below) when you expand your asset universe. For example, with gold and commodities, you may also capture diversification potential during times of high inflation, where a pure stock-bond portfolio might have suffered. Tony Robbins had a piece on YahooFinance to describe this in more detail.
So, this all makes intuitive sense. Risk Parity is a serious enough proposal to check if it can help us improve the performance of retirement portfolios. Let’s get started…
Risk Parity Simulations – Basics
If you’re an avid reader of this blog, you’ll know that I already simulated Risk Parity portfolios a long time ago when I studied gold in Part 34. The results back then were mostly disappointing. I’ve made some improvements in my Google Sheet since then, so let’s take another look at Risk Parity and see if anything has changed. I will test a variety of portfolios, namely:
- The classic 60/40 portfolio: 60% US large-cap equities (e.g., SPY, VOO, or IVV tickers), 40% US Intermediate Treasury Bonds (e.g., VGIT or IEF tickers).
- My preferred portfolio: 75% equities, 25% bonds, which has a bit more earnings power and, despite the higher short-term fluctuations from the larger equity portion, actually reduces the risk of running out of money during a multi-decade-long retirement.
- The “OG” Risk Parity portfolio, i.e., Ray Dalio’s/Tony Robbins’s All Seasons/All Weather portfolio: 30% equities, 40% long-term bonds (e.g., TLT, GOVZ, or VGLT tickers), 15% short-term bonds/T-bills (e.g., SHY or BIL tickers), 7.5% gold (GLD, GLDM, etc.), and 7.5% commodities (e.g., GSG)
- The Golden Butterfly portfolio: 20% each in US large-cap blend (e.g., VOO, IVV, SPY), US Small-Cap Value Stocks (e.g., VIOV), 3-month T-Bills (BIL, SHY), Long-term Treasury Bonds (e.g., TLT, VGLT), and Gold (GLD, GLDM).
- Frank Vasquez’s Golden Ratio Portfolio: 42% in US equities: 21% each in large-cap growth (e.g., VUG) and small-cap value (e.g., VIOV), 26% in long-term Treasury Bonds (e.g., VGLT), 16% gold (e.g., GLD), 10% in managed futures (e.g., DBMF), and 6% in 3-month T-Bills. More information here.
- Expense ratios for these portfolios are 0.03% for the 60/40 and 75/25 portfolios, and 0.0773% for the All Weather portfolio. The two “Golden” portfolios have expense ratios that depend on how exactly we model the small-cap value portions; more on that below.
Note that historically, Frank has had different allocations for the 10% portion in his Golden Ratio portfolio. He first started with international REITs (ticker REET), but then switched to DBMF, which is a trend-following strategy. The DBMF creates a lot of headaches: We have no backtest for the DBMF fund, so we’d need to make some assumptions here. The easy way out would be to assume that DBMF simply mimics a commodity fund, like the GSG ETF, that would line up very well with the Commodity asset class in my SWR toolkit. But that’s a bad fit. The recent returns of the DBMF don’t look anything like the GSG ETF, by design. The DBMF regularly shifts in and out of different asset classes depending on asset class momentum.
So, I propose the following method to replicate the DBMF: First, I use the actual DBMF returns since 2019 but also backfill the returns to include the simulated returns used on the Tesfolio.io site that Frank uses for his return simulations (Ticker DBMFSIM). I ran a factor model to see if and how much exposure the DBMF has to the trend-following strategy I proposed in my post last year (Part 63). There is quite a bit of correlation between the strategies. The best fit I could generate used 67% of my momentum strategy, -28% S&P 500, -18% 10Y Benchmark Bonds, and +5% Gold (and thus about 73% T-bills to fill the portfolio’s net exposure to 100%). The DBMF and simulated return chart are below. Notice that the replication is a lot smoother than the actual DBMF fund, so we should take the SWR simulations involving this factor with a grain of salt. With the actual DBMF, if it had been available during the entire simulation window, we would have had more sequence risk due to higher volatility and worse drawdowns.
I also like to simulate two different retirement scenarios:
- The standard, 30-year retirement horizon with zero final asset value target, as in the Bengen and Trinity studies.
- A more FIRE-realistic scenario: a 50-year horizon and a 25% final portfolio value target, i.e., you like to leave a bequest worth one quarter of today’s portfolio value, adjusted for CPI inflation. That’s the scenario I currently use for my wife and me.
A few more simulation details: The 60/40, 75/25, and All Weather portfolios are very easy to model in the SWR Simulation Toolkit. The relevant portfolio percentage allocation translates one-for-one into the asset class schema in my toolkit. For the Golden Butterfly and Golden Ratio, there is more explaining to do. The Golden Butterfly has a total of 40% in US equities, but because 20 percentage points come from small-cap value (SCV) stocks, I need to include the Fama-French factor exposures of the VIOV ETF, which are 1.03 on the SMB and 0.57 on the HML. Multiply those by 20% allocated to the VIOV, and we get 20.60% and 11.40% allocated to the SMB and HML factor, respectively. Likewise, for the Golden Ratio portfolio, 21% each goes to the VUG and VIOV. But VUG has SMB=0.04 and HML=-0.36, while VIOV again has SMB=1.03 and HML=0.57. That means the net SML exposure is 22.47% and the net HML exposure is 4.41%. Also note the 10% allocation to the DBMF Sim series. As an additional test case for SCV, I also include a 75/25 portfolio where 55% of the portfolio is in S&P 500, and the remaining 20% goes to Small-Cap-value stocks; more on that later.
Disentangling Risk Parity vs. Stock Picking
Two of the three Risk Parity portfolios rely on Small-Cap Value stocks. In addition, Frank balances the 21% small-cap value stocks with 21% large-cap growth stocks. As I’ve written in a previous post, Part 62, the inclusion of such stock-picking strategies raises a few issues.
Of course, we can always just plug in those Fama-French factor returns. But that comes with two caveats: First, back in the 1920s nobody knew about the efficacy of small-cap value stocks. Unless someone traveled forward in time to read the relevant academic research from the 1980s on the small-cap premium and the 1990s on the value premium, and then back in time to the 1920s, nobody would have implemented that strategy. Second, as I pointed out in a post in late 2024, the reliable outperformance in small-cap and value stocks gave way to rather disappointing returns recently: small-cap stocks have not outperformed large-cap stocks for the last 45 years.
Even worse, over the past 20 years, value stocks have underperformed growth stocks. I am not saying that they will never turn around again, but there has been no statistically significant outperformance in SCV in the 1980-2026 return window. How anyone would want to confidently budget for a vast SCV outperformance going forward is a mystery to me.
So, in light of these SCV headaches, let’s start by simulating the Risk Parity strategies without the stock-picking flavor. Specifically, I like to understand how much of the Risk Parity performance in historical simulations is truly due to better portfolio construction and how much is due to the alpha (fancy finance lingo for outperformance) of your stock-picking strategy. Everything else would be junk science, i.e., throwing a lot of stuff at the wall, trying until something comes out I like, and then declaring victory without understanding why exactly your new strategy works. If Risk Parity is such a brilliant new asset allocation and diversification tool, it should not have to rely on stock picking. After all, VIOV, VUG, and VOO are all highly correlated, so lumping them together in one VOO allocation shouldn’t invalidate any of the Risk Parity qualities.
So, let’s look at the simulations while, for now, ignoring the SCV stock-picking alpha…
Safe Withdrawal Rate Simulations: Risk Parity Only, no Stock Picking
In this base case, I simulate five portfolios: 60/40, 75/25, All-Weather/Seasons, Golden Butterfly Light, and the Golden Ratio Light; i.e., the Risk Parity portfolios use only the broad equity benchmark, with no stock picking included. But I do include the momentum/managed-futures return series in the Golden Ratio portfolio.
The weighted expense ratios for the two Golden portfolios are 0.0680% for Golden Butterfly and 0.0454% for the Golden Ratio. These are quite low because both portfolios rely only on the very inexpensive ETFs. Also, the DBMF (which would have a substantial 0.85% expense ratio) has its expense ratio already baked into the factor model estimates.
Let’s start with the classic Bengen-style and Trinity-Study exercise: 30 years retirement, with a $0 final value success criterion.
- The 75/25 portfolio had a failsafe withdrawal rate of 3.83%. A 60/40 had slightly lower rates, but they were surprisingly close. All Risk Parity portfolios were below that, with safe withdrawal rates between 3.35% and 3.51%. But admittedly, results look a little bit better if focusing on post-1926 cohorts. The All-Weather and Golden Ratio portfolios slightly beat the 75/25 and 60/40.
- Failure rates of the 4% Rule were all above 10% for the Risk Parity portfolios, but still acceptable at 6.13% for the 75/25.
- No person in their right mind would retire with a 5% withdrawal rate. The failure probabilities are all between 39% and 50%.
- Notice that the failure probabilities for both the 4% and 5% WR are higher if the CAPE is elevated (greater than 20) than in the unconditional case. For example, the All-Weather portfolio has a 6.33% failure probability overall but a 14.48% failure probability when the CAPE is high. This flies in the face of the prominent claim that Risk Parity makes you independent of macroeconomic shocks and equity valuations. Apparently, the Risk Parity portfolio cannot hedge out all the risk from overvalued stocks!
And next, the results for the 50-year horizon, with a 25% final portfolio value cushion:
- The 75/25 has a safe withdrawal (failsafe) rate of 3.24%. 60/40 is markedly below at 3.01%. All Risk Parity portfolios are terrible, ranging from 2.45% to 2.71%.
- A 5% withdrawal rate would be unacceptable and almost certain ruin for all five portfolios, whether simple 60/40, 75/25, or the three Risk Parity portfolios. Even 4% withdrawals would be way too aggressive and lead to very high failure probabilities.
- Failure probabilities rise when the CAPE is higher. Even for the Risk Parity portfolios.
Without SCV, the Risk Parity strategies are all pretty much nothing-burgers: suboptimal retirement portfolios. The All-Weather/All-Season portfolio is thus retired (pardon the pun) as a completely useless portfolio for my retirement. Sorry, Tony Robbins and Ray Dalio! But there is still hope for the “golden” portfolios with some stock picking alpha, so let’s keep simulating…
Safe Withdrawal Rate Simulations: Risk Parity plus Stock Picking
Because the GB portfolio now relies on the slightly more expensive VIOV fund, I adjust the expense ratio to 0.082% for the Golden Butterfly portfolio. The GR portfolio uses the VOG and VIOV funds, so the expense ratio is now 0.0601%. When adding the stock picking, I consider three different cases in my simulations that I can all model with my SWR Toolkit, as described in Part 62:
- No alpha from HML and SMB. I keep the HML and SMB factors in the simulations, but I use the HP-filtered returns to shut down the historical alpha to a zero average. The idea here is that both factors are now well-known, have essentially zero correlation with any macroeconomic risks (like growth and inflation), and thus no longer demand nor deserve a risk premium. If we simulate how today’s retirees might fare if historical return patterns repeat, we’d assume that the SCV is over and switch it off with the HP-Filter, while keeping the rest of the return patterns, i.e., equity and bond risk premiums alive.
- Same as option 1, but I do add an annualized alpha of 0.7% to both the HMB and SMB factors. For the VIOV ETF, that would generate a weighted outperformance of just above 1.12% per year. That’s not bad. Paul Merriman once told me he’d be happy if SCV outperforms by 1% every year going forward.
- Historical Fama-French SMB and HML returns. As I pointed out, the Fama-French factor returns were amazingly attractive over the first several decades, but have since moderated. It’s unlikely that we’re going to repeat the impressive returns pre-2006 (for HML) and pre-1982 (for SMB). So, use the simulation results with a grain of salt.
Let’s start with the Golden Butterfly portfolio. I first report the simulation results for the 30-year horizon. To declutter the tables, I now include only the 75/25 baseline and the GB portfolios with the three assumptions about the SCV factor returns.
- Failsafe withdrawal rates are between 3.33% and 3.44% for the GB portfolio, far below that of a simple 75/25. True, you could do a little bit better when a) using the raw HML and SMB returns and b) looking only at the post-1926 era with a 4.08% withdrawal rate. But you’re still miles away from a 5% safe withdrawal rate!
- Also notice that while Risk Parity with historical SCV patterns looked really great for the 1929 cohort (5.92% vs. 3.85% in the 75/25 portfolio), you now introduce another headache, namely the 4.08% failsafe rate for the 1930s decade (for the 1937 retirement cohort) and the 4.28% rate during the 1940s. So, even though you solved the initial Great Depression stock market fall, you then introduce worse results in the subsequent events. It’s like squeezing a balloon.
- Clearly, the Golden Butterfly portfolio performed well during the 1960s and 70s, almost by construction because it was tailored to perform well during the inflationary post-1970 period.
- Intriguingly, two of the GB portfolios had lower failure rates for a 5% WR when conditioning on CAPE>20 than in the unconditional case. However, the 4% Rule failure rates do show the familiar pattern where higher equity valuations cause greater failure rates.
- Regardless, all failure probabilities of the 5% Rule are unacceptably high for most retirees.
Moving on to the more FIRE-appropriate 50-year horizon:
- Golden Butterfly portfolios had inferior safe withdrawal rates, between 2.4% and 2.57% during the entire simulation period. Even focusing only on the post-1926 era, and assuming the raw SCV data, you stay below the failsafe of the 75/25.
- The failure rates of a 5% withdrawal rate are all insanely high: about 67% for 75/25 and almost certain ruin for the Golden Butterfly portfolio: between 72% and 82%.
Next, the same tables for the Golden Ratio portfolio, starting with a 30-year horizon:
- The Golden Ratio portfolios had failsafe withdrawal rates in the 3.49% to 3.58% range, much lower than the 75/25 portfolio. But admittedly, post-1926, the results look better. The best-case scenario would be a 4.22% withdrawal rate for the most optimistic Golden Ratio portfolio. Not bad!
- Quite intriguingly, the two GR portfolios with the SCV raw returns did better in terms of failure probabilities under the 5% withdrawal rate.
- Just like the Golden Butterfly, the Golden Ratio portfolio performed well during the 1960s and 70s, again by construction. It was backfitted to optimize the portfolio stats during the 1970s, so by extension it also solves some of the 1960s headaches.
Finally, the 50-year horizon stats:
- Over the entire simulation period, the Golden Ratio portfolio performed very poorly, with safe withdrawal rates between 2.66% and 2.78%, almost 50 basis points below the 75/25. That’s a 14%-18% lower retirement budget than the 75/25 portfolio.
- By “scrubbing” the data a little bit, i.e., looking at only the raw data SMB and HML and only post-1926, you get to a 3.29% SWR, slightly better than the 75/25 portfolio.
- The Failure Probabilities of all Golden Ratio portfolios were terrible at 50%+. Not even the 1970s produced a 5% safe withdrawal rate, but you got close at 4.87% in the ideal case with raw Fama-French returns.
Safe Withdrawal Rate Simulations: Stock Picking, but no Risk Parity
If you so enjoy small-cap value stocks and you believe that we’ll have a renaissance of this style again, you don’t need Risk Parity to get that benefit. Risk Parity doesn’t have a monopoly or patent on Small-Cap Value. Paul Merriman and Larry Swedroe certainly wouldn’t think so. Thus, let me do the same exercise for the simple 75/25 portfolio: I replace 20 percentage points of the S&P 500 equity portion with small-cap-value stocks (e.g., the VIOV ETF). As before, I simulate portfolios under the three alternative SMB and HML return assumptions, i.e., HP-filtered, HP-Filtered plus 0.7% alpha, and historical (=raw) returns.
Here are the results for the 30-year retirement:
- Not surprisingly, the HP-filtered SMB and HML series don’t produce any gains. Quite the opposite: adding more risk with zero extra return slightly decreases the failsafe. But you also slightly decrease the failure probabilities. So, you make the worst outcomes slightly worse, but alleviate the not-so-bad outcomes. Interesting result!
- The SCV with a 0.7% annualized alpha (about 1.12% extra returns estimate for the VIOV ETF) does almost as well as the 75/25.
- Finally, the SCV raw returns provide some relief from Sequence Risk. You increase the SWR to 3.94% (all cohorts) and 4.13% (post-1926). The failure rates of the 4% Rule improved with the 0.7% alpha and raw returns SMB+HML. If you’re a strong believer in the small-cap value factor, I won’t judge you for mixing this into your retirement portfolio.
And next, the results for the 50-year horizon, with a 25% final portfolio value cushion:
- The 75/25 with the most optimistic SCV assumption (raw return) performed quite well. The failsafe rate is now 3.31% (unconditionally) and 3.74% if considering only post-1926 data. That’s a significant boost relative to the 75/25 portfolio. But you know what? That’s also a significant boost relative to the most optimistic SCV assumptions under Golden Butterfly (3.16%) and Golden Ratio (4.13%).
So the lesson learned: The Risk Parity strategy performance poses a truly intriguing catch-22: Either you do only Risk Parity, in which case all return stats suck, across the board, whether All-Seasons, Golden Butterfly, or Golden Ratio. Or you hope and pray for the miracle of Small-Cap-Value stocks, in which you do have a slight edge over the classic 75/25. However, in that case, all the Risk Parity plus SCV still underperforms a 75/25 plus SCV. It’s almost comical how hopeless the case for Risk Parity is.
Objections from the Peanut Gallery
I already know how some Risk Parity fans are going to attempt to dispute my research, so here are my replies to the standard fare of the cheap shots from the Peanut Gallery:
“Portfoliocharts has better-looking simulation results!”
Yes, but that’s because the simulation window there starts only in 1970, so that site conveniently ignores some of the historical worst-case scenarios: 1907, 1911, 1929, 1937, 1964, 1965, and 1968. PortfolioCharts, while clearly visually very appealing with beautiful charts and tables, is really only a garbage-in, garbage-out affair. If you want your portfolio to do well in PortfolioCharts.com simulations, you will seek all the usual suspects: small-cap stocks (which had a fantastic run in the 1970s), gold (which also had a nice run after gold ownership was legalized in 1975 and the pent-up demand caused a great run and nice diversification benefits during the terrible 1970s malaise), and commodities. But that will overfit your asset allocation. If the future looks different from the 1970s, you could have some very frustrating outcomes. It’s much safer to pick an allocation that’s been robust to all of the various market events, not just the 1970s.
“SCV is such a crucial ingredient in Risk Parity, you can’t disentangle Risk Parity and SCV!”
If you think that, you demonstrate that you don’t understand the mechanism behind Risk Parity (or finance in general). The funds VOO, VIOV, and VUG are all highly correlated, much more so than the inter-asset-class correlations, say, between stocks, bonds, and gold. Lumping the equity funds together in one simple fund will not undo the benefits of Risk Parity, which relies on spreading risk around the maximum range of asset classes. The fact that Ray Dalio’s All-Weather OG Risk Parity portfolio doesn’t mention SCV tells you that you can indeed disentangle the two items.
“We should ignore the pre 1926 simulation results because the Fama-French database didn’t start until July 1926!”
That is a very weak argument for several reasons. First, I have the simulation assuming 1.12% annual outperformance of the SCV equity style (i.e., scenario 2 with 0.7% alpha in the SMB and HML), and those simulations are still disappointing. Second, a portfolio that started in 1919 still has most of its history subject to the HML and SMB factors (especially in the 50-year simulation), even though there were a few zeros over the first few years. As long as all the other return series are available, we should run the simulations and assume that the 1900-1925 equity returns were just the index return. If Fama and French couldn’t find the pre-1926 data with today’s research methods, the average retail investor wouldn’t have found the stats to put together a small-cap value portfolio at that time either, so it’s a natural assumption to set the HML+SMB to 0% pre-1926. Lastly, I can guarantee you that if the shoe were on the other foot, i.e., the 1900s and 1910s results had looked much better for Risk Parity than for 75/25, the influencers would be touting that fact and would insist on including the earlier period, citing the very reasons I spelled out above. Besides, even the post-1926 return stats aren’t that impressive for Risk Parity.
Addendum: July 27, 7:30 AM
People in the comments section raised a few issues:
- Rebalancing: My portfolios are rebalanced monthly. As I showed in Part 39, there is no discernible alpha (excess return) from varying that frequency: If you go to quarterly or annual rebalancing, you may slightly increase or decrease your SWR. It’s a crapshoot. It’s mostly the luck of the draw when you rebalance around the turning points. Sometimes you get lucky with less frequent rebalancing. But then you’d likely also get lucky with a 75/25 portfolio and infrequent rebalancing. So, the rebalance risk would apply to both RP portfolios and the 75/25.
- Frank’s Portfolios aren’t really Risk Parity: I know. I will point that out in the upcoming “How To Lie” post, as issue #3. Unfortunately, going more toward Risk Parity will only worsen the outcomes. I show that in issue #7: Risk Parity portfolios are inherently inefficient because they only look at variances and covariances, not average returns, so the more you go toward Risk Parity, the worse your Sharpe Ratio gets.
- Leverage: None of the portfolios I presented here have leverage. Introducing leverage would be a terrible idea. You increase your average returns but also increase your volatility, thus worsening your Sequence Risk. I simulated Frank’s “Aggressive 50-50” portfolio (33% each in UPRO and TMF and 17% each in PFFV and VGIT), and the safe withdrawal rate dropped to below 2% in the 1960s for both 30- and 50-year horizons. The reason for this terrible performance is that, for example, the UPRO increases volatility 3x, but excess returns over the risk-free asset grow much more slowly than 3x due to trading costs, churning, whipsaw effects, etc. More leverage means worse Sharpe Ratios and worse retirement outcomes.
Summary
There you have it. I’m not very convinced that Risk Parity is a useful strategy for retirees, and I believe it’s even more dangerous for people in the accumulation phase, but that’s a topic for another day. Pure Risk Parity without stock-picking alpha in the form of Small-Cap Value stocks is largely useless. The All Weather/All Season strategy is systematically much worse than 75/25. The other two strategies, GB and GR without SCV, are potentially competitive with 75/25 over a 30-year retirement horizon and for retirees who are willing to deplete their portfolio if you ignore the pre-1926 era. A 4% withdrawal rate isn’t completely safe, but some retirees with flexibility could potentially try it.
The GB and GR portfolio with SCV had no failures of the 4% rule in the post-1926 simulations, but note the caveats: there is no guarantee that the SCV premium is going to repeat again, now that all smart money chasing easy alpha knows about this flavor. I also want to reiterate the same caveat from above: the DBMF had worse return characteristics, i.e., more volatility and more extreme drawdowns than the momentum strategy I used in the simulations. The actual Golden Ratio performance may not be as good as in the simulations because of that. With all those caveats, I would not increase my withdrawal rate over that of the simple 75/25 portfolio if using the Golden Ratio portfolio.
Over a longer horizon (50 years) and with a modest bequest target, the All-Weather strategy is hopelessly inferior to 75/25 and 60/40, and even the two Golden portfolios are significantly behind the simple 75/25 portfolio. Under no circumstances would a 5% withdrawal rate be safe over the longer horizon. The failure rates of, say, the Golden Ratio portfolio would have been around 60%. So, not even the best possible stock-picking scenario, using raw returns for HML and SMB, would have been anywhere close to safe for a retiree. The Golden Butterfly failure rates would have been between 72 and 82%. Only a certifiable charlatan would recommend this strategy to unsuspecting retirees and claim a 5% withdrawal rate is safe. In fact, even a 4% withdrawal rate would be pushing your luck over 50 years.
In my professional opinion, Risk Parity as a retirement portfolio is a scam because it hides a completely different asset allocation trick, i.e., backfilling your simulation data with the Small-Cap Value stock-picking flavor. The purported outperformance of that stock-picking alpha creates all the benefit. So, the great irony is that even in the few instances where Frank’s Golden Ratio portfolio or Tyler’s Golden Butterfly portfolio looked competitive, it’s not because of the Risk Parity part. It’s entirely due to the SCV part, while the Risk Parity element of your portfolio actually hurts you at the margin. Let’s repeat, everyone…
The parts of the Golden Butterfly and Golden Ratio simulations that did well did so not because of Risk Parity, but despite it.
I like the diagram below showcasing the Risk Parity shell game. While it’s true that the Golden Ratio portfolio had a decent safe withdrawal rate over 50 years (3.29%, post-1926), even slightly higher than the 75/25 portfolio (though a far cry from the 5% claim), looking at the marginal(!) impacts of making one change at a time, you notice that SCV always improves the results and Risk Parity always reduces your Safe Withdrawal Rate:
Thus, intriguingly, even if you believe that SCV will work again and will work as well as in the 1920s through 1980s, however insane that assumption may be, the best path forward is not Risk Parity: Rather, you’d simply still prefer a 75/25 where the 75% equities consist of 55% VOO and 20% VIOV. Not that I would recommend that, but conditionally on your SCV wet dreams, you should just do SCV only and ignore the Risk Parity part.
So, long story short, I warn my readers, in the sharpest possible tone, to stay away from Risk Parity. It’s an inferior and suboptimal allocation method. People who push this approach and claim the safe withdrawal rate is 5% have cooked the books to come to that conclusion. I will have some more fun facts about the false advertising and misunderstandings that the Risk Parity crowd likes to spread. It will appear in a future installment of my “How to Lie with Personal Finance Series.” Stay tuned!
Please leave your comments and suggestions below! Additionally, be sure to explore the other parts of the series; see here for a guide to the different parts so far!
Title Picture Credit: WordPress AI + ERN edits
Thank you very much for this very interesting analysis… Yet I believe you are barking up the wrong tree.
In my opinion, the main culprit in your risk parity allocations, driving the SWR down is not diversification away from stocks – it is that all these portfolio have recorded a lower volatility (and performance) vs. the traditional portfolios you are comparing them too. And as we know, LT CAGR is one of the most important drivers of SWR. I believe this is also why SCV improves the picture in your examples.
Now try to add a bit of leverage to your portfolios so as to match the 60/40 or 75/25’s volatilities. I expect the picture to vastly turn in favor of the other allocations. Even better: push the volatility to become higher than the 60/40, and I believe the RP portfolios will all trounce the 60/40.
Also, these portfolios are not really risk parity portfolios. Shifting the allocation to a true risk-parity one should improve the risk-adjusted returns… And thus, the SWR once the volatility is brought back to the same level as the traditional portfolios.
François
Leverage will make things worse. You increase the drawdowns in the bear markets. Also, the leverage available to retail inventors through UPRO, TMF, etc. is not that good. If you factor in the inefficiency (high expense ratio, t-costs, churning, etc.) you will make things worse not better.
Correct. That is issue #3 I will mention in my “How To Lie” Post: none of Frank’s portfolios are RP.S Stay tuned. But, alas, true RP will not make the issues go away either.
False. That’s issue #7 in my upcoming post. RP is inherently inefficient from the risk-adjusted return perspective. The more you move away toward RP the more inefficient you get, even more inefficient than Frank’s portfolios.
To your point on leverage making things worse: that’s not what the data suggests, to the best of my knowledge at least
One simple example, with a RP portfolio of stocks/bonds/trend/gold backtested since 1987: https://testfol.io/?s=lfXOabgKCXf
30-year SWR (failsafe rate):
– 75/25: 5.75%
– RP portfolio (unlevered): 6.02%
– RP portfolio (levered to reach the 75/25’s vol): 8.27%
I get a comparable outcome through Monte-Carlo simulations. Of course as you rightfully mentioned, this is provided you can access cheap leverage (ie, box spreads)
Looking forward to your upcoming posts on the topic, as I am looking for arguments to convince me not to go all-in on RP.
I just see that you added a NB to your article partly addressing the topic of leverage. I believe the “agressive” UPRO/TMF mix you are referring to has a very low SWR because…It is indeed way too volatile! (31% SD when I backtest it, almost 2x a 100% stock portfolio).
I was more thinking of leverage as a tool to reach the same volatility as traditional portfolios, perhaps slightly more, not to double it. All things in life are good, provided they are not consumed in excess 🙂
But if done moderately, it does work!
Thanks for your comments,
François
At the margin, you’re still left with replacing some of the equity funds with UPRO and or bond bunds with TMF, and the leverage causes the volatility to go up in line with the higher beta, but the average return goes up much less due to the major drag in the UPRO and TMF funds (see the ETF factor exposures in the SWR sheet, especially the atrocious alpha of UPRO, TMF).
I have been trying to increase the SWR using leverage in Ern’s spreadsheet and have not had a lot of success. If you only go back to 1926 then gold helps, but there is not much value in leveraging the portfolio. For my personal situation (40 year horizon) a 65/25/10 stocks/bonds/gold works best. That looks a bit like a risk parity portfolio but its higher in equities. I tried lowering the equity ratio and leveraging the entire thing (using futures on the bonds and gold) but it didn’t really help.
That’s one single example and it’s not representative. With leverage, especially with the wildly overleveraged portfolios like Aggressive 50-50, etc. you would have increased your Sequence Risk.
Was your risk parity analysis unlevered? I skimmed through the post but I didn’t see any mention of leverage or borrowing. I personally invest in the accumulation phase of a mix of 2x HFEA(hedgefundie’s excellent adventure) (120% VTI and 80% TLT targets) and 3x HFEA/hell on fire (180% VTI and 120% TLT – ie the classic 60/40 UPRO/TMF.) I’d really appreciate a repeat analysis of leveraged various risk parity and on SWR rates.
I run about 25% annual contributions a year into the 3x variant. The 2x variant is the rest of my portfolio. I started a high frequency trading firm a year ago that trades ontop of 2x strategy on portfolio margin. I “bucket” both strategies where I never rebalance across them. I’ll hold both until age 67 then figure out derisking. I’ve done my own studies on this strategy and safe withdrawal rate analysis. The intermediate term treasuries version of the portfolio does well on 1970+. Long term 1985+. You can argue one epic long bond bull market from 1985+ but my stance on it is long term treasuries had structural issues before that date because the older issued 20 year bonds in 1970s were callable after a 5 year period which removes all the upside to long term treasuries 🙄. The 30 year didn’t launch until the mid 80s.
If we focus on the 2000 lost decade bear market withdrawing 4% SWR gets insanely dicey for 60/40 and 100% stocks (forgot how 75/25 did.) However 4% SWR on 2x 60/40 coasts right on through 😍. Then I’d personally would never retire on 3x 60/40 but some decades has up to a 11% SWR😅.
The vast majority of my wealth is now in a taxable account. I also pivot to doing a bit of an international allocation to help increase 1970s after inflation resiliency by a tilt to international – 30% VXUS 90% VTI and 80% TLT. I also squeeze another 1%~ from doing 67% of my TLT target (53.6% on 2x) duration matching EDV (20-30 year zero coupon STRIPS.) I’ve found so far whenever gold does well so does international stocks due to the dollar weakening. So this is a forward-looking allocation but it’s holding up well for me.
So yeah Karsten I’m with you that unlevered risk parity is garbage but also honestly if you’re only comparing it unlevered – of course it’s garbage as you’re not getting enough risk offsetting movement unleveled.
Leverage will make things worse. You notice that in Frank’s simulations on his site: While the Golden Butterfly and Golden Ratio portfolio are holding up all right in his simulations, the leveraged portfolios got totally clobbered since 2020. More leverage means more drawdowns and worse Sequence Risk.
Also: a lot of the levered funds (UPRO, TMF, etc.) have terrible drag built in, not just from expense ratio, but also t-costs, churning, whipsaw risk, etc.
Thank you Karsten for this thorough review that was sorely needed. I’ve tried studying the topic myself but between the 12 year old playing with a sound board on the podcast and Frank barking like a rabid dog at anyone on the ChooseFI Facebook group it was difficult to assemble an understanding of the topic. Looking forward to your “How to Lie with Personal Finance Series” and I’ll keep my portfolio away from the risky parity “troubled waters” for now.
Yeah, Frank can be a bit caustic in the Facebook comments sometimes. I’ve stopped going to the forums there a long time ago.
The ChooseFI forums have largely moved to the ChooseFI community website and app (though I’m sure stragglers remain in the Facebook forums). Unfortunately, Frank remains caustic at times there as well.
Gotcha! Thanks for sharing!
I didn’t see it explicitly mentioned, but how do you account for rebalancing over the 30 / 50 years? If I understand correctly, part of risk parity is to lower and shorten some of the worst downturns. Say in a recession, the LT govt bonds increase and now you are selling and rebalancing into stocks?
All my portfolios are rebalanced monthly. There is no clear directional effect from rebalancing monthly vs. quarterly vs. annually. I showed that in Part 39 of the series.
Good idea to exclude SCV from the RP portfolios to look at the portfolio construction benefits, thank you.
I don’t get testing gold-heavy portfolios using historical performance before 1970, though (i.e., in a pre-fiat money system). I would believe the results more if you had replaced the historical returns of gold before 1970 with a simulated series (perhaps something with a real CAGR of 0%, matching its volatility and correlations). This is especially true since you already simulate other assets.
Also, I have a question: does your simulation choose which specific assets to sell based on some rules? Or does it just decrease the total value and multiply it by the target weights?
Anyway, that was an interesting read!
Thanks!
My simulations assume rebalancing. You buy/sell shares of assets that are under/over-represented in the portfolio. So, yes, I “just decrease the total value and multiply it by the target weights” as you suspect.
True, between 1933 and 1972, gold returns were negative. So, assuming a CAGR=0 would have helped portfolios with gold for the 1929 cohort. But: you would then hurt the 1970s cohorts because the rebound in gold prices in the 1970s would have been much smaller. Gold would have helped much less in the 1960s and 1970s retirement cohorts because they would have bought gold at much higher cost basis. It’s like squeezing a balloon!
Cumulative Total Returns, Real CPI-Adjusted. Stocks, vs. Commodities vs. Gold.
Karsten, thanks for the new post. It is good to read differing viewpoints of any portfolio strategy, as it gives all of us a better understanding of the potential pitfalls.
I understand your point regarding SCV that a retiree should not rely on one factor to drive results for the SWR – makes sense.
I believe it was shown in Bengen’s 2025 book that holding the entire equity market more evenly (I believe he held 11% each of large caps, mid caps, small caps, micro-caps, and international…so 55% equities) resulted in an improved SWR as opposed to just holding large caps. In making that change he improved his SWR results from 4.2% to 4.7%. Holding a cap weighted overall market fund gets you a small step closer to Bengen, but because it is weighted the S&P 500 is just too dominant to the total, which waters down the potential benefit. I don’t believe you have tested that yet?
In one of your previous posts of this series, didn’t you conclude that holding 5%-15% gold in a portfolio improved the SWR results? Are you saying that is no longer the case? I also struggle with trying to evaluate gold prior to 1970 due to the gold standard. Did you just put gold in at a 0% return over all those decades when evaluating the portfolios?
Playing with Karsten’s Google Sheets calculator, there is definitely a band from 6-11% (depending on time horizons, pension-like incomes, and desired equity allocations) where gold does have a Maxi-Min effect of increasing the failsafe SWR rate (though at a slight penalty to SWR in better market conditions that happen most of the time). This is still the case, but I haven’t experimented with forcing pre-1970 returns on the asset class down to zero.
Most of the historical overperformance of SCVs is likely a relic of the past, and decades from now will replaced with silly stuff like people cherrypicking overperformance of AI stocks, or mega-caps during current era as a portfolio scheme justification… but information is relatively available, and easy alpha is hard to find when anybody with access to an LLM can write a trading bot to scrape out marginal gains.
Yes, good reply. I also replied separately below before seeing this excellent comment.
Correct: we should not rely on potentially unreliable historical factor returns.
But the same criticism applies to Bengen’s new work too: You don’t hold the entire market more evenly. You hold the entire market evenly if you hold VTI. Any mixing in of small-cap and micro cap will undo that evenness and create concentration and also less diversification and more risk. See my post from 2025, especially “Lie 2: More ETFs = More Diversification”
https://earlyretirementnow.com/2025/12/10/how-to-lie-with-personal-finance-diversification/
I concluded that gold could provide some diversification benefit. But there is caveat that gold was not tradable between 1933 and 1975. The simulations don’t lie: the 1960s cohorts and 1972/73 would have benefited from gold. If you could have bought it at those prices when it wasn’t available to purchase for retail investors.
Hi Karsten,
Thank you for adding this analysis to the conversation. I get quite a few financial advising clients who are persuaded that they want or need a small cap tilt, value tilt, or risk parity portfolio due to hearing about it over and over from Frank Vasquez, Paul Merriman, and others.
In some cases I am able to persuade them to let it go, in others cases they insist on keeping it and I have to adjust other elements of the plan to accommodate the lower expected return.
I thought your article “My thoughts on Small-Cap and Value Stocks” put this to bed back in 2019! The flatlining of the SCV since the early 1980s and the decline of the HML since the mid-2000’s illustrates this very clearly to me, but the additional rigor in this article adds substantially to the evidence that this tree isn’t worth barking up.
I think there is a pattern where people in the accumulation phase, the so-called “boring middle”, are looking for the next shiny object, technique, or concept that will catapult them to FI.
To some extent this is great because it stimulates ongoing learning, but the effect is that portfolios get overly complicated and are not well maintained, especially post-FI.
As you showed, a 75/25 portfolio (or rising equity glidepath) with a withdrawal rate adjusted for future cashflows (e.g. mortgage paid off, Social Security starting, etc.) would do the job not just well, but better than a 10-fund portfolio, SCV/HML, risk parity, etc.
Thanks Aubrey, very valid point! I noticed too that people have the natural tendency to tinker with their finances. Especially in the “boring middle” as you describe it, where people get impatient and want to force the higher returns to get to FIRE faster.
Karsten, Thanks again for providing much needed rigor/scholarly analysis and broader context (e.g. FIRE timeframes, longer analysis windows, etc) to Risk Parity!
You’ve applied this scholarly rigor to myth bust many assertions in social media regarding SWR, and I very much appreciate that, as it is badly needed.
1) FYI, Your URL link to “Golden Ratio Portfolio” is broken – because it duplicated the URL.
2) Wouldn’t small cap still justify a risk premium simply because small companies are inherently riskier than large companies (higher failure rates, harder to achieve profitability)?
3) Lately I’ve been hearing a refinement of SC investing approach to say a quality (profitability) filter needs to be applied along with small to capture a premium. Bengen admits to validity of criticism that a lot of small companies are remaining private now, so investors have less access to what may be the best small cap investments. So it might follow (I’m hearing this elsewhere, not Bengen): given that SCV premium factor is now well known, overcoming it may require filtering out the many poor performing small companies. Have you done any analysis adding a quality/profitability filter along with SC? If not, could you to see if that saves the SML premium?
Thanks again!
On 2) My personal tendency is to treat these as already priced in, not unlike Bond Funds. Some will fail, but that’s baked into the pricing on aggregate, so over time that market equilibrium will revert to somewhat of a historical mean.
3) While I suspect LLM-type tools scraping rapidly through SC data and producing higher quality estimates on expected returns can normalize premiums on pricing for those, especially as private equity continues to voraciously hunt down anything that smells like the alpha from the last couple of decades, if anything the end result is going to be more normal-looking returns that more closely index broader segments of the market due to more shared fundamentals.
2) According to the charts on Yardeni.com, the small cap S&P 600 trades for a forward PE of 15.9 compared to the large cap S&P 500 at a forward PE of 19.9. Thus, there is a huge -20% discount being applied to small caps right now. Maybe part of this discounts represents a risk premium? But still, even small caps these days have market caps in the hundreds of millions, and at that scale I don’t think they go bankrupt often.
3) If we wanted to apply a quality filter, the easiest way is to simply use the S&P 600 instead of the Russell 2000. The difference between these two small-cap indices is that the S&P 600 only includes profitable companies, whereas the Russell 2000 contains lots of money-losing names.
However, we should temper our expectations because the two indices have moved in almost lockstep together over the years.
2: that PE of 15.9 vs. 19.9 is not free money. Again, the market is smart enough and efficient enough to gauge the 1y forward and all additional future earnings and assign a correct value.
3: And amazingly, the IWN Russell 2000 Value is one of the best performing SCV funds over the last year, much better than DFSV.
1: Thanks! I fixed the broken link!
2: Between SC and Value, I’m surprised that SC doesn’t work anymore. It could be a legitimate risk premium because small companies face more macro risk. But alas, SC hasn’t worked since the Banz (1991) paper.
3: I’m troubled by ever more backfitting. They say the same about SCV, where a quality screen will weed out the value traps. Makes sense. But how come SCV worked so well before without these screens? Also, some of the “dumb” SCV funds without such screen recently performed better than the new smart funds, e.g., IWN outperformed DFSV and AVUV over the last 1Y. I’m afraid that these new flavors are just more overfitting and backtest-optimization traps that explain the most recent trend, but then fizzle in the out-of-sample case.
I just have to thank you for the vindication of using a 50-year horizon with a 25% target.
I don’t want to jump to the grey part of the Rich/Broke/Dead chart because I’m stressed about having less than 7 years of nominal spending funds in my portfolio.
Also, the practical upshot of these targets is a way more reasonably risk-adjusted scenario where the 99% success rates line up well on virtually every simulation I run.
Yes, so true, that 25% cushion adds to retirement confidence!
Maybe the goal of owning counter-correlated assets is a good one, but the problem is the assets aren’t actually counter correlated.
E.g. portfoliovisualizer.com/asset-correlations shows five-year monthly correlations of 0.82 between SPY and VIOV, 0.37 between SPY and GSG, and and 0.25 between TLT and GLD. In medical or social sciences research, those would be considered strong positive correlations between the variables. Only between TLT and GSG do we see a strongly negative correlation of -0.40, so perhaps a true RP portfolio would only have these two assets!
So maybe these diversification schemes are failing because they’re buying correlations instead of counter-correlations? And perhaps assets that look counter-correlated in one period of time swing into correlation in other times? E.g. when rates fall SPY and TLT may usually go different directions, and then when rates rise, both SPY and TLT are hurt together.
For these reasons, I use options instead of asset allocation to firmly define the possible range of outcomes for my portfolio.
Consider, for example, a strategy of using a 2 year costless collar on SPY, with perhaps 15% maximum upside and -13% maximum downside, plus dividends, rolled annually (I’ve observed costless spreads in this range for several years). This portfolio would have a beta around 0.5 and be immunized against SORR events that do not look like -13% losses dragged out over several years.
With such a portfolio, we could re-imagine 1974 as involving half the losses. Or merely experiencing corrections in 1930, 1931, and 1937 instead of utter portfolio destruction.
To model this, you have to create your own dataset. If there is no rolling going on, just convert all returns in excess of the collar’s range to the biggest/smallest number within the collar’s range. E.g. 1931’s price return of -47.07% becomes -13%, and 1980’s price return of +25.77% becomes +15%.
Because the collar is costless, there’s be no additional expenses (maybe $20/year commissions) if the options are held to expiration.
To model a rolling strategy, or algorithms such as dropping the hedge after a period of negative stock returns (both recommended), you’d have to calculate option values from historical data, which is beyond my skill set. Perhaps there are calculators on the web that generate imaginary option prices from the past?
I really wish I could know the SWR’s of such a strategy, which uses mathematically and contractually certain counter-correlated assets that work in any circumstances, instead of guesswork about asset classes maybe offsetting one another.
Karsten,
What are your thoughts about TIPS these days? The 10 year TIPS yield is around 2.43%, which is a level we haven’t seen on a sustained basis since 2002.
These would have definitely helped the 1970s cohort, had they been available. As it is, there’s no good backtesting history.
And 30y TIPS at 2.96%. That could afford you a real WR of almost 5% over 30y, though with certain asset depletion.
Yes, for some retirees, that would be the best option. Not a good option for early retirees with a 50y horizon and 25% bequest target.
All good points. I dabble in options and found that short-term options are overpriced and it’s best to sell. See my separate series on that: https://earlyretirementnow.com/options/
I’m with you in that longer-term options display fairer prices. There is also the great feature that you can likely time the market and wait for the annual recurring lull period when everybody gets complacent and the IV is low and insurance is cheap to buy.
Careful, though with the supposedly costless collars. There is negative skewness, so need to sell the upside with less % OTM than you can buy protection on the downside. So, for example, a costless collar may be -20% on the put side and only +13% on the call side.
But in general, I like the collar idea.
“Careful, though with the supposedly costless collars. There is negative skewness, so need to sell the upside with less % OTM than you can buy protection on the downside. So, for example, a costless collar may be -20% on the put side and only +13% on the call side.”
The skew is negative if you collar for weeks or months at a time. However, I’ve found the skew goes the opposite way when going waaaay out in time (1 to 2.5 years). This is because it is much more likely for indices to be up the farther out in time we go.
E.g. SPY options expiring 12/15/28, or 2.38 years from now.
Current price is 741. If I pick 75% of that price as my put strike, I’ll buy the put at the 555 strike. This is $186 below the current price. I’ll pay about $22.88 for that option. If I pick 125% of today’s price as my call strike, I’ll sell the call at the 930 strike. This is $189 above the current price. I’ll pay about $34.20 for that option.
The net $11.32 per share credit is mine to keep, and I got $3 more upside than downside. The net 14.32 tilt is worth an additional +1.93%.
So my best case upside if I hold until expiration is 930 plus 11.32, or 941.32. That’s 27% higher than today’s price.
My worst case downside exit price is 555 plus 11.32, or 566.32. That’s -23.6% lower than today’s price.
This is at VIX=18.2 . Expect a spread more like 5% when VIX is below 16. As you noted, the price of insurance gets better when IV is low.
Going out in time is key. A collar with the same strikes expiring in December 2026 would cost a debit instead of generating a credit!
Exactly: That happens because what you consider at-the-money rises due to a positive risk-free rate of return. Compared to that neutral ATM level, there is still negative skewness. But it looks like great insurance compared to today’s index level, I agree.
Separating small/large and value/growth is not “stock picking”. [parts of the comment were redacted. this is a family-friendly blog. KJ]
So, picking certain stocks based on their underlying characteristics is not stock picking? You are saying that someone picking stocks is not a stock picker. Got it!
I recommend my popular post from two years ago: https://earlyretirementnow.com/2024/12/02/small-cap-value-stocks-diversification-or-diworsefication/
You should re-read it. Everybody I know found it awesome.
But you’ve also picked the stocks in these portfolios based solely on their underlying characteristics? the S&P500 has rules about domicile, market cap, profitability, public float, liquidity etc. Is this not stock picking too?
Kind of silly argument. He generally uses S&P500 and whole market index interchangeably because the difference between the two in these kinds of calculations are negligible.
Hi Karsten, thanks for another rigorous instalment. I replicated parts of your analysis (monthly cohorts, rigid real withdrawals, same portfolio definitions) and I read your conclusion differently than your tables do.
1. The failsafe verdict is made entirely by the gold-standard cohorts. In the 1900s-1910s decades that set your “All years” failsafes, gold IS the money: a fixed nominal price by definition, so any gold allocation is dead weight, and the Golden Butterfly or All Weather aren’t definable portfolios in that world. Same problem, milder, from 1934 to 1968 ($35 by decree, illegal to hold until 1974) while T-bills sat rate-capped under the Fed-Treasury accord. Your own by-decade rows tell that story: Risk Parity loses exactly the administered-price decades (1930s-40s) and wins the decades that actually kill retirees, 4.47%/4.46% vs 3.83% in the 1960s and 4.93%/5.04% vs 4.32% in the 1970s (GB Light / GR Light vs 75/25). The window objection you rightly raise against PortfolioCharts cuts both ways.
2. Since 1926, your tables rank Risk Parity above 75/25 even before any SCV backfill. Failsafe: Golden Ratio Light 3.95%, All Weather 3.92% vs 75/25 3.83%. Failure rates at 4%: 0.19%, 1.30%, 1.39% vs 1.94%. With raw SCV it widens (GR 4.22%, GB 4.08% vs 3.83% at 30 years). So “the parts that did well did so despite Risk Parity” doesn’t match the no-SCV columns. Where your case is really strong is the 50-year horizon, where the cash/gold drag compounds mercilessly (I replicate that), and the CAPE>20 conditioning (6.13% vs 14.48% failure is a real warning). “30y ambiguous-to-favorable since 1926, 50y decisive against, and no valuation hedge” seems like the defensible headline, rather than scam.
3. Managed futures never get isolated. The maximum tested is 10%, inside a portfolio where 26% long bonds, 16% gold and 6% cash do the damage; and Part 63 itself concluded in momentum’s favor. On the DBMF backfill: a 67% loading on a 4-asset monthly rotation is both smoother than the real fund (you note it) and narrower. No commodity, FX or short-rate legs, which is where 1970s and 2022 trend profits actually came from. Two ideas that could be interesting: validate any backfill against BTOP50 over 1987-2000 (real net CTA returns predating both SG indices), and vol/drawdown-match the proxy to actual DBMF post-2019 so sequence risk isn’t understated. I’d love to see 65/25/10 and 60/20/20 with a properly rough trend series before a category verdict.
On SCV, one channel worth separating from the factor-alpha bet: at CAPE ~40 concentrated in mega-cap growth, small value is the equity that isn’t the bubble. 2000-2007 was that mechanism, not SMB/HML alpha, and it survives even under your zero-alpha assumption.
Appreciate your blog for insights as I am trying to move from an accumulation to a drawdown portfolio.
Another consideration of the more complex Risk Parity portfolios is the necessity of understanding what asset should be performing best in the current economic “climate” quadrant and then selling the ‘right’ asset as needed.
How much lower could a SWR be for a RPP if withdrawals are not optimal?
Well, RP claims to solve that asset allocation problem. The issue is that the portfolio that’s acceptable in all climates will be only mediocre in the environment we’re in the most time, i.e., economic expansion, stable inflation.
I want to be gracious and simply assign the same SWR to RP as with my portfolio. Maybe you don’t need much of a haircut, if any.
No, we mostly came to the same conclusions:
The 1929 cohort benefited from gold, I have always conceded that, see Part 34. The 1937 cohort did not.
I conceded that RP is competitive with 75/25 in the post-1926 era when looking at 30y windows. I think we both agree that a 5% WR would not be a good idea even in that case.
In the more FIRE-relevant case of 50y, RP is awful without SCV and looks quite good with raw SCV returns.
I still call it a scam if a) 5% seems really too aggressive over 30y and irresponsible over 50y, and b) the gains came mostly from SCV not RP.
Not sure what you mean by isolating MF. Momentum works, but likely not in the form provided by DBMF. I replicated the DBMF with just commodities (Taylor at PortfolioCharts does that on his site) and the results were inferior.
But I agree: more research could be done to study this.
I can understand your concern in the last point: The last time we had CAPE above 40 we had a good decade for SCV and RP. Maybe this will repeat. I would not bet my retirement on that.
Thanks for the reply, and agreed on more of what you wrote: 5% rigid is indefensible over 30y and reckless over 50y, and without SCV the 50y numbers are unambiguous.
One residual quibble on attribution: for 30y since 1926 the no-SCV columns already edge out 75/25 (failsafe 3.95%/3.92% vs 3.83%, 4% failure 0.19%/1.30%/1.39% vs 1.94%), so on that horizon the gains can’t come mostly from SCV. It seems SCV widens an edge that exists without it. Your 50y point stands, and it’s the FIRE-relevant one; an interesting question is why the ranking flips with horizon (my read: the cash/gold drag compounds with years while the sequence-risk protection is front-loaded).
By “isolating MF” I meant the marginal, dose-response test: 75/25 vs 65/25/10 vs 55/25/20, same bonds, so the sleeve isn’t judged together with 26% long bonds and 16% gold. I ran it on US data 1989+ (a trend proxy pinned to net SG Trend, so deliberately conservative): clean monotone dose-response in the tails (ruin at 4%/30y: 1.7% -> 0.9% -> 0.4% for 0/10/20% trend; similar at 50y), flat failsafe, with the cost showing up only in the median. And it still doesn’t rescue a 5% WR. Caveat freely admitted: 1989+ misses the 1970s, and historical cohorts on that window show nothing (no cohort fails at 4%); the effect appears under resampling. A commodities-only replication tests a different animal, by the way: DBMF’s book is mostly rates, equity indices and FX, where the 2022 profits sat.
Fair point on CAPE>40 and 2000-2007. I’d frame SCV as intra-equity diversification your own zero-alpha scenario prices at roughly nothing, rather than a bet on a rerun.
Noted.
Thanks for clarifying. If I add 10% momentum to the 75/25 portfolio (68% equities, 22% bonds, 10% DBMF simulation), then I get some really nice results, both for 1929 and 1960s. So, this is a real winner. If I use the full ERN momentum strategy, it’s even better. But 5% SWR will be hard to achieve, especially over 50y.
Karsten I think you should reach out to Brad at ChooseFi and discuss these results on his podcast. I fear there are too many novice investors who are going all in on this shiny new object (risk parity) even during the accumulation stage based on Frank’s convincing confident style.
I was always suspicious of this approach and it seems to me that Frank’s perspective is more about pushing his life philosophy against what he deems “hoarding money” rather than sound investing advice.
I’ll second Jim’s recommendation.
Frank’s style is to bully people into not publicly disagreeing with him, so there are far too few contrarian voices out there.
And ChooseFI would be a great platform for you (Karsten) to set the record straight on both Risk Parity and SCV premium.
Thanks! With two requests now I might do that! 🙂
Make that three.
Thanks! I will do that soon! 🙂
I could do that. And maybe all the other podcasters. I will have a debate with Frank about SWRs on Bigger Pockets later this week, maybe the RP issue will come up there. But I’m not really good at inviting myself and pushing myself onto people shows, haha. I will see if I have the guts to reach out to Brad, Paula, etc.
Oh boy…I ran backtests…the SWR is higher and drawdowns shallower. That’s all that matters to me and I dont plan to use 5% SWR, only 3.5%!
Good for you. If 3.5% is all you need then why would you do RP?
Lower drawdowns and sleep well at night basically. 75/25 is too much of a roller-coaster for my taste
Risk Parity recently had larger drawdowns than 75/25, i.e., during the 2022 bear market. But maybe you used a different version of Risk Parity than Frank. Good for you.
Good luck, and God Bless You!
This is extremely annoying. Two prominent figures of the FIRE movement disagreeing in something so important for everyone. It’s mind-blowing. I respect them both but has become a feud where we all are being mislead. Shame on both of them. Childish behavior
Thanks for your respect. I don’t quite see where I mislead anyone, though.
I agree with Karsten. I don’t believe he is misleading anyone, so nothing to be ashamed of.
I don’t think Frank is intentionally misleading people either (I believe he really believes what he espouses). But his debate style is very caustic/combative, bullying people to get them to not disagree with him publicly and insulting those who disagree with him, rather than debating arguments on their merits, which I believe results in him misleading people unintentionally because he shuts down valid criticisms by bullying. I agree Frank’s debate style is childish.
Karsten is also far more qualified to conduct scholarly SWR analysis, as a Ph.D. economist himself, having taught Ph.D level economics, and a former quant for the US Federal Reserve, whereas Frank is a former lawyer, a substantial skilled profession, but not the one I would rely on for econometric research.
So all Paul Merriman’s studies we can all just ignore because Karsten said so? Not so fast, not so fast. Do your own study and don’t trust anything you read in these blogs
Merriman would be happy with 1% p.a. outperformance in SCV. I know him. I’ve had dinner with him. Have you?
But 1% is not going to suffice to raise your SWR to 5% over 50 years.
Thank you so much for your work and your passion! 🙂
I hadn’t planned on deviating from the 75/25 split, but with every new article, I become more convinced that I shouldn’t change a thing. Instead, I’m simply continuing to enjoy my early retirement.
Greetings from Germany
Thanks, Michael! Viele Gruesse zurueck nach Deutschland!
Definitely another “Drop a mic post”. 75/25 allocation and carry on! Thanks Karsten for keeping the world on the right path. My thought is that many other folks, while well read, educated, and well meaning; simply don’t have the quantified proof that you have.
Thanks, Karl! Index plus a little bit of CBOE SPX options is hugely profitable, so, why rock the boat?!
Ern, you kind of already wrote a post about leveraging risk parity (Lower risk through leverage). A stock/bond portfolio is arguably the most basic version of risk parity. The trick is to use futures on the lower risk assets. I played around with the SWR spreadsheet and if I only go back to 1926 I can maximize my SWR (40 year horizon with SS and pension) by leveraging a 50/25/25 (stocks/bonds/gold) by 1.4. If I go back further, the gold makes it worse and I don’t get much value with leverage in general. As somebody who has most of their savings in 401K/IRA I can actually emulate this using NTSX and GDE (and the expense ratios are not bad at .2).
As a quick check on my methodology, I put 70% for stocks, 35% for bonds and -40% for cash (gold defaulted to 35%). Is this correct way to emulate leverage in the spreadsheet?
Correct: the cash weight is -1 times the excess over 100%.
Actually I also tried just removing the formula on the tax column and entering over 100% and got different results. Its surprisingly harder to lower the SWR with this method. My guess is that during some of the rough times inflation was high and cash was negative so doing -40% cash was actually helping the portfolio.
Two kids indeed.
The irony is criticizing others for relying on history while relying heavily on historical data himself. Nobody has a crystal ball. Every SWR model is a bet on assumptions. Frank has backtests; Karsten has simulations. Both are imperfect views of an unknowable future. The debate should be about methodology, not pretending one side owns the truth.
I don’t criticize him for “history” but for using only one single window and not the entire history.
I’d like to see more studies using Monte Carlo simulations or dataset generation methodologies. When you go that route, you can front-test(?) millions of lifetimes instead of the last few decades.
The thing holding back practitioners is the focus on asset allocation. Made-up datasets might not exhibit the same correlations and countercorrelation swings between stocks and bonds, for example. Of course, the issue with artificial numbers for gold’s value in the historical dataset is at least equally worrisome. So even if you get the dataset’s variance, factors, ranges, etc. right, what do you do about the covariances or the patterns that might exist in reality but not in your dataset?
If we asked ourselves for ways other than altering our AA or taking on leverage to improve our expected SWR, I suspect methods other than historical back testing would be preferred. E.g. if I’m eliminating outlier returns by hedging with options or futures, what does that look like?
I agree with Chris that it would be great if you could include some Monte Carlo SWR analysis for FIRE time horizons to complement your backtest-based analysis.
An argument can be made that backtesting is overly optimistic for prediction (because “past performance is not a guarantee of future results”) and Monte Carlo analysis is somewhat overly pessimistic (because generated random data, even with appropriate statistical distributions may contain some crazy sequences that would never occur in real economies or equity markets).
So a mix of Monte Carlo and backtest would be more robust that backtest alone.
Is back testing overly optimistic? That can be a real danger when you are looking to maximize returns, but maximizing SWR is a different story. At that point the focus is on worst case scenarios that arguably should not happen again (like the great depression). You can make an argument that back testing (especially when going back 100+ years) is pessimistic.
If you stipulate that some Monte Carlo scenarios would never happen in the real world, I’m struggling with how you get value out of that?
> Is back testing overly optimistic?
I believe so (as does Wade Pfau, who like Karsten, is a Ph.D. economist, and he is the principal creator of the Retirement Income Certified Professional (RICP) credential), in that future worst case could be worse than past worst case. Given that most countries SWRs are lower than US SWRs, there is certainly room for worst case US SWR to degrade relative to past history.
I am not arguing that Monte Carlo is superior, but it is by far the most widely used modeling methodology to test retirement portfolio success rates for a reason, so including it along with backtesting is a more robust overall picture. All the major brokerages use Monte Carlo simulations.
It is possible a future worse case could be worse than the past, but I think its unlikely. The market has been growing at roughly the same rate for a very long time. The Fed has gotten better at handling recessions which have been trending shorter duration for a long time. Even without trending in the right direction its not likely the worst case is going to happen in the next 10 years (the real danger) vs the prior 150. It seems like a simple way to account for this is aim for higher final value (as Ern did in his 50 year test case).
The case against Monte Carlo is just too strong for me. In the effort to increase sample size they are adding a lot of potential junk into the data. I would rather work with a smaller sample of real data than a larger sample of questionable data.
I also think it is unlikely, but so is being in the cohort experiencing the worst SWR, so that isn’t a counterargument, as SWR is all about worst cases, not expected cases.
Higher final value serves a different purpose (legacy, avoiding a scary ride towards portfolio depletion at the end). While it moves in the same direction, it isn’t really a substitute.
It can serve both purposes. What is your alternative? Run Monte Carlo with impossible scenarios and drive off that? There could be catastrophic events that cause the underlying Monte Carlo assumptions to no longer apply. How do you account for that? There is no limit to how careful you can be. Personally I think something this bad has not happened in the last 150 years is a reasonable line to draw.
Tony Robbins has made millions for himself selling his self help jingo-ism and holding rallies for his followers. The article was really just an advertorial for the next thing he’s selling:
“His first book in over 20 years, MONEY Master the Game: 7 Simple Steps to Financial Freedom, is out November 18th.”
I appreciate you doing your part to flag the issues with it, but I imagine this will become the new rage in certain corner of the internet!
I also heard he dabbled with nutrition advice once and almost killed himself with heavy metal poison. So, I will stay away from this financial guru, haha!
At this point I think you have completed the collection of static SWR analysis, great job!
This is a bit off-tangent, but there seems to be a resurgence in dynamic SWR with recent risk-based guardrails that claims to be an improvement over Guyton-Klinger and similar methods of that era. At least to me it’s an intuitive expansion of SWR, that sets readjust the SWR when you hit upper/lower guardrails that can be defined as hitting 100% and 50% probability of success. It seems to be more principled/intuitive that traditional dynamic SWR that just make up the rule of thumb e.g. Guyton-Klinger.
There’s a software company IncomeLabs that simplifies this and many advisors seem to be onboard lately, and it’s written about in Kitces blog. Karsten, any chance of us doing a deep dive into these more “modern” dynamic SWR methods?
Many of the cases where Risk Parity is inferior to 75/25 are based on the 1900s/1910s market. That was a very different world: the dollar was pegged to gold, the Federal Reserve wasn’t established until 1913, and Depression-era financial market regulations were not yet in place.
I’ll leave it to the PhDs to tell me if that market is representative of the future, but if we’re saying the world has changed and SCV alpha is not going to persist based on a really tubular paper, seems reasonable to me to question what aspects of the 1900s/1910s will persist.
Separately, Frank Vasquez gets to 5% SWR with a variable withdrawal rate strategy and inflation adjustments less than CPI. Clearly there are drawbacks to both, but it’s a straw man to say he claims a 5% Trinity-style SWR just via Risk Parity.
(By the way, you were my intro to FIRE (thank you) and I’m a regular listener to Frank’s podcast. Clearly you agree on the vast majority of points, and I find a rigorous debate on the disagreements educational.)
> Frank Vasquez gets to 5% SWR with a variable withdrawal rate strategy and inflation adjustments less than CPI.
No, Frank claims perpetual 5% fixed or more with at least some of the portfolios:
See https://www.riskparityradio.com/portfolios. Click on Golden Butterfly or Golden Ratio, and look for “an expected permanent safe withdrawal rate of”. In both cases 5.3% and 5.0% respectively. He is claiming at least 5% before applying dynamic spending or assuming reduced spending with age.
I see, from the Golden Butterfly info page:
> Since 1970, it has a compounded annual growth rate (after inflation) of 6.4%, and an expected permanent safe withdrawal rate of 5.3%.
Karsten reaches essentially the same conclusion above using raw SCV returns (Karsten is slightly lower but Frank is not as transparent with his methodology so there are various potential explanations). But I agree with you, Frank’s use of the word “permanent” doesn’t make sense right next to “Since 1970,” and is potentially misleading.
I do hear more nuanced explanation on his podcast, which is potentially an indictment of a poorly-written website.
Stupid idea: buy stocks with dividends. No need to sell the goose that lays the golden eggs.
Is there any testing for only buying stocks that have a yield?
Have index funds ruined value in the stock market because it inflated businesses with no return?
To me, buying something in hopes it goes up is gambling. Instead I would rather invest in things that pay me (real estate rentals) once I buy I earn forever, without having to sell.
All of this looks fun, but at what time do we stop focusing on projecting, back testing, gambling strategies and start focusing on quality businesses?
I find it quite amusing how all of the “financial gurus” are all over the map on what percentage of equities to hold in retirement.
Let me see if I have this straight:
Morningstar 30-50% equities has the highest SWR (latest December 2025 report); Portfolio Charts (Tyler) – Golden Butterfly 40%; Risk Parity Radio (Frank) – Golden Ratio 42%; Bill Bengen 55% (latest book); the classic financial industry 60/40 – 60%; your recommendation 75%; Warren Buffett 90% (his wife’s trust) and Cederburg 100%.
No wonder there is so much confusion out there for the retail investor!
It makes more sense when you realize the goals are different. Most of those guys are focused on the generic 30 year case. Ern is focused on early retirement, so he is looking at a longer window (like 50 years). The longer your portfolio needs to last the more it needs to earn and thus the higher equity percentage. That is the biggest point of this article. These risk parity portfolios don’t have the long term staying power because the equities percentage is too small.
Are you saying Ern has a different recommendation for a 30 year vs a 50 year retirement goal? I’m not seeing that. From what I’m reading, it’s 75/25 for either one.
75/25 is a good generic portfolio that works well across different timeframes and historical data, so that is what he used as a baseline. Most studies focus on a 30 year horizon and most of his readers have longer horizons so he needs to take that into account. To be clear, he doesn’t generally recommend specific portfolios. He simply has views on what makes sense and what doesn’t. As you have read, he is skeptical of the SCV factor and believes the example risk parity portfolios need more equity, especially if you have a longer horizon.
Hi Karsten,
First off, I just want to say that I am a huge fan. I have been following your blog for years and read almost every article you publish.
I really appreciate having your voice in the FIRE community. When it comes to critical topics like guardrails, variable withdrawal methods, and the general push for “flexibility” to allow for higher initial withdrawal rates, there is so much noise out there. Many bloggers, YouTubers, podcasters, and CFPs promote these methods, but few seem to apply the mathematical rigor required to truly test them—especially for the long horizons of early retirees.
As a retail investor, it’s easy to get lost in these overly optimistic proposals. Your blog is one of the very few sources I truly trust to help me make concrete decisions for my retirement.
Because of this, I have a request. I would love it if you could write a post analyzing what seems to be the hottest topic in the community right now (perhaps even more so than Risk Parity!): Risk-Based Guardrails.
As you probably know, Derek Tharp wrote several articles about this on the Kitces blog, and since then, it feels like the financial community hasn’t stopped talking about it. Aubrey Williams (who I believe has commented here before) frequently discusses it on podcasts as well.
Listening to the proponents of Risk-Based Guardrails, you’d think they’ve finally found the panacea for Sequence of Return Risk. It’s everywhere on YouTube, frequently recommended by CFPs, and is being integrated into retirement software like Income Lab and Boldin. They make it sound like you can safely withdraw much more money, and that any income reductions during a severe market crash will be relatively minor and short-lived.
However, when I played around with this free simulation tool (https://fire-guardrails.streamlit.app), the required income reductions were actually much larger and lasted significantly longer than what is usually claimed.
Maybe I am missing something or don’t fully understand the mechanics, but I have yet to see a thorough, critical review of this method anywhere. It just feels too optimistic to me.
I would be thrilled if you could do a deep dive, cut through the hype, and show us the real math behind Risk-Based Guardrails.
Thanks for everything you do!